diff --git a/build/sbt b/build/sbt index 28ebb64f7197c..cc3203d79bccd 100755 --- a/build/sbt +++ b/build/sbt @@ -125,4 +125,32 @@ loadConfigFile() { [[ -f "$etc_sbt_opts_file" ]] && set -- $(loadConfigFile "$etc_sbt_opts_file") "$@" [[ -f "$sbt_opts_file" ]] && set -- $(loadConfigFile "$sbt_opts_file") "$@" +exit_status=127 +saved_stty="" + +restoreSttySettings() { + stty $saved_stty + saved_stty="" +} + +onExit() { + if [[ "$saved_stty" != "" ]]; then + restoreSttySettings + fi + exit $exit_status +} + +saveSttySettings() { + saved_stty=$(stty -g 2>/dev/null) + if [[ ! $? ]]; then + saved_stty="" + fi +} + +saveSttySettings +trap onExit INT + run "$@" + +exit_status=$? +onExit diff --git a/build/sbt-launch-lib.bash b/build/sbt-launch-lib.bash index 5e0c640fa5919..504be48b358fa 100755 --- a/build/sbt-launch-lib.bash +++ b/build/sbt-launch-lib.bash @@ -81,7 +81,7 @@ execRunner () { echo "" } - exec "$@" + "$@" } addJava () { diff --git a/core/pom.xml b/core/pom.xml index 4daaf88147142..66180035e61f1 100644 --- a/core/pom.xml +++ b/core/pom.xml @@ -243,6 +243,14 @@ io.dropwizard.metrics metrics-graphite + + com.fasterxml.jackson.core + jackson-databind + + + com.fasterxml.jackson.module + jackson-module-scala_2.10 + org.apache.derby derby diff --git a/core/src/main/resources/org/apache/spark/ui/static/webui.css b/core/src/main/resources/org/apache/spark/ui/static/webui.css index f23ba9dba167f..68b33b5f0d7c7 100644 --- a/core/src/main/resources/org/apache/spark/ui/static/webui.css +++ b/core/src/main/resources/org/apache/spark/ui/static/webui.css @@ -103,6 +103,12 @@ span.expand-details { float: right; } +span.rest-uri { + font-size: 10pt; + font-style: italic; + color: gray; +} + pre { font-size: 0.8em; } diff --git a/core/src/main/scala/org/apache/spark/CacheManager.scala b/core/src/main/scala/org/apache/spark/CacheManager.scala index a0c0372b7f0ef..a96d754744a05 100644 --- a/core/src/main/scala/org/apache/spark/CacheManager.scala +++ b/core/src/main/scala/org/apache/spark/CacheManager.scala @@ -47,10 +47,15 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging { val inputMetrics = blockResult.inputMetrics val existingMetrics = context.taskMetrics .getInputMetricsForReadMethod(inputMetrics.readMethod) - existingMetrics.addBytesRead(inputMetrics.bytesRead) - - new InterruptibleIterator(context, blockResult.data.asInstanceOf[Iterator[T]]) + existingMetrics.incBytesRead(inputMetrics.bytesRead) + val iter = blockResult.data.asInstanceOf[Iterator[T]] + new InterruptibleIterator[T](context, iter) { + override def next(): T = { + existingMetrics.incRecordsRead(1) + delegate.next() + } + } case None => // Acquire a lock for loading this partition // If another thread already holds the lock, wait for it to finish return its results diff --git a/core/src/main/scala/org/apache/spark/ExecutorAllocationClient.scala b/core/src/main/scala/org/apache/spark/ExecutorAllocationClient.scala index a46a81eabd965..443830f8d03b6 100644 --- a/core/src/main/scala/org/apache/spark/ExecutorAllocationClient.scala +++ b/core/src/main/scala/org/apache/spark/ExecutorAllocationClient.scala @@ -19,24 +19,32 @@ package org.apache.spark /** * A client that communicates with the cluster manager to request or kill executors. + * This is currently supported only in YARN mode. */ private[spark] trait ExecutorAllocationClient { + /** + * Express a preference to the cluster manager for a given total number of executors. + * This can result in canceling pending requests or filing additional requests. + * @return whether the request is acknowledged by the cluster manager. + */ + private[spark] def requestTotalExecutors(numExecutors: Int): Boolean + /** * Request an additional number of executors from the cluster manager. - * Return whether the request is acknowledged by the cluster manager. + * @return whether the request is acknowledged by the cluster manager. */ def requestExecutors(numAdditionalExecutors: Int): Boolean /** * Request that the cluster manager kill the specified executors. - * Return whether the request is acknowledged by the cluster manager. + * @return whether the request is acknowledged by the cluster manager. */ def killExecutors(executorIds: Seq[String]): Boolean /** * Request that the cluster manager kill the specified executor. - * Return whether the request is acknowledged by the cluster manager. + * @return whether the request is acknowledged by the cluster manager. */ def killExecutor(executorId: String): Boolean = killExecutors(Seq(executorId)) } diff --git a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala index 02d54bf3b53cc..998695b6ac8ab 100644 --- a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala +++ b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala @@ -201,18 +201,34 @@ private[spark] class ExecutorAllocationManager( } /** - * If the add time has expired, request new executors and refresh the add time. - * If the remove time for an existing executor has expired, kill the executor. + * The number of executors we would have if the cluster manager were to fulfill all our existing + * requests. + */ + private def targetNumExecutors(): Int = + numExecutorsPending + executorIds.size - executorsPendingToRemove.size + + /** + * The maximum number of executors we would need under the current load to satisfy all running + * and pending tasks, rounded up. + */ + private def maxNumExecutorsNeeded(): Int = { + val numRunningOrPendingTasks = listener.totalPendingTasks + listener.totalRunningTasks + (numRunningOrPendingTasks + tasksPerExecutor - 1) / tasksPerExecutor + } + + /** + * This is called at a fixed interval to regulate the number of pending executor requests + * and number of executors running. + * + * First, adjust our requested executors based on the add time and our current needs. + * Then, if the remove time for an existing executor has expired, kill the executor. + * * This is factored out into its own method for testing. */ private def schedule(): Unit = synchronized { val now = clock.getTimeMillis - if (addTime != NOT_SET && now >= addTime) { - addExecutors() - logDebug(s"Starting timer to add more executors (to " + - s"expire in $sustainedSchedulerBacklogTimeout seconds)") - addTime += sustainedSchedulerBacklogTimeout * 1000 - } + + addOrCancelExecutorRequests(now) removeTimes.retain { case (executorId, expireTime) => val expired = now >= expireTime @@ -223,59 +239,89 @@ private[spark] class ExecutorAllocationManager( } } + /** + * Check to see whether our existing allocation and the requests we've made previously exceed our + * current needs. If so, let the cluster manager know so that it can cancel pending requests that + * are unneeded. + * + * If not, and the add time has expired, see if we can request new executors and refresh the add + * time. + * + * @return the delta in the target number of executors. + */ + private def addOrCancelExecutorRequests(now: Long): Int = synchronized { + val currentTarget = targetNumExecutors + val maxNeeded = maxNumExecutorsNeeded + + if (maxNeeded < currentTarget) { + // The target number exceeds the number we actually need, so stop adding new + // executors and inform the cluster manager to cancel the extra pending requests. + val newTotalExecutors = math.max(maxNeeded, minNumExecutors) + client.requestTotalExecutors(newTotalExecutors) + numExecutorsToAdd = 1 + updateNumExecutorsPending(newTotalExecutors) + } else if (addTime != NOT_SET && now >= addTime) { + val delta = addExecutors(maxNeeded) + logDebug(s"Starting timer to add more executors (to " + + s"expire in $sustainedSchedulerBacklogTimeout seconds)") + addTime += sustainedSchedulerBacklogTimeout * 1000 + delta + } else { + 0 + } + } + /** * Request a number of executors from the cluster manager. * If the cap on the number of executors is reached, give up and reset the * number of executors to add next round instead of continuing to double it. - * Return the number actually requested. + * + * @param maxNumExecutorsNeeded the maximum number of executors all currently running or pending + * tasks could fill + * @return the number of additional executors actually requested. */ - private def addExecutors(): Int = synchronized { - // Do not request more executors if we have already reached the upper bound - val numExistingExecutors = executorIds.size + numExecutorsPending - if (numExistingExecutors >= maxNumExecutors) { + private def addExecutors(maxNumExecutorsNeeded: Int): Int = { + // Do not request more executors if it would put our target over the upper bound + val currentTarget = targetNumExecutors + if (currentTarget >= maxNumExecutors) { logDebug(s"Not adding executors because there are already ${executorIds.size} " + s"registered and $numExecutorsPending pending executor(s) (limit $maxNumExecutors)") numExecutorsToAdd = 1 return 0 } - // The number of executors needed to satisfy all pending tasks is the number of tasks pending - // divided by the number of tasks each executor can fit, rounded up. - val maxNumExecutorsPending = - (listener.totalPendingTasks() + tasksPerExecutor - 1) / tasksPerExecutor - if (numExecutorsPending >= maxNumExecutorsPending) { - logDebug(s"Not adding executors because there are already $numExecutorsPending " + - s"pending and pending tasks could only fill $maxNumExecutorsPending") - numExecutorsToAdd = 1 - return 0 - } - - // It's never useful to request more executors than could satisfy all the pending tasks, so - // cap request at that amount. - // Also cap request with respect to the configured upper bound. - val maxNumExecutorsToAdd = math.min( - maxNumExecutorsPending - numExecutorsPending, - maxNumExecutors - numExistingExecutors) - assert(maxNumExecutorsToAdd > 0) - - val actualNumExecutorsToAdd = math.min(numExecutorsToAdd, maxNumExecutorsToAdd) - - val newTotalExecutors = numExistingExecutors + actualNumExecutorsToAdd - val addRequestAcknowledged = testing || client.requestExecutors(actualNumExecutorsToAdd) + val actualMaxNumExecutors = math.min(maxNumExecutors, maxNumExecutorsNeeded) + val newTotalExecutors = math.min(currentTarget + numExecutorsToAdd, actualMaxNumExecutors) + val addRequestAcknowledged = testing || client.requestTotalExecutors(newTotalExecutors) if (addRequestAcknowledged) { - logInfo(s"Requesting $actualNumExecutorsToAdd new executor(s) because " + - s"tasks are backlogged (new desired total will be $newTotalExecutors)") - numExecutorsToAdd = - if (actualNumExecutorsToAdd == numExecutorsToAdd) numExecutorsToAdd * 2 else 1 - numExecutorsPending += actualNumExecutorsToAdd - actualNumExecutorsToAdd + val delta = updateNumExecutorsPending(newTotalExecutors) + logInfo(s"Requesting $delta new executor(s) because tasks are backlogged" + + s" (new desired total will be $newTotalExecutors)") + numExecutorsToAdd = if (delta == numExecutorsToAdd) { + numExecutorsToAdd * 2 + } else { + 1 + } + delta } else { - logWarning(s"Unable to reach the cluster manager " + - s"to request $actualNumExecutorsToAdd executors!") + logWarning( + s"Unable to reach the cluster manager to request $newTotalExecutors total executors!") 0 } } + /** + * Given the new target number of executors, update the number of pending executor requests, + * and return the delta from the old number of pending requests. + */ + private def updateNumExecutorsPending(newTotalExecutors: Int): Int = { + val newNumExecutorsPending = + newTotalExecutors - executorIds.size + executorsPendingToRemove.size + val delta = newNumExecutorsPending - numExecutorsPending + numExecutorsPending = newNumExecutorsPending + delta + } + /** * Request the cluster manager to remove the given executor. * Return whether the request is received. @@ -415,6 +461,8 @@ private[spark] class ExecutorAllocationManager( private val stageIdToNumTasks = new mutable.HashMap[Int, Int] private val stageIdToTaskIndices = new mutable.HashMap[Int, mutable.HashSet[Int]] private val executorIdToTaskIds = new mutable.HashMap[String, mutable.HashSet[Long]] + // Number of tasks currently running on the cluster. Should be 0 when no stages are active. + private var numRunningTasks: Int = _ override def onStageSubmitted(stageSubmitted: SparkListenerStageSubmitted): Unit = { val stageId = stageSubmitted.stageInfo.stageId @@ -435,6 +483,10 @@ private[spark] class ExecutorAllocationManager( // This is needed in case the stage is aborted for any reason if (stageIdToNumTasks.isEmpty) { allocationManager.onSchedulerQueueEmpty() + if (numRunningTasks != 0) { + logWarning("No stages are running, but numRunningTasks != 0") + numRunningTasks = 0 + } } } } @@ -446,6 +498,7 @@ private[spark] class ExecutorAllocationManager( val executorId = taskStart.taskInfo.executorId allocationManager.synchronized { + numRunningTasks += 1 // This guards against the race condition in which the `SparkListenerTaskStart` // event is posted before the `SparkListenerBlockManagerAdded` event, which is // possible because these events are posted in different threads. (see SPARK-4951) @@ -475,7 +528,8 @@ private[spark] class ExecutorAllocationManager( val executorId = taskEnd.taskInfo.executorId val taskId = taskEnd.taskInfo.taskId allocationManager.synchronized { - // If the executor is no longer running scheduled any tasks, mark it as idle + numRunningTasks -= 1 + // If the executor is no longer running any scheduled tasks, mark it as idle if (executorIdToTaskIds.contains(executorId)) { executorIdToTaskIds(executorId) -= taskId if (executorIdToTaskIds(executorId).isEmpty) { @@ -514,6 +568,11 @@ private[spark] class ExecutorAllocationManager( }.sum } + /** + * The number of tasks currently running across all stages. + */ + def totalRunningTasks(): Int = numRunningTasks + /** * Return true if an executor is not currently running a task, and false otherwise. * diff --git a/core/src/main/scala/org/apache/spark/SparkConf.scala b/core/src/main/scala/org/apache/spark/SparkConf.scala index 13aa9960ac33a..0dbd26146cb13 100644 --- a/core/src/main/scala/org/apache/spark/SparkConf.scala +++ b/core/src/main/scala/org/apache/spark/SparkConf.scala @@ -18,6 +18,7 @@ package org.apache.spark import java.util.concurrent.ConcurrentHashMap +import java.util.concurrent.atomic.AtomicBoolean import scala.collection.JavaConverters._ import scala.collection.mutable.LinkedHashSet @@ -67,7 +68,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging { if (value == null) { throw new NullPointerException("null value for " + key) } - settings.put(key, value) + settings.put(translateConfKey(key, warn = true), value) this } @@ -139,7 +140,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging { /** Set a parameter if it isn't already configured */ def setIfMissing(key: String, value: String): SparkConf = { - settings.putIfAbsent(key, value) + settings.putIfAbsent(translateConfKey(key, warn = true), value) this } @@ -175,7 +176,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging { /** Get a parameter as an Option */ def getOption(key: String): Option[String] = { - Option(settings.get(key)) + Option(settings.get(translateConfKey(key))) } /** Get all parameters as a list of pairs */ @@ -228,7 +229,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging { def getAppId: String = get("spark.app.id") /** Does the configuration contain a given parameter? */ - def contains(key: String): Boolean = settings.containsKey(key) + def contains(key: String): Boolean = settings.containsKey(translateConfKey(key)) /** Copy this object */ override def clone: SparkConf = { @@ -285,7 +286,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging { // Validate memory fractions val memoryKeys = Seq( "spark.storage.memoryFraction", - "spark.shuffle.memoryFraction", + "spark.shuffle.memoryFraction", "spark.shuffle.safetyFraction", "spark.storage.unrollFraction", "spark.storage.safetyFraction") @@ -351,9 +352,20 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging { def toDebugString: String = { getAll.sorted.map{case (k, v) => k + "=" + v}.mkString("\n") } + } -private[spark] object SparkConf { +private[spark] object SparkConf extends Logging { + + private val deprecatedConfigs: Map[String, DeprecatedConfig] = { + val configs = Seq( + DeprecatedConfig("spark.files.userClassPathFirst", "spark.executor.userClassPathFirst", + "1.3"), + DeprecatedConfig("spark.yarn.user.classpath.first", null, "1.3", + "Use spark.{driver,executor}.userClassPathFirst instead.")) + configs.map { x => (x.oldName, x) }.toMap + } + /** * Return whether the given config is an akka config (e.g. akka.actor.provider). * Note that this does not include spark-specific akka configs (e.g. spark.akka.timeout). @@ -380,4 +392,63 @@ private[spark] object SparkConf { def isSparkPortConf(name: String): Boolean = { (name.startsWith("spark.") && name.endsWith(".port")) || name.startsWith("spark.port.") } + + /** + * Translate the configuration key if it is deprecated and has a replacement, otherwise just + * returns the provided key. + * + * @param userKey Configuration key from the user / caller. + * @param warn Whether to print a warning if the key is deprecated. Warnings will be printed + * only once for each key. + */ + def translateConfKey(userKey: String, warn: Boolean = false): String = { + deprecatedConfigs.get(userKey) + .map { deprecatedKey => + if (warn) { + deprecatedKey.warn() + } + deprecatedKey.newName.getOrElse(userKey) + }.getOrElse(userKey) + } + + /** + * Holds information about keys that have been deprecated or renamed. + * + * @param oldName Old configuration key. + * @param newName New configuration key, or `null` if key has no replacement, in which case the + * deprecated key will be used (but the warning message will still be printed). + * @param version Version of Spark where key was deprecated. + * @param deprecationMessage Message to include in the deprecation warning; mandatory when + * `newName` is not provided. + */ + private case class DeprecatedConfig( + oldName: String, + _newName: String, + version: String, + deprecationMessage: String = null) { + + private val warned = new AtomicBoolean(false) + val newName = Option(_newName) + + if (newName == null && (deprecationMessage == null || deprecationMessage.isEmpty())) { + throw new IllegalArgumentException("Need new config name or deprecation message.") + } + + def warn(): Unit = { + if (warned.compareAndSet(false, true)) { + if (newName != null) { + val message = Option(deprecationMessage).getOrElse( + s"Please use the alternative '$newName' instead.") + logWarning( + s"The configuration option '$oldName' has been replaced as of Spark $version and " + + s"may be removed in the future. $message") + } else { + logWarning( + s"The configuration option '$oldName' has been deprecated as of Spark $version and " + + s"may be removed in the future. $deprecationMessage") + } + } + } + + } } diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala index 5623587c36fa6..04ca5d1019e4b 100644 --- a/core/src/main/scala/org/apache/spark/SparkContext.scala +++ b/core/src/main/scala/org/apache/spark/SparkContext.scala @@ -1103,10 +1103,27 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli listenerBus.addListener(listener) } + /** + * Express a preference to the cluster manager for a given total number of executors. + * This can result in canceling pending requests or filing additional requests. + * This is currently only supported in YARN mode. Return whether the request is received. + */ + private[spark] override def requestTotalExecutors(numExecutors: Int): Boolean = { + assert(master.contains("yarn") || dynamicAllocationTesting, + "Requesting executors is currently only supported in YARN mode") + schedulerBackend match { + case b: CoarseGrainedSchedulerBackend => + b.requestTotalExecutors(numExecutors) + case _ => + logWarning("Requesting executors is only supported in coarse-grained mode") + false + } + } + /** * :: DeveloperApi :: * Request an additional number of executors from the cluster manager. - * This is currently only supported in Yarn mode. Return whether the request is received. + * This is currently only supported in YARN mode. Return whether the request is received. */ @DeveloperApi override def requestExecutors(numAdditionalExecutors: Int): Boolean = { @@ -1124,7 +1141,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli /** * :: DeveloperApi :: * Request that the cluster manager kill the specified executors. - * This is currently only supported in Yarn mode. Return whether the request is received. + * This is currently only supported in YARN mode. Return whether the request is received. */ @DeveloperApi override def killExecutors(executorIds: Seq[String]): Boolean = { @@ -1420,6 +1437,9 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli val callSite = getCallSite val cleanedFunc = clean(func) logInfo("Starting job: " + callSite.shortForm) + if (conf.getBoolean("spark.logLineage", false)) { + logInfo("RDD's recursive dependencies:\n" + rdd.toDebugString) + } dagScheduler.runJob(rdd, cleanedFunc, partitions, callSite, allowLocal, resultHandler, localProperties.get) progressBar.foreach(_.finishAll()) @@ -2110,7 +2130,7 @@ object SparkContext extends Logging { val scheduler = new TaskSchedulerImpl(sc) val localCluster = new LocalSparkCluster( - numSlaves.toInt, coresPerSlave.toInt, memoryPerSlaveInt) + numSlaves.toInt, coresPerSlave.toInt, memoryPerSlaveInt, sc.conf) val masterUrls = localCluster.start() val backend = new SparkDeploySchedulerBackend(scheduler, sc, masterUrls) scheduler.initialize(backend) diff --git a/core/src/main/scala/org/apache/spark/TestUtils.scala b/core/src/main/scala/org/apache/spark/TestUtils.scala index be081c3825566..35b324ba6f573 100644 --- a/core/src/main/scala/org/apache/spark/TestUtils.scala +++ b/core/src/main/scala/org/apache/spark/TestUtils.scala @@ -17,12 +17,13 @@ package org.apache.spark -import java.io.{File, FileInputStream, FileOutputStream} +import java.io.{ByteArrayInputStream, File, FileInputStream, FileOutputStream} import java.net.{URI, URL} import java.util.jar.{JarEntry, JarOutputStream} import scala.collection.JavaConversions._ +import com.google.common.base.Charsets.UTF_8 import com.google.common.io.{ByteStreams, Files} import javax.tools.{JavaFileObject, SimpleJavaFileObject, ToolProvider} @@ -59,6 +60,22 @@ private[spark] object TestUtils { createJar(files1 ++ files2, jarFile) } + /** + * Create a jar file containing multiple files. The `files` map contains a mapping of + * file names in the jar file to their contents. + */ + def createJarWithFiles(files: Map[String, String], dir: File = null): URL = { + val tempDir = Option(dir).getOrElse(Utils.createTempDir()) + val jarFile = File.createTempFile("testJar", ".jar", tempDir) + val jarStream = new JarOutputStream(new FileOutputStream(jarFile)) + files.foreach { case (k, v) => + val entry = new JarEntry(k) + jarStream.putNextEntry(entry) + ByteStreams.copy(new ByteArrayInputStream(v.getBytes(UTF_8)), jarStream) + } + jarStream.close() + jarFile.toURI.toURL + } /** * Create a jar file that contains this set of files. All files will be located at the root diff --git a/core/src/main/scala/org/apache/spark/deploy/Client.scala b/core/src/main/scala/org/apache/spark/deploy/Client.scala index 38b3da0b13756..237d26fc6bd0e 100644 --- a/core/src/main/scala/org/apache/spark/deploy/Client.scala +++ b/core/src/main/scala/org/apache/spark/deploy/Client.scala @@ -68,8 +68,9 @@ private class ClientActor(driverArgs: ClientArguments, conf: SparkConf) .map(Utils.splitCommandString).getOrElse(Seq.empty) val sparkJavaOpts = Utils.sparkJavaOpts(conf) val javaOpts = sparkJavaOpts ++ extraJavaOpts - val command = new Command(mainClass, Seq("{{WORKER_URL}}", driverArgs.mainClass) ++ - driverArgs.driverOptions, sys.env, classPathEntries, libraryPathEntries, javaOpts) + val command = new Command(mainClass, + Seq("{{WORKER_URL}}", "{{USER_JAR}}", driverArgs.mainClass) ++ driverArgs.driverOptions, + sys.env, classPathEntries, libraryPathEntries, javaOpts) val driverDescription = new DriverDescription( driverArgs.jarUrl, diff --git a/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala b/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala index e5873ce724b9f..415bd50591692 100644 --- a/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala +++ b/core/src/main/scala/org/apache/spark/deploy/ClientArguments.scala @@ -29,8 +29,7 @@ import org.apache.spark.util.{IntParam, MemoryParam} * Command-line parser for the driver client. */ private[spark] class ClientArguments(args: Array[String]) { - val defaultCores = 1 - val defaultMemory = 512 + import ClientArguments._ var cmd: String = "" // 'launch' or 'kill' var logLevel = Level.WARN @@ -39,9 +38,9 @@ private[spark] class ClientArguments(args: Array[String]) { var master: String = "" var jarUrl: String = "" var mainClass: String = "" - var supervise: Boolean = false - var memory: Int = defaultMemory - var cores: Int = defaultCores + var supervise: Boolean = DEFAULT_SUPERVISE + var memory: Int = DEFAULT_MEMORY + var cores: Int = DEFAULT_CORES private var _driverOptions = ListBuffer[String]() def driverOptions = _driverOptions.toSeq @@ -50,7 +49,7 @@ private[spark] class ClientArguments(args: Array[String]) { parse(args.toList) - def parse(args: List[String]): Unit = args match { + private def parse(args: List[String]): Unit = args match { case ("--cores" | "-c") :: IntParam(value) :: tail => cores = value parse(tail) @@ -106,9 +105,10 @@ private[spark] class ClientArguments(args: Array[String]) { |Usage: DriverClient kill | |Options: - | -c CORES, --cores CORES Number of cores to request (default: $defaultCores) - | -m MEMORY, --memory MEMORY Megabytes of memory to request (default: $defaultMemory) + | -c CORES, --cores CORES Number of cores to request (default: $DEFAULT_CORES) + | -m MEMORY, --memory MEMORY Megabytes of memory to request (default: $DEFAULT_MEMORY) | -s, --supervise Whether to restart the driver on failure + | (default: $DEFAULT_SUPERVISE) | -v, --verbose Print more debugging output """.stripMargin System.err.println(usage) @@ -117,6 +117,10 @@ private[spark] class ClientArguments(args: Array[String]) { } object ClientArguments { + private[spark] val DEFAULT_CORES = 1 + private[spark] val DEFAULT_MEMORY = 512 // MB + private[spark] val DEFAULT_SUPERVISE = false + def isValidJarUrl(s: String): Boolean = { try { val uri = new URI(s) diff --git a/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala b/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala index 243d8edb72ed3..7f600d89604a2 100644 --- a/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala +++ b/core/src/main/scala/org/apache/spark/deploy/DeployMessage.scala @@ -148,15 +148,22 @@ private[deploy] object DeployMessages { // Master to MasterWebUI - case class MasterStateResponse(host: String, port: Int, workers: Array[WorkerInfo], - activeApps: Array[ApplicationInfo], completedApps: Array[ApplicationInfo], - activeDrivers: Array[DriverInfo], completedDrivers: Array[DriverInfo], - status: MasterState) { + case class MasterStateResponse( + host: String, + port: Int, + restPort: Option[Int], + workers: Array[WorkerInfo], + activeApps: Array[ApplicationInfo], + completedApps: Array[ApplicationInfo], + activeDrivers: Array[DriverInfo], + completedDrivers: Array[DriverInfo], + status: MasterState) { Utils.checkHost(host, "Required hostname") assert (port > 0) def uri = "spark://" + host + ":" + port + def restUri: Option[String] = restPort.map { p => "spark://" + host + ":" + p } } // WorkerWebUI to Worker diff --git a/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala b/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala index 9a7a113c95715..0401b15446a7b 100644 --- a/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala +++ b/core/src/main/scala/org/apache/spark/deploy/LocalSparkCluster.scala @@ -33,7 +33,11 @@ import org.apache.spark.util.Utils * fault recovery without spinning up a lot of processes. */ private[spark] -class LocalSparkCluster(numWorkers: Int, coresPerWorker: Int, memoryPerWorker: Int) +class LocalSparkCluster( + numWorkers: Int, + coresPerWorker: Int, + memoryPerWorker: Int, + conf: SparkConf) extends Logging { private val localHostname = Utils.localHostName() @@ -43,9 +47,11 @@ class LocalSparkCluster(numWorkers: Int, coresPerWorker: Int, memoryPerWorker: I def start(): Array[String] = { logInfo("Starting a local Spark cluster with " + numWorkers + " workers.") + // Disable REST server on Master in this mode unless otherwise specified + val _conf = conf.clone().setIfMissing("spark.master.rest.enabled", "false") + /* Start the Master */ - val conf = new SparkConf(false) - val (masterSystem, masterPort, _) = Master.startSystemAndActor(localHostname, 0, 0, conf) + val (masterSystem, masterPort, _, _) = Master.startSystemAndActor(localHostname, 0, 0, _conf) masterActorSystems += masterSystem val masterUrl = "spark://" + localHostname + ":" + masterPort val masters = Array(masterUrl) diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala index 9d25e647f1703..c4bc5054d61a1 100644 --- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala +++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala @@ -18,25 +18,35 @@ package org.apache.spark.deploy import java.io.{File, PrintStream} -import java.lang.reflect.{Modifier, InvocationTargetException} +import java.lang.reflect.{InvocationTargetException, Modifier} import java.net.URL + import scala.collection.mutable.{ArrayBuffer, HashMap, Map} import org.apache.hadoop.fs.Path import org.apache.ivy.Ivy import org.apache.ivy.core.LogOptions -import org.apache.ivy.core.module.descriptor.{DefaultExcludeRule, DefaultDependencyDescriptor, DefaultModuleDescriptor} -import org.apache.ivy.core.module.id.{ModuleId, ArtifactId, ModuleRevisionId} +import org.apache.ivy.core.module.descriptor._ +import org.apache.ivy.core.module.id.{ArtifactId, ModuleId, ModuleRevisionId} import org.apache.ivy.core.report.ResolveReport -import org.apache.ivy.core.resolve.{IvyNode, ResolveOptions} +import org.apache.ivy.core.resolve.ResolveOptions import org.apache.ivy.core.retrieve.RetrieveOptions import org.apache.ivy.core.settings.IvySettings import org.apache.ivy.plugins.matcher.GlobPatternMatcher import org.apache.ivy.plugins.resolver.{ChainResolver, IBiblioResolver} -import org.apache.spark.executor.ExecutorURLClassLoader -import org.apache.spark.util.Utils -import org.apache.spark.executor.ChildExecutorURLClassLoader -import org.apache.spark.executor.MutableURLClassLoader + +import org.apache.spark.deploy.rest._ +import org.apache.spark.executor._ +import org.apache.spark.util.{ChildFirstURLClassLoader, MutableURLClassLoader, Utils} + +/** + * Whether to submit, kill, or request the status of an application. + * The latter two operations are currently supported only for standalone cluster mode. + */ +private[spark] object SparkSubmitAction extends Enumeration { + type SparkSubmitAction = Value + val SUBMIT, KILL, REQUEST_STATUS = Value +} /** * Main gateway of launching a Spark application. @@ -83,21 +93,74 @@ object SparkSubmit { if (appArgs.verbose) { printStream.println(appArgs) } - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) - launch(childArgs, classpath, sysProps, mainClass, appArgs.verbose) + appArgs.action match { + case SparkSubmitAction.SUBMIT => submit(appArgs) + case SparkSubmitAction.KILL => kill(appArgs) + case SparkSubmitAction.REQUEST_STATUS => requestStatus(appArgs) + } + } + + /** Kill an existing submission using the REST protocol. Standalone cluster mode only. */ + private def kill(args: SparkSubmitArguments): Unit = { + new StandaloneRestClient() + .killSubmission(args.master, args.submissionToKill) } /** - * @return a tuple containing - * (1) the arguments for the child process, - * (2) a list of classpath entries for the child, - * (3) a list of system properties and env vars, and - * (4) the main class for the child + * Request the status of an existing submission using the REST protocol. + * Standalone cluster mode only. */ - private[spark] def createLaunchEnv(args: SparkSubmitArguments) - : (ArrayBuffer[String], ArrayBuffer[String], Map[String, String], String) = { + private def requestStatus(args: SparkSubmitArguments): Unit = { + new StandaloneRestClient() + .requestSubmissionStatus(args.master, args.submissionToRequestStatusFor) + } - // Values to return + /** + * Submit the application using the provided parameters. + * + * This runs in two steps. First, we prepare the launch environment by setting up + * the appropriate classpath, system properties, and application arguments for + * running the child main class based on the cluster manager and the deploy mode. + * Second, we use this launch environment to invoke the main method of the child + * main class. + */ + private[spark] def submit(args: SparkSubmitArguments): Unit = { + val (childArgs, childClasspath, sysProps, childMainClass) = prepareSubmitEnvironment(args) + // In standalone cluster mode, there are two submission gateways: + // (1) The traditional Akka gateway using o.a.s.deploy.Client as a wrapper + // (2) The new REST-based gateway introduced in Spark 1.3 + // The latter is the default behavior as of Spark 1.3, but Spark submit will fail over + // to use the legacy gateway if the master endpoint turns out to be not a REST server. + if (args.isStandaloneCluster && args.useRest) { + try { + printStream.println("Running Spark using the REST application submission protocol.") + runMain(childArgs, childClasspath, sysProps, childMainClass) + } catch { + // Fail over to use the legacy submission gateway + case e: SubmitRestConnectionException => + printWarning(s"Master endpoint ${args.master} was not a REST server. " + + "Falling back to legacy submission gateway instead.") + args.useRest = false + submit(args) + } + // In all other modes, just run the main class as prepared + } else { + runMain(childArgs, childClasspath, sysProps, childMainClass) + } + } + + /** + * Prepare the environment for submitting an application. + * This returns a 4-tuple: + * (1) the arguments for the child process, + * (2) a list of classpath entries for the child, + * (3) a map of system properties, and + * (4) the main class for the child + * Exposed for testing. + */ + private[spark] def prepareSubmitEnvironment(args: SparkSubmitArguments) + : (Seq[String], Seq[String], Map[String, String], String) = { + // Return values val childArgs = new ArrayBuffer[String]() val childClasspath = new ArrayBuffer[String]() val sysProps = new HashMap[String, String]() @@ -235,10 +298,13 @@ object SparkSubmit { sysProp = "spark.driver.extraLibraryPath"), // Standalone cluster only + // Do not set CL arguments here because there are multiple possibilities for the main class OptionAssigner(args.jars, STANDALONE, CLUSTER, sysProp = "spark.jars"), OptionAssigner(args.ivyRepoPath, STANDALONE, CLUSTER, sysProp = "spark.jars.ivy"), - OptionAssigner(args.driverMemory, STANDALONE, CLUSTER, clOption = "--memory"), - OptionAssigner(args.driverCores, STANDALONE, CLUSTER, clOption = "--cores"), + OptionAssigner(args.driverMemory, STANDALONE, CLUSTER, sysProp = "spark.driver.memory"), + OptionAssigner(args.driverCores, STANDALONE, CLUSTER, sysProp = "spark.driver.cores"), + OptionAssigner(args.supervise.toString, STANDALONE, CLUSTER, + sysProp = "spark.driver.supervise"), // Yarn client only OptionAssigner(args.queue, YARN, CLIENT, sysProp = "spark.yarn.queue"), @@ -279,7 +345,6 @@ object SparkSubmit { if (args.childArgs != null) { childArgs ++= args.childArgs } } - // Map all arguments to command-line options or system properties for our chosen mode for (opt <- options) { if (opt.value != null && @@ -301,14 +366,21 @@ object SparkSubmit { sysProps.put("spark.jars", jars.mkString(",")) } - // In standalone-cluster mode, use Client as a wrapper around the user class - if (clusterManager == STANDALONE && deployMode == CLUSTER) { - childMainClass = "org.apache.spark.deploy.Client" - if (args.supervise) { - childArgs += "--supervise" + // In standalone cluster mode, use the REST client to submit the application (Spark 1.3+). + // All Spark parameters are expected to be passed to the client through system properties. + if (args.isStandaloneCluster) { + if (args.useRest) { + childMainClass = "org.apache.spark.deploy.rest.StandaloneRestClient" + childArgs += (args.primaryResource, args.mainClass) + } else { + // In legacy standalone cluster mode, use Client as a wrapper around the user class + childMainClass = "org.apache.spark.deploy.Client" + if (args.supervise) { childArgs += "--supervise" } + Option(args.driverMemory).foreach { m => childArgs += ("--memory", m) } + Option(args.driverCores).foreach { c => childArgs += ("--cores", c) } + childArgs += "launch" + childArgs += (args.master, args.primaryResource, args.mainClass) } - childArgs += "launch" - childArgs += (args.master, args.primaryResource, args.mainClass) if (args.childArgs != null) { childArgs ++= args.childArgs } @@ -345,7 +417,7 @@ object SparkSubmit { // Ignore invalid spark.driver.host in cluster modes. if (deployMode == CLUSTER) { - sysProps -= ("spark.driver.host") + sysProps -= "spark.driver.host" } // Resolve paths in certain spark properties @@ -374,9 +446,15 @@ object SparkSubmit { (childArgs, childClasspath, sysProps, childMainClass) } - private def launch( - childArgs: ArrayBuffer[String], - childClasspath: ArrayBuffer[String], + /** + * Run the main method of the child class using the provided launch environment. + * + * Note that this main class will not be the one provided by the user if we're + * running cluster deploy mode or python applications. + */ + private def runMain( + childArgs: Seq[String], + childClasspath: Seq[String], sysProps: Map[String, String], childMainClass: String, verbose: Boolean = false) { @@ -389,11 +467,11 @@ object SparkSubmit { } val loader = - if (sysProps.getOrElse("spark.files.userClassPathFirst", "false").toBoolean) { - new ChildExecutorURLClassLoader(new Array[URL](0), + if (sysProps.getOrElse("spark.driver.userClassPathFirst", "false").toBoolean) { + new ChildFirstURLClassLoader(new Array[URL](0), Thread.currentThread.getContextClassLoader) } else { - new ExecutorURLClassLoader(new Array[URL](0), + new MutableURLClassLoader(new Array[URL](0), Thread.currentThread.getContextClassLoader) } Thread.currentThread.setContextClassLoader(loader) @@ -697,7 +775,7 @@ private[spark] object SparkSubmitUtils { * Provides an indirection layer for passing arguments as system properties or flags to * the user's driver program or to downstream launcher tools. */ -private[spark] case class OptionAssigner( +private case class OptionAssigner( value: String, clusterManager: Int, deployMode: Int, diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala index 5cadc534f4baa..bd0ae26fd8210 100644 --- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala +++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala @@ -22,6 +22,7 @@ import java.util.jar.JarFile import scala.collection.mutable.{ArrayBuffer, HashMap} +import org.apache.spark.deploy.SparkSubmitAction._ import org.apache.spark.util.Utils /** @@ -39,8 +40,6 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St var driverExtraClassPath: String = null var driverExtraLibraryPath: String = null var driverExtraJavaOptions: String = null - var driverCores: String = null - var supervise: Boolean = false var queue: String = null var numExecutors: String = null var files: String = null @@ -56,8 +55,16 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St var verbose: Boolean = false var isPython: Boolean = false var pyFiles: String = null + var action: SparkSubmitAction = null val sparkProperties: HashMap[String, String] = new HashMap[String, String]() + // Standalone cluster mode only + var supervise: Boolean = false + var driverCores: String = null + var submissionToKill: String = null + var submissionToRequestStatusFor: String = null + var useRest: Boolean = true // used internally + /** Default properties present in the currently defined defaults file. */ lazy val defaultSparkProperties: HashMap[String, String] = { val defaultProperties = new HashMap[String, String]() @@ -82,7 +89,7 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St // Use `sparkProperties` map along with env vars to fill in any missing parameters loadEnvironmentArguments() - checkRequiredArguments() + validateArguments() /** * Merge values from the default properties file with those specified through --conf. @@ -107,6 +114,15 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St .orElse(sparkProperties.get("spark.master")) .orElse(env.get("MASTER")) .orNull + driverExtraClassPath = Option(driverExtraClassPath) + .orElse(sparkProperties.get("spark.driver.extraClassPath")) + .orNull + driverExtraJavaOptions = Option(driverExtraJavaOptions) + .orElse(sparkProperties.get("spark.driver.extraJavaOptions")) + .orNull + driverExtraLibraryPath = Option(driverExtraLibraryPath) + .orElse(sparkProperties.get("spark.driver.extraLibraryPath")) + .orNull driverMemory = Option(driverMemory) .orElse(sparkProperties.get("spark.driver.memory")) .orElse(env.get("SPARK_DRIVER_MEMORY")) @@ -166,10 +182,21 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St if (name == null && primaryResource != null) { name = Utils.stripDirectory(primaryResource) } + + // Action should be SUBMIT unless otherwise specified + action = Option(action).getOrElse(SUBMIT) } /** Ensure that required fields exists. Call this only once all defaults are loaded. */ - private def checkRequiredArguments(): Unit = { + private def validateArguments(): Unit = { + action match { + case SUBMIT => validateSubmitArguments() + case KILL => validateKillArguments() + case REQUEST_STATUS => validateStatusRequestArguments() + } + } + + private def validateSubmitArguments(): Unit = { if (args.length == 0) { printUsageAndExit(-1) } @@ -192,6 +219,29 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St } } + private def validateKillArguments(): Unit = { + if (!master.startsWith("spark://")) { + SparkSubmit.printErrorAndExit("Killing submissions is only supported in standalone mode!") + } + if (submissionToKill == null) { + SparkSubmit.printErrorAndExit("Please specify a submission to kill.") + } + } + + private def validateStatusRequestArguments(): Unit = { + if (!master.startsWith("spark://")) { + SparkSubmit.printErrorAndExit( + "Requesting submission statuses is only supported in standalone mode!") + } + if (submissionToRequestStatusFor == null) { + SparkSubmit.printErrorAndExit("Please specify a submission to request status for.") + } + } + + def isStandaloneCluster: Boolean = { + master.startsWith("spark://") && deployMode == "cluster" + } + override def toString = { s"""Parsed arguments: | master $master @@ -300,6 +350,22 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St propertiesFile = value parse(tail) + case ("--kill") :: value :: tail => + submissionToKill = value + if (action != null) { + SparkSubmit.printErrorAndExit(s"Action cannot be both $action and $KILL.") + } + action = KILL + parse(tail) + + case ("--status") :: value :: tail => + submissionToRequestStatusFor = value + if (action != null) { + SparkSubmit.printErrorAndExit(s"Action cannot be both $action and $REQUEST_STATUS.") + } + action = REQUEST_STATUS + parse(tail) + case ("--supervise") :: tail => supervise = true parse(tail) @@ -372,7 +438,10 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St outStream.println("Unknown/unsupported param " + unknownParam) } outStream.println( - """Usage: spark-submit [options] [app options] + """Usage: spark-submit [options] [app arguments] + |Usage: spark-submit --kill [submission ID] --master [spark://...] + |Usage: spark-submit --status [submission ID] --master [spark://...] + | |Options: | --master MASTER_URL spark://host:port, mesos://host:port, yarn, or local. | --deploy-mode DEPLOY_MODE Whether to launch the driver program locally ("client") or @@ -413,6 +482,8 @@ private[spark] class SparkSubmitArguments(args: Seq[String], env: Map[String, St | Spark standalone with cluster deploy mode only: | --driver-cores NUM Cores for driver (Default: 1). | --supervise If given, restarts the driver on failure. + | --kill SUBMISSION_ID If given, kills the driver specified. + | --status SUBMISSION_ID If given, requests the status of the driver specified. | | Spark standalone and Mesos only: | --total-executor-cores NUM Total cores for all executors. diff --git a/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala b/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala index 92125f2df7d10..868c63d30a202 100644 --- a/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala +++ b/core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala @@ -194,7 +194,7 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis None } } - .sortBy { info => (-info.endTime, -info.startTime) } + .sortWith(compareAppInfo) lastModifiedTime = newLastModifiedTime @@ -214,7 +214,7 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis val newIterator = logInfos.iterator.buffered val oldIterator = applications.values.iterator.buffered while (newIterator.hasNext && oldIterator.hasNext) { - if (newIterator.head.endTime > oldIterator.head.endTime) { + if (compareAppInfo(newIterator.head, oldIterator.head)) { addIfAbsent(newIterator.next) } else { addIfAbsent(oldIterator.next) @@ -230,6 +230,17 @@ private[history] class FsHistoryProvider(conf: SparkConf) extends ApplicationHis } } + /** + * Comparison function that defines the sort order for the application listing. + * + * @return Whether `i1` should precede `i2`. + */ + private def compareAppInfo( + i1: FsApplicationHistoryInfo, + i2: FsApplicationHistoryInfo): Boolean = { + if (i1.endTime != i2.endTime) i1.endTime >= i2.endTime else i1.startTime >= i2.startTime + } + /** * Replays the events in the specified log file and returns information about the associated * application. diff --git a/core/src/main/scala/org/apache/spark/deploy/master/Master.scala b/core/src/main/scala/org/apache/spark/deploy/master/Master.scala index 5eeb9fe526248..53e453990f8c7 100644 --- a/core/src/main/scala/org/apache/spark/deploy/master/Master.scala +++ b/core/src/main/scala/org/apache/spark/deploy/master/Master.scala @@ -43,6 +43,7 @@ import org.apache.spark.deploy.history.HistoryServer import org.apache.spark.deploy.master.DriverState.DriverState import org.apache.spark.deploy.master.MasterMessages._ import org.apache.spark.deploy.master.ui.MasterWebUI +import org.apache.spark.deploy.rest.StandaloneRestServer import org.apache.spark.metrics.MetricsSystem import org.apache.spark.scheduler.{EventLoggingListener, ReplayListenerBus} import org.apache.spark.ui.SparkUI @@ -52,12 +53,12 @@ private[spark] class Master( host: String, port: Int, webUiPort: Int, - val securityMgr: SecurityManager) + val securityMgr: SecurityManager, + val conf: SparkConf) extends Actor with ActorLogReceive with Logging with LeaderElectable { import context.dispatcher // to use Akka's scheduler.schedule() - val conf = new SparkConf val hadoopConf = SparkHadoopUtil.get.newConfiguration(conf) def createDateFormat = new SimpleDateFormat("yyyyMMddHHmmss") // For application IDs @@ -121,6 +122,17 @@ private[spark] class Master( throw new SparkException("spark.deploy.defaultCores must be positive") } + // Alternative application submission gateway that is stable across Spark versions + private val restServerEnabled = conf.getBoolean("spark.master.rest.enabled", true) + private val restServer = + if (restServerEnabled) { + val port = conf.getInt("spark.master.rest.port", 6066) + Some(new StandaloneRestServer(host, port, self, masterUrl, conf)) + } else { + None + } + private val restServerBoundPort = restServer.map(_.start()) + override def preStart() { logInfo("Starting Spark master at " + masterUrl) logInfo(s"Running Spark version ${org.apache.spark.SPARK_VERSION}") @@ -174,6 +186,7 @@ private[spark] class Master( recoveryCompletionTask.cancel() } webUi.stop() + restServer.foreach(_.stop()) masterMetricsSystem.stop() applicationMetricsSystem.stop() persistenceEngine.close() @@ -421,7 +434,9 @@ private[spark] class Master( } case RequestMasterState => { - sender ! MasterStateResponse(host, port, workers.toArray, apps.toArray, completedApps.toArray, + sender ! MasterStateResponse( + host, port, restServerBoundPort, + workers.toArray, apps.toArray, completedApps.toArray, drivers.toArray, completedDrivers.toArray, state) } @@ -429,8 +444,8 @@ private[spark] class Master( timeOutDeadWorkers() } - case RequestWebUIPort => { - sender ! WebUIPortResponse(webUi.boundPort) + case BoundPortsRequest => { + sender ! BoundPortsResponse(port, webUi.boundPort, restServerBoundPort) } } @@ -656,7 +671,7 @@ private[spark] class Master( def registerApplication(app: ApplicationInfo): Unit = { val appAddress = app.driver.path.address - if (addressToWorker.contains(appAddress)) { + if (addressToApp.contains(appAddress)) { logInfo("Attempted to re-register application at same address: " + appAddress) return } @@ -851,7 +866,7 @@ private[spark] object Master extends Logging { SignalLogger.register(log) val conf = new SparkConf val args = new MasterArguments(argStrings, conf) - val (actorSystem, _, _) = startSystemAndActor(args.host, args.port, args.webUiPort, conf) + val (actorSystem, _, _, _) = startSystemAndActor(args.host, args.port, args.webUiPort, conf) actorSystem.awaitTermination() } @@ -875,19 +890,26 @@ private[spark] object Master extends Logging { Address(protocol, systemName, host, port) } + /** + * Start the Master and return a four tuple of: + * (1) The Master actor system + * (2) The bound port + * (3) The web UI bound port + * (4) The REST server bound port, if any + */ def startSystemAndActor( host: String, port: Int, webUiPort: Int, - conf: SparkConf): (ActorSystem, Int, Int) = { + conf: SparkConf): (ActorSystem, Int, Int, Option[Int]) = { val securityMgr = new SecurityManager(conf) val (actorSystem, boundPort) = AkkaUtils.createActorSystem(systemName, host, port, conf = conf, securityManager = securityMgr) - val actor = actorSystem.actorOf(Props(classOf[Master], host, boundPort, webUiPort, - securityMgr), actorName) + val actor = actorSystem.actorOf( + Props(classOf[Master], host, boundPort, webUiPort, securityMgr, conf), actorName) val timeout = AkkaUtils.askTimeout(conf) - val respFuture = actor.ask(RequestWebUIPort)(timeout) - val resp = Await.result(respFuture, timeout).asInstanceOf[WebUIPortResponse] - (actorSystem, boundPort, resp.webUIBoundPort) + val portsRequest = actor.ask(BoundPortsRequest)(timeout) + val portsResponse = Await.result(portsRequest, timeout).asInstanceOf[BoundPortsResponse] + (actorSystem, boundPort, portsResponse.webUIPort, portsResponse.restPort) } } diff --git a/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala b/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala index db72d8ae9bdaf..15c6296888f70 100644 --- a/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala +++ b/core/src/main/scala/org/apache/spark/deploy/master/MasterMessages.scala @@ -36,7 +36,7 @@ private[master] object MasterMessages { case object CompleteRecovery - case object RequestWebUIPort + case object BoundPortsRequest - case class WebUIPortResponse(webUIBoundPort: Int) + case class BoundPortsResponse(actorPort: Int, webUIPort: Int, restPort: Option[Int]) } diff --git a/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala b/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala index 7ca3b08a28728..fd514f07664a9 100644 --- a/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala +++ b/core/src/main/scala/org/apache/spark/deploy/master/ui/MasterPage.scala @@ -46,19 +46,19 @@ private[spark] class MasterPage(parent: MasterWebUI) extends WebUIPage("") { val stateFuture = (master ? RequestMasterState)(timeout).mapTo[MasterStateResponse] val state = Await.result(stateFuture, timeout) - val workerHeaders = Seq("Id", "Address", "State", "Cores", "Memory") + val workerHeaders = Seq("Worker Id", "Address", "State", "Cores", "Memory") val workers = state.workers.sortBy(_.id) val workerTable = UIUtils.listingTable(workerHeaders, workerRow, workers) - val appHeaders = Seq("ID", "Name", "Cores", "Memory per Node", "Submitted Time", "User", - "State", "Duration") + val appHeaders = Seq("Application ID", "Name", "Cores", "Memory per Node", "Submitted Time", + "User", "State", "Duration") val activeApps = state.activeApps.sortBy(_.startTime).reverse val activeAppsTable = UIUtils.listingTable(appHeaders, appRow, activeApps) val completedApps = state.completedApps.sortBy(_.endTime).reverse val completedAppsTable = UIUtils.listingTable(appHeaders, appRow, completedApps) - val driverHeaders = Seq("ID", "Submitted Time", "Worker", "State", "Cores", "Memory", - "Main Class") + val driverHeaders = Seq("Submission ID", "Submitted Time", "Worker", "State", "Cores", + "Memory", "Main Class") val activeDrivers = state.activeDrivers.sortBy(_.startTime).reverse val activeDriversTable = UIUtils.listingTable(driverHeaders, driverRow, activeDrivers) val completedDrivers = state.completedDrivers.sortBy(_.startTime).reverse @@ -73,6 +73,14 @@ private[spark] class MasterPage(parent: MasterWebUI) extends WebUIPage("") {
  • URL: {state.uri}
    • + { + state.restUri.map { uri => +
  • + REST URL: {uri} + (cluster mode) +
    • + }.getOrElse { Seq.empty } + }
  • Workers: {state.workers.size}
  • Cores: {state.workers.map(_.cores).sum} Total, {state.workers.map(_.coresUsed).sum} Used
    • @@ -188,7 +196,7 @@ private[spark] class MasterPage(parent: MasterWebUI) extends WebUIPage("") { {Utils.megabytesToString(driver.desc.mem.toLong)} - {driver.desc.command.arguments(1)} + {driver.desc.command.arguments(2)} } } diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala new file mode 100644 index 0000000000000..115aa5278bb62 --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestClient.scala @@ -0,0 +1,307 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.io.{DataOutputStream, FileNotFoundException} +import java.net.{HttpURLConnection, SocketException, URL} + +import scala.io.Source + +import com.fasterxml.jackson.databind.JsonMappingException +import com.google.common.base.Charsets + +import org.apache.spark.{Logging, SparkConf, SPARK_VERSION => sparkVersion} + +/** + * A client that submits applications to the standalone Master using a REST protocol. + * This client is intended to communicate with the [[StandaloneRestServer]] and is + * currently used for cluster mode only. + * + * In protocol version v1, the REST URL takes the form http://[host:port]/v1/submissions/[action], + * where [action] can be one of create, kill, or status. Each type of request is represented in + * an HTTP message sent to the following prefixes: + * (1) submit - POST to /submissions/create + * (2) kill - POST /submissions/kill/[submissionId] + * (3) status - GET /submissions/status/[submissionId] + * + * In the case of (1), parameters are posted in the HTTP body in the form of JSON fields. + * Otherwise, the URL fully specifies the intended action of the client. + * + * Since the protocol is expected to be stable across Spark versions, existing fields cannot be + * added or removed, though new optional fields can be added. In the rare event that forward or + * backward compatibility is broken, Spark must introduce a new protocol version (e.g. v2). + * + * The client and the server must communicate using the same version of the protocol. If there + * is a mismatch, the server will respond with the highest protocol version it supports. A future + * implementation of this client can use that information to retry using the version specified + * by the server. + */ +private[spark] class StandaloneRestClient extends Logging { + import StandaloneRestClient._ + + /** + * Submit an application specified by the parameters in the provided request. + * + * If the submission was successful, poll the status of the submission and report + * it to the user. Otherwise, report the error message provided by the server. + */ + def createSubmission( + master: String, + request: CreateSubmissionRequest): SubmitRestProtocolResponse = { + logInfo(s"Submitting a request to launch an application in $master.") + validateMaster(master) + val url = getSubmitUrl(master) + val response = postJson(url, request.toJson) + response match { + case s: CreateSubmissionResponse => + reportSubmissionStatus(master, s) + handleRestResponse(s) + case unexpected => + handleUnexpectedRestResponse(unexpected) + } + response + } + + /** Request that the server kill the specified submission. */ + def killSubmission(master: String, submissionId: String): SubmitRestProtocolResponse = { + logInfo(s"Submitting a request to kill submission $submissionId in $master.") + validateMaster(master) + val response = post(getKillUrl(master, submissionId)) + response match { + case k: KillSubmissionResponse => handleRestResponse(k) + case unexpected => handleUnexpectedRestResponse(unexpected) + } + response + } + + /** Request the status of a submission from the server. */ + def requestSubmissionStatus( + master: String, + submissionId: String, + quiet: Boolean = false): SubmitRestProtocolResponse = { + logInfo(s"Submitting a request for the status of submission $submissionId in $master.") + validateMaster(master) + val response = get(getStatusUrl(master, submissionId)) + response match { + case s: SubmissionStatusResponse => if (!quiet) { handleRestResponse(s) } + case unexpected => handleUnexpectedRestResponse(unexpected) + } + response + } + + /** Construct a message that captures the specified parameters for submitting an application. */ + def constructSubmitRequest( + appResource: String, + mainClass: String, + appArgs: Array[String], + sparkProperties: Map[String, String], + environmentVariables: Map[String, String]): CreateSubmissionRequest = { + val message = new CreateSubmissionRequest + message.clientSparkVersion = sparkVersion + message.appResource = appResource + message.mainClass = mainClass + message.appArgs = appArgs + message.sparkProperties = sparkProperties + message.environmentVariables = environmentVariables + message.validate() + message + } + + /** Send a GET request to the specified URL. */ + private def get(url: URL): SubmitRestProtocolResponse = { + logDebug(s"Sending GET request to server at $url.") + val conn = url.openConnection().asInstanceOf[HttpURLConnection] + conn.setRequestMethod("GET") + readResponse(conn) + } + + /** Send a POST request to the specified URL. */ + private def post(url: URL): SubmitRestProtocolResponse = { + logDebug(s"Sending POST request to server at $url.") + val conn = url.openConnection().asInstanceOf[HttpURLConnection] + conn.setRequestMethod("POST") + readResponse(conn) + } + + /** Send a POST request with the given JSON as the body to the specified URL. */ + private def postJson(url: URL, json: String): SubmitRestProtocolResponse = { + logDebug(s"Sending POST request to server at $url:\n$json") + val conn = url.openConnection().asInstanceOf[HttpURLConnection] + conn.setRequestMethod("POST") + conn.setRequestProperty("Content-Type", "application/json") + conn.setRequestProperty("charset", "utf-8") + conn.setDoOutput(true) + val out = new DataOutputStream(conn.getOutputStream) + out.write(json.getBytes(Charsets.UTF_8)) + out.close() + readResponse(conn) + } + + /** + * Read the response from the server and return it as a validated [[SubmitRestProtocolResponse]]. + * If the response represents an error, report the embedded message to the user. + */ + private def readResponse(connection: HttpURLConnection): SubmitRestProtocolResponse = { + try { + val responseJson = Source.fromInputStream(connection.getInputStream).mkString + logDebug(s"Response from the server:\n$responseJson") + val response = SubmitRestProtocolMessage.fromJson(responseJson) + response.validate() + response match { + // If the response is an error, log the message + case error: ErrorResponse => + logError(s"Server responded with error:\n${error.message}") + error + // Otherwise, simply return the response + case response: SubmitRestProtocolResponse => response + case unexpected => + throw new SubmitRestProtocolException( + s"Message received from server was not a response:\n${unexpected.toJson}") + } + } catch { + case unreachable @ (_: FileNotFoundException | _: SocketException) => + throw new SubmitRestConnectionException( + s"Unable to connect to server ${connection.getURL}", unreachable) + case malformed @ (_: SubmitRestProtocolException | _: JsonMappingException) => + throw new SubmitRestProtocolException( + "Malformed response received from server", malformed) + } + } + + /** Return the REST URL for creating a new submission. */ + private def getSubmitUrl(master: String): URL = { + val baseUrl = getBaseUrl(master) + new URL(s"$baseUrl/create") + } + + /** Return the REST URL for killing an existing submission. */ + private def getKillUrl(master: String, submissionId: String): URL = { + val baseUrl = getBaseUrl(master) + new URL(s"$baseUrl/kill/$submissionId") + } + + /** Return the REST URL for requesting the status of an existing submission. */ + private def getStatusUrl(master: String, submissionId: String): URL = { + val baseUrl = getBaseUrl(master) + new URL(s"$baseUrl/status/$submissionId") + } + + /** Return the base URL for communicating with the server, including the protocol version. */ + private def getBaseUrl(master: String): String = { + val masterUrl = master.stripPrefix("spark://").stripSuffix("/") + s"http://$masterUrl/$PROTOCOL_VERSION/submissions" + } + + /** Throw an exception if this is not standalone mode. */ + private def validateMaster(master: String): Unit = { + if (!master.startsWith("spark://")) { + throw new IllegalArgumentException("This REST client is only supported in standalone mode.") + } + } + + /** Report the status of a newly created submission. */ + private def reportSubmissionStatus( + master: String, + submitResponse: CreateSubmissionResponse): Unit = { + if (submitResponse.success) { + val submissionId = submitResponse.submissionId + if (submissionId != null) { + logInfo(s"Submission successfully created as $submissionId. Polling submission state...") + pollSubmissionStatus(master, submissionId) + } else { + // should never happen + logError("Application successfully submitted, but submission ID was not provided!") + } + } else { + val failMessage = Option(submitResponse.message).map { ": " + _ }.getOrElse("") + logError("Application submission failed" + failMessage) + } + } + + /** + * Poll the status of the specified submission and log it. + * This retries up to a fixed number of times before giving up. + */ + private def pollSubmissionStatus(master: String, submissionId: String): Unit = { + (1 to REPORT_DRIVER_STATUS_MAX_TRIES).foreach { _ => + val response = requestSubmissionStatus(master, submissionId, quiet = true) + val statusResponse = response match { + case s: SubmissionStatusResponse => s + case _ => return // unexpected type, let upstream caller handle it + } + if (statusResponse.success) { + val driverState = Option(statusResponse.driverState) + val workerId = Option(statusResponse.workerId) + val workerHostPort = Option(statusResponse.workerHostPort) + val exception = Option(statusResponse.message) + // Log driver state, if present + driverState match { + case Some(state) => logInfo(s"State of driver $submissionId is now $state.") + case _ => logError(s"State of driver $submissionId was not found!") + } + // Log worker node, if present + (workerId, workerHostPort) match { + case (Some(id), Some(hp)) => logInfo(s"Driver is running on worker $id at $hp.") + case _ => + } + // Log exception stack trace, if present + exception.foreach { e => logError(e) } + return + } + Thread.sleep(REPORT_DRIVER_STATUS_INTERVAL) + } + logError(s"Error: Master did not recognize driver $submissionId.") + } + + /** Log the response sent by the server in the REST application submission protocol. */ + private def handleRestResponse(response: SubmitRestProtocolResponse): Unit = { + logInfo(s"Server responded with ${response.messageType}:\n${response.toJson}") + } + + /** Log an appropriate error if the response sent by the server is not of the expected type. */ + private def handleUnexpectedRestResponse(unexpected: SubmitRestProtocolResponse): Unit = { + logError(s"Error: Server responded with message of unexpected type ${unexpected.messageType}.") + } +} + +private[spark] object StandaloneRestClient { + val REPORT_DRIVER_STATUS_INTERVAL = 1000 + val REPORT_DRIVER_STATUS_MAX_TRIES = 10 + val PROTOCOL_VERSION = "v1" + + /** Submit an application, assuming Spark parameters are specified through system properties. */ + def main(args: Array[String]): Unit = { + if (args.size < 2) { + sys.error("Usage: StandaloneRestClient [app resource] [main class] [app args*]") + sys.exit(1) + } + val appResource = args(0) + val mainClass = args(1) + val appArgs = args.slice(2, args.size) + val conf = new SparkConf + val master = conf.getOption("spark.master").getOrElse { + throw new IllegalArgumentException("'spark.master' must be set.") + } + val sparkProperties = conf.getAll.toMap + val environmentVariables = sys.env.filter { case (k, _) => k.startsWith("SPARK_") } + val client = new StandaloneRestClient + val submitRequest = client.constructSubmitRequest( + appResource, mainClass, appArgs, sparkProperties, environmentVariables) + client.createSubmission(master, submitRequest) + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala new file mode 100644 index 0000000000000..6e4486e20fcba --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/StandaloneRestServer.scala @@ -0,0 +1,449 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.io.{DataOutputStream, File} +import java.net.InetSocketAddress +import javax.servlet.http.{HttpServlet, HttpServletRequest, HttpServletResponse} + +import scala.io.Source + +import akka.actor.ActorRef +import com.fasterxml.jackson.databind.JsonMappingException +import com.google.common.base.Charsets +import org.eclipse.jetty.server.Server +import org.eclipse.jetty.servlet.{ServletHolder, ServletContextHandler} +import org.eclipse.jetty.util.thread.QueuedThreadPool +import org.json4s._ +import org.json4s.jackson.JsonMethods._ + +import org.apache.spark.{Logging, SparkConf, SPARK_VERSION => sparkVersion} +import org.apache.spark.util.{AkkaUtils, Utils} +import org.apache.spark.deploy.{Command, DeployMessages, DriverDescription} +import org.apache.spark.deploy.ClientArguments._ + +/** + * A server that responds to requests submitted by the [[StandaloneRestClient]]. + * This is intended to be embedded in the standalone Master and used in cluster mode only. + * + * This server responds with different HTTP codes depending on the situation: + * 200 OK - Request was processed successfully + * 400 BAD REQUEST - Request was malformed, not successfully validated, or of unexpected type + * 468 UNKNOWN PROTOCOL VERSION - Request specified a protocol this server does not understand + * 500 INTERNAL SERVER ERROR - Server throws an exception internally while processing the request + * + * The server always includes a JSON representation of the relevant [[SubmitRestProtocolResponse]] + * in the HTTP body. If an error occurs, however, the server will include an [[ErrorResponse]] + * instead of the one expected by the client. If the construction of this error response itself + * fails, the response will consist of an empty body with a response code that indicates internal + * server error. + * + * @param host the address this server should bind to + * @param requestedPort the port this server will attempt to bind to + * @param masterActor reference to the Master actor to which requests can be sent + * @param masterUrl the URL of the Master new drivers will attempt to connect to + * @param masterConf the conf used by the Master + */ +private[spark] class StandaloneRestServer( + host: String, + requestedPort: Int, + masterActor: ActorRef, + masterUrl: String, + masterConf: SparkConf) + extends Logging { + + import StandaloneRestServer._ + + private var _server: Option[Server] = None + private val baseContext = s"/$PROTOCOL_VERSION/submissions" + + // A mapping from servlets to the URL prefixes they are responsible for + private val servletToContext = Map[StandaloneRestServlet, String]( + new SubmitRequestServlet(masterActor, masterUrl, masterConf) -> s"$baseContext/create/*", + new KillRequestServlet(masterActor, masterConf) -> s"$baseContext/kill/*", + new StatusRequestServlet(masterActor, masterConf) -> s"$baseContext/status/*", + new ErrorServlet -> "/" // default handler + ) + + /** Start the server and return the bound port. */ + def start(): Int = { + val (server, boundPort) = Utils.startServiceOnPort[Server](requestedPort, doStart, masterConf) + _server = Some(server) + logInfo(s"Started REST server for submitting applications on port $boundPort") + boundPort + } + + /** + * Map the servlets to their corresponding contexts and attach them to a server. + * Return a 2-tuple of the started server and the bound port. + */ + private def doStart(startPort: Int): (Server, Int) = { + val server = new Server(new InetSocketAddress(host, startPort)) + val threadPool = new QueuedThreadPool + threadPool.setDaemon(true) + server.setThreadPool(threadPool) + val mainHandler = new ServletContextHandler + mainHandler.setContextPath("/") + servletToContext.foreach { case (servlet, prefix) => + mainHandler.addServlet(new ServletHolder(servlet), prefix) + } + server.setHandler(mainHandler) + server.start() + val boundPort = server.getConnectors()(0).getLocalPort + (server, boundPort) + } + + def stop(): Unit = { + _server.foreach(_.stop()) + } +} + +private object StandaloneRestServer { + val PROTOCOL_VERSION = StandaloneRestClient.PROTOCOL_VERSION + val SC_UNKNOWN_PROTOCOL_VERSION = 468 +} + +/** + * An abstract servlet for handling requests passed to the [[StandaloneRestServer]]. + */ +private abstract class StandaloneRestServlet extends HttpServlet with Logging { + + /** Service a request. If an exception is thrown in the process, indicate server error. */ + protected override def service( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + try { + super.service(request, response) + } catch { + case e: Exception => + logError("Exception while handling request", e) + response.setStatus(HttpServletResponse.SC_INTERNAL_SERVER_ERROR) + } + } + + /** + * Serialize the given response message to JSON and send it through the response servlet. + * This validates the response before sending it to ensure it is properly constructed. + */ + protected def sendResponse( + responseMessage: SubmitRestProtocolResponse, + responseServlet: HttpServletResponse): Unit = { + val message = validateResponse(responseMessage, responseServlet) + responseServlet.setContentType("application/json") + responseServlet.setCharacterEncoding("utf-8") + responseServlet.setStatus(HttpServletResponse.SC_OK) + val content = message.toJson.getBytes(Charsets.UTF_8) + val out = new DataOutputStream(responseServlet.getOutputStream) + out.write(content) + out.close() + } + + /** + * Return any fields in the client request message that the server does not know about. + * + * The mechanism for this is to reconstruct the JSON on the server side and compare the + * diff between this JSON and the one generated on the client side. Any fields that are + * only in the client JSON are treated as unexpected. + */ + protected def findUnknownFields( + requestJson: String, + requestMessage: SubmitRestProtocolMessage): Array[String] = { + val clientSideJson = parse(requestJson) + val serverSideJson = parse(requestMessage.toJson) + val Diff(_, _, unknown) = clientSideJson.diff(serverSideJson) + unknown match { + case j: JObject => j.obj.map { case (k, _) => k }.toArray + case _ => Array.empty[String] // No difference + } + } + + /** Return a human readable String representation of the exception. */ + protected def formatException(e: Throwable): String = { + val stackTraceString = e.getStackTrace.map { "\t" + _ }.mkString("\n") + s"$e\n$stackTraceString" + } + + /** Construct an error message to signal the fact that an exception has been thrown. */ + protected def handleError(message: String): ErrorResponse = { + val e = new ErrorResponse + e.serverSparkVersion = sparkVersion + e.message = message + e + } + + /** + * Validate the response to ensure that it is correctly constructed. + * + * If it is, simply return the message as is. Otherwise, return an error response instead + * to propagate the exception back to the client and set the appropriate error code. + */ + private def validateResponse( + responseMessage: SubmitRestProtocolResponse, + responseServlet: HttpServletResponse): SubmitRestProtocolResponse = { + try { + responseMessage.validate() + responseMessage + } catch { + case e: Exception => + responseServlet.setStatus(HttpServletResponse.SC_INTERNAL_SERVER_ERROR) + handleError("Internal server error: " + formatException(e)) + } + } +} + +/** + * A servlet for handling kill requests passed to the [[StandaloneRestServer]]. + */ +private class KillRequestServlet(masterActor: ActorRef, conf: SparkConf) + extends StandaloneRestServlet { + + /** + * If a submission ID is specified in the URL, have the Master kill the corresponding + * driver and return an appropriate response to the client. Otherwise, return error. + */ + protected override def doPost( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + val submissionId = request.getPathInfo.stripPrefix("/") + val responseMessage = + if (submissionId.nonEmpty) { + handleKill(submissionId) + } else { + response.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError("Submission ID is missing in kill request.") + } + sendResponse(responseMessage, response) + } + + private def handleKill(submissionId: String): KillSubmissionResponse = { + val askTimeout = AkkaUtils.askTimeout(conf) + val response = AkkaUtils.askWithReply[DeployMessages.KillDriverResponse]( + DeployMessages.RequestKillDriver(submissionId), masterActor, askTimeout) + val k = new KillSubmissionResponse + k.serverSparkVersion = sparkVersion + k.message = response.message + k.submissionId = submissionId + k.success = response.success + k + } +} + +/** + * A servlet for handling status requests passed to the [[StandaloneRestServer]]. + */ +private class StatusRequestServlet(masterActor: ActorRef, conf: SparkConf) + extends StandaloneRestServlet { + + /** + * If a submission ID is specified in the URL, request the status of the corresponding + * driver from the Master and include it in the response. Otherwise, return error. + */ + protected override def doGet( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + val submissionId = request.getPathInfo.stripPrefix("/") + val responseMessage = + if (submissionId.nonEmpty) { + handleStatus(submissionId) + } else { + response.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError("Submission ID is missing in status request.") + } + sendResponse(responseMessage, response) + } + + private def handleStatus(submissionId: String): SubmissionStatusResponse = { + val askTimeout = AkkaUtils.askTimeout(conf) + val response = AkkaUtils.askWithReply[DeployMessages.DriverStatusResponse]( + DeployMessages.RequestDriverStatus(submissionId), masterActor, askTimeout) + val message = response.exception.map { s"Exception from the cluster:\n" + formatException(_) } + val d = new SubmissionStatusResponse + d.serverSparkVersion = sparkVersion + d.submissionId = submissionId + d.success = response.found + d.driverState = response.state.map(_.toString).orNull + d.workerId = response.workerId.orNull + d.workerHostPort = response.workerHostPort.orNull + d.message = message.orNull + d + } +} + +/** + * A servlet for handling submit requests passed to the [[StandaloneRestServer]]. + */ +private class SubmitRequestServlet( + masterActor: ActorRef, + masterUrl: String, + conf: SparkConf) + extends StandaloneRestServlet { + + /** + * Submit an application to the Master with parameters specified in the request. + * + * The request is assumed to be a [[SubmitRestProtocolRequest]] in the form of JSON. + * If the request is successfully processed, return an appropriate response to the + * client indicating so. Otherwise, return error instead. + */ + protected override def doPost( + requestServlet: HttpServletRequest, + responseServlet: HttpServletResponse): Unit = { + val responseMessage = + try { + val requestMessageJson = Source.fromInputStream(requestServlet.getInputStream).mkString + val requestMessage = SubmitRestProtocolMessage.fromJson(requestMessageJson) + // The response should have already been validated on the client. + // In case this is not true, validate it ourselves to avoid potential NPEs. + requestMessage.validate() + handleSubmit(requestMessageJson, requestMessage, responseServlet) + } catch { + // The client failed to provide a valid JSON, so this is not our fault + case e @ (_: JsonMappingException | _: SubmitRestProtocolException) => + responseServlet.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError("Malformed request: " + formatException(e)) + } + sendResponse(responseMessage, responseServlet) + } + + /** + * Handle the submit request and construct an appropriate response to return to the client. + * + * This assumes that the request message is already successfully validated. + * If the request message is not of the expected type, return error to the client. + */ + private def handleSubmit( + requestMessageJson: String, + requestMessage: SubmitRestProtocolMessage, + responseServlet: HttpServletResponse): SubmitRestProtocolResponse = { + requestMessage match { + case submitRequest: CreateSubmissionRequest => + val askTimeout = AkkaUtils.askTimeout(conf) + val driverDescription = buildDriverDescription(submitRequest) + val response = AkkaUtils.askWithReply[DeployMessages.SubmitDriverResponse]( + DeployMessages.RequestSubmitDriver(driverDescription), masterActor, askTimeout) + val submitResponse = new CreateSubmissionResponse + submitResponse.serverSparkVersion = sparkVersion + submitResponse.message = response.message + submitResponse.success = response.success + submitResponse.submissionId = response.driverId.orNull + val unknownFields = findUnknownFields(requestMessageJson, requestMessage) + if (unknownFields.nonEmpty) { + // If there are fields that the server does not know about, warn the client + submitResponse.unknownFields = unknownFields + } + submitResponse + case unexpected => + responseServlet.setStatus(HttpServletResponse.SC_BAD_REQUEST) + handleError(s"Received message of unexpected type ${unexpected.messageType}.") + } + } + + /** + * Build a driver description from the fields specified in the submit request. + * + * This involves constructing a command that takes into account memory, java options, + * classpath and other settings to launch the driver. This does not currently consider + * fields used by python applications since python is not supported in standalone + * cluster mode yet. + */ + private def buildDriverDescription(request: CreateSubmissionRequest): DriverDescription = { + // Required fields, including the main class because python is not yet supported + val appResource = Option(request.appResource).getOrElse { + throw new SubmitRestMissingFieldException("Application jar is missing.") + } + val mainClass = Option(request.mainClass).getOrElse { + throw new SubmitRestMissingFieldException("Main class is missing.") + } + + // Optional fields + val sparkProperties = request.sparkProperties + val driverMemory = sparkProperties.get("spark.driver.memory") + val driverCores = sparkProperties.get("spark.driver.cores") + val driverExtraJavaOptions = sparkProperties.get("spark.driver.extraJavaOptions") + val driverExtraClassPath = sparkProperties.get("spark.driver.extraClassPath") + val driverExtraLibraryPath = sparkProperties.get("spark.driver.extraLibraryPath") + val superviseDriver = sparkProperties.get("spark.driver.supervise") + val appArgs = request.appArgs + val environmentVariables = request.environmentVariables + + // Construct driver description + val conf = new SparkConf(false) + .setAll(sparkProperties) + .set("spark.master", masterUrl) + val extraClassPath = driverExtraClassPath.toSeq.flatMap(_.split(File.pathSeparator)) + val extraLibraryPath = driverExtraLibraryPath.toSeq.flatMap(_.split(File.pathSeparator)) + val extraJavaOpts = driverExtraJavaOptions.map(Utils.splitCommandString).getOrElse(Seq.empty) + val sparkJavaOpts = Utils.sparkJavaOpts(conf) + val javaOpts = sparkJavaOpts ++ extraJavaOpts + val command = new Command( + "org.apache.spark.deploy.worker.DriverWrapper", + Seq("{{WORKER_URL}}", "{{USER_JAR}}", mainClass) ++ appArgs, // args to the DriverWrapper + environmentVariables, extraClassPath, extraLibraryPath, javaOpts) + val actualDriverMemory = driverMemory.map(Utils.memoryStringToMb).getOrElse(DEFAULT_MEMORY) + val actualDriverCores = driverCores.map(_.toInt).getOrElse(DEFAULT_CORES) + val actualSuperviseDriver = superviseDriver.map(_.toBoolean).getOrElse(DEFAULT_SUPERVISE) + new DriverDescription( + appResource, actualDriverMemory, actualDriverCores, actualSuperviseDriver, command) + } +} + +/** + * A default servlet that handles error cases that are not captured by other servlets. + */ +private class ErrorServlet extends StandaloneRestServlet { + private val serverVersion = StandaloneRestServer.PROTOCOL_VERSION + + /** Service a faulty request by returning an appropriate error message to the client. */ + protected override def service( + request: HttpServletRequest, + response: HttpServletResponse): Unit = { + val path = request.getPathInfo + val parts = path.stripPrefix("/").split("/").toSeq + var versionMismatch = false + var msg = + parts match { + case Nil => + // http://host:port/ + "Missing protocol version." + case `serverVersion` :: Nil => + // http://host:port/correct-version + "Missing the /submissions prefix." + case `serverVersion` :: "submissions" :: Nil => + // http://host:port/correct-version/submissions + "Missing an action: please specify one of /create, /kill, or /status." + case unknownVersion :: _ => + // http://host:port/unknown-version/* + versionMismatch = true + s"Unknown protocol version '$unknownVersion'." + case _ => + // never reached + s"Malformed path $path." + } + msg += s" Please submit requests through http://[host]:[port]/$serverVersion/submissions/..." + val error = handleError(msg) + // If there is a version mismatch, include the highest protocol version that + // this server supports in case the client wants to retry with our version + if (versionMismatch) { + error.highestProtocolVersion = serverVersion + response.setStatus(StandaloneRestServer.SC_UNKNOWN_PROTOCOL_VERSION) + } else { + response.setStatus(HttpServletResponse.SC_BAD_REQUEST) + } + sendResponse(error, response) + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala new file mode 100644 index 0000000000000..d7a0bdbe10778 --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolException.scala @@ -0,0 +1,36 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +/** + * An exception thrown in the REST application submission protocol. + */ +private[spark] class SubmitRestProtocolException(message: String, cause: Throwable = null) + extends Exception(message, cause) + +/** + * An exception thrown if a field is missing from a [[SubmitRestProtocolMessage]]. + */ +private[spark] class SubmitRestMissingFieldException(message: String) + extends SubmitRestProtocolException(message) + +/** + * An exception thrown if the REST client cannot reach the REST server. + */ +private[spark] class SubmitRestConnectionException(message: String, cause: Throwable) + extends SubmitRestProtocolException(message, cause) diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala new file mode 100644 index 0000000000000..b877898231e3e --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolMessage.scala @@ -0,0 +1,146 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import scala.util.Try + +import com.fasterxml.jackson.annotation._ +import com.fasterxml.jackson.annotation.JsonAutoDetect.Visibility +import com.fasterxml.jackson.annotation.JsonInclude.Include +import com.fasterxml.jackson.databind.{DeserializationFeature, ObjectMapper, SerializationFeature} +import com.fasterxml.jackson.module.scala.DefaultScalaModule +import org.json4s.JsonAST._ +import org.json4s.jackson.JsonMethods._ + +import org.apache.spark.util.Utils + +/** + * An abstract message exchanged in the REST application submission protocol. + * + * This message is intended to be serialized to and deserialized from JSON in the exchange. + * Each message can either be a request or a response and consists of three common fields: + * (1) the action, which fully specifies the type of the message + * (2) the Spark version of the client / server + * (3) an optional message + */ +@JsonInclude(Include.NON_NULL) +@JsonAutoDetect(getterVisibility = Visibility.ANY, setterVisibility = Visibility.ANY) +@JsonPropertyOrder(alphabetic = true) +private[spark] abstract class SubmitRestProtocolMessage { + @JsonIgnore + val messageType = Utils.getFormattedClassName(this) + + val action: String = messageType + var message: String = null + + // For JSON deserialization + private def setAction(a: String): Unit = { } + + /** + * Serialize the message to JSON. + * This also ensures that the message is valid and its fields are in the expected format. + */ + def toJson: String = { + validate() + SubmitRestProtocolMessage.mapper.writeValueAsString(this) + } + + /** + * Assert the validity of the message. + * If the validation fails, throw a [[SubmitRestProtocolException]]. + */ + final def validate(): Unit = { + try { + doValidate() + } catch { + case e: Exception => + throw new SubmitRestProtocolException(s"Validation of message $messageType failed!", e) + } + } + + /** Assert the validity of the message */ + protected def doValidate(): Unit = { + if (action == null) { + throw new SubmitRestMissingFieldException(s"The action field is missing in $messageType") + } + } + + /** Assert that the specified field is set in this message. */ + protected def assertFieldIsSet[T](value: T, name: String): Unit = { + if (value == null) { + throw new SubmitRestMissingFieldException(s"'$name' is missing in message $messageType.") + } + } + + /** + * Assert a condition when validating this message. + * If the assertion fails, throw a [[SubmitRestProtocolException]]. + */ + protected def assert(condition: Boolean, failMessage: String): Unit = { + if (!condition) { throw new SubmitRestProtocolException(failMessage) } + } +} + +/** + * Helper methods to process serialized [[SubmitRestProtocolMessage]]s. + */ +private[spark] object SubmitRestProtocolMessage { + private val packagePrefix = this.getClass.getPackage.getName + private val mapper = new ObjectMapper() + .configure(DeserializationFeature.FAIL_ON_UNKNOWN_PROPERTIES, false) + .enable(SerializationFeature.INDENT_OUTPUT) + .registerModule(DefaultScalaModule) + + /** + * Parse the value of the action field from the given JSON. + * If the action field is not found, throw a [[SubmitRestMissingFieldException]]. + */ + def parseAction(json: String): String = { + parse(json).asInstanceOf[JObject].obj + .find { case (f, _) => f == "action" } + .map { case (_, v) => v.asInstanceOf[JString].s } + .getOrElse { + throw new SubmitRestMissingFieldException(s"Action field not found in JSON:\n$json") + } + } + + /** + * Construct a [[SubmitRestProtocolMessage]] from its JSON representation. + * + * This method first parses the action from the JSON and uses it to infer the message type. + * Note that the action must represent one of the [[SubmitRestProtocolMessage]]s defined in + * this package. Otherwise, a [[ClassNotFoundException]] will be thrown. + */ + def fromJson(json: String): SubmitRestProtocolMessage = { + val className = parseAction(json) + val clazz = Class.forName(packagePrefix + "." + className) + .asSubclass[SubmitRestProtocolMessage](classOf[SubmitRestProtocolMessage]) + fromJson(json, clazz) + } + + /** + * Construct a [[SubmitRestProtocolMessage]] from its JSON representation. + * + * This method determines the type of the message from the class provided instead of + * inferring it from the action field. This is useful for deserializing JSON that + * represents custom user-defined messages. + */ + def fromJson[T <: SubmitRestProtocolMessage](json: String, clazz: Class[T]): T = { + mapper.readValue(json, clazz) + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala new file mode 100644 index 0000000000000..9e1fd8c40cabd --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolRequest.scala @@ -0,0 +1,78 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import scala.util.Try + +import org.apache.spark.util.Utils + +/** + * An abstract request sent from the client in the REST application submission protocol. + */ +private[spark] abstract class SubmitRestProtocolRequest extends SubmitRestProtocolMessage { + var clientSparkVersion: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(clientSparkVersion, "clientSparkVersion") + } +} + +/** + * A request to launch a new application in the REST application submission protocol. + */ +private[spark] class CreateSubmissionRequest extends SubmitRestProtocolRequest { + var appResource: String = null + var mainClass: String = null + var appArgs: Array[String] = null + var sparkProperties: Map[String, String] = null + var environmentVariables: Map[String, String] = null + + protected override def doValidate(): Unit = { + super.doValidate() + assert(sparkProperties != null, "No Spark properties set!") + assertFieldIsSet(appResource, "appResource") + assertPropertyIsSet("spark.app.name") + assertPropertyIsBoolean("spark.driver.supervise") + assertPropertyIsNumeric("spark.driver.cores") + assertPropertyIsNumeric("spark.cores.max") + assertPropertyIsMemory("spark.driver.memory") + assertPropertyIsMemory("spark.executor.memory") + } + + private def assertPropertyIsSet(key: String): Unit = + assertFieldIsSet(sparkProperties.getOrElse(key, null), key) + + private def assertPropertyIsBoolean(key: String): Unit = + assertProperty[Boolean](key, "boolean", _.toBoolean) + + private def assertPropertyIsNumeric(key: String): Unit = + assertProperty[Int](key, "numeric", _.toInt) + + private def assertPropertyIsMemory(key: String): Unit = + assertProperty[Int](key, "memory", Utils.memoryStringToMb) + + /** Assert that a Spark property can be converted to a certain type. */ + private def assertProperty[T](key: String, valueType: String, convert: (String => T)): Unit = { + sparkProperties.get(key).foreach { value => + Try(convert(value)).getOrElse { + throw new SubmitRestProtocolException( + s"Property '$key' expected $valueType value: actual was '$value'.") + } + } + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala new file mode 100644 index 0000000000000..16dfe041d4bea --- /dev/null +++ b/core/src/main/scala/org/apache/spark/deploy/rest/SubmitRestProtocolResponse.scala @@ -0,0 +1,85 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.lang.Boolean + +/** + * An abstract response sent from the server in the REST application submission protocol. + */ +private[spark] abstract class SubmitRestProtocolResponse extends SubmitRestProtocolMessage { + var serverSparkVersion: String = null + var success: Boolean = null + var unknownFields: Array[String] = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(serverSparkVersion, "serverSparkVersion") + } +} + +/** + * A response to a [[CreateSubmissionRequest]] in the REST application submission protocol. + */ +private[spark] class CreateSubmissionResponse extends SubmitRestProtocolResponse { + var submissionId: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(success, "success") + } +} + +/** + * A response to a kill request in the REST application submission protocol. + */ +private[spark] class KillSubmissionResponse extends SubmitRestProtocolResponse { + var submissionId: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(submissionId, "submissionId") + assertFieldIsSet(success, "success") + } +} + +/** + * A response to a status request in the REST application submission protocol. + */ +private[spark] class SubmissionStatusResponse extends SubmitRestProtocolResponse { + var submissionId: String = null + var driverState: String = null + var workerId: String = null + var workerHostPort: String = null + + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(submissionId, "submissionId") + assertFieldIsSet(success, "success") + } +} + +/** + * An error response message used in the REST application submission protocol. + */ +private[spark] class ErrorResponse extends SubmitRestProtocolResponse { + // The highest protocol version that the server knows about + // This is set when the client specifies an unknown version + var highestProtocolVersion: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(message, "message") + } +} diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala b/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala index 28e9662db5da9..3e013c32096c5 100644 --- a/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala +++ b/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala @@ -115,9 +115,19 @@ object CommandUtils extends Logging { val userClassPath = command.classPathEntries ++ Seq(classPath) val javaVersion = System.getProperty("java.version") - val permGenOpt = if (!javaVersion.startsWith("1.8")) Some("-XX:MaxPermSize=128m") else None + + val javaOpts = workerLocalOpts ++ command.javaOpts + + val permGenOpt = + if (!javaVersion.startsWith("1.8") && !javaOpts.exists(_.startsWith("-XX:MaxPermSize="))) { + // do not specify -XX:MaxPermSize if it was already specified by user + Some("-XX:MaxPermSize=128m") + } else { + None + } + Seq("-cp", userClassPath.filterNot(_.isEmpty).mkString(File.pathSeparator)) ++ - permGenOpt ++ workerLocalOpts ++ command.javaOpts ++ memoryOpts + permGenOpt ++ javaOpts ++ memoryOpts } /** Spawn a thread that will redirect a given stream to a file */ diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/DriverRunner.scala b/core/src/main/scala/org/apache/spark/deploy/worker/DriverRunner.scala index 28cab36c7b9e2..b964a09bdb218 100644 --- a/core/src/main/scala/org/apache/spark/deploy/worker/DriverRunner.scala +++ b/core/src/main/scala/org/apache/spark/deploy/worker/DriverRunner.scala @@ -74,10 +74,15 @@ private[spark] class DriverRunner( val driverDir = createWorkingDirectory() val localJarFilename = downloadUserJar(driverDir) - // Make sure user application jar is on the classpath + def substituteVariables(argument: String): String = argument match { + case "{{WORKER_URL}}" => workerUrl + case "{{USER_JAR}}" => localJarFilename + case other => other + } + // TODO: If we add ability to submit multiple jars they should also be added here val builder = CommandUtils.buildProcessBuilder(driverDesc.command, driverDesc.mem, - sparkHome.getAbsolutePath, substituteVariables, Seq(localJarFilename)) + sparkHome.getAbsolutePath, substituteVariables) launchDriver(builder, driverDir, driverDesc.supervise) } catch { @@ -111,12 +116,6 @@ private[spark] class DriverRunner( } } - /** Replace variables in a command argument passed to us */ - private def substituteVariables(argument: String): String = argument match { - case "{{WORKER_URL}}" => workerUrl - case other => other - } - /** * Creates the working directory for this driver. * Will throw an exception if there are errors preparing the directory. diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/DriverWrapper.scala b/core/src/main/scala/org/apache/spark/deploy/worker/DriverWrapper.scala index 05e242e6df702..ab467a5ee8c6c 100644 --- a/core/src/main/scala/org/apache/spark/deploy/worker/DriverWrapper.scala +++ b/core/src/main/scala/org/apache/spark/deploy/worker/DriverWrapper.scala @@ -17,10 +17,12 @@ package org.apache.spark.deploy.worker +import java.io.File + import akka.actor._ import org.apache.spark.{SecurityManager, SparkConf} -import org.apache.spark.util.{AkkaUtils, Utils} +import org.apache.spark.util.{AkkaUtils, ChildFirstURLClassLoader, MutableURLClassLoader, Utils} /** * Utility object for launching driver programs such that they share fate with the Worker process. @@ -28,21 +30,31 @@ import org.apache.spark.util.{AkkaUtils, Utils} object DriverWrapper { def main(args: Array[String]) { args.toList match { - case workerUrl :: mainClass :: extraArgs => + case workerUrl :: userJar :: mainClass :: extraArgs => val conf = new SparkConf() val (actorSystem, _) = AkkaUtils.createActorSystem("Driver", Utils.localHostName(), 0, conf, new SecurityManager(conf)) actorSystem.actorOf(Props(classOf[WorkerWatcher], workerUrl), name = "workerWatcher") + val currentLoader = Thread.currentThread.getContextClassLoader + val userJarUrl = new File(userJar).toURI().toURL() + val loader = + if (sys.props.getOrElse("spark.driver.userClassPathFirst", "false").toBoolean) { + new ChildFirstURLClassLoader(Array(userJarUrl), currentLoader) + } else { + new MutableURLClassLoader(Array(userJarUrl), currentLoader) + } + Thread.currentThread.setContextClassLoader(loader) + // Delegate to supplied main class - val clazz = Class.forName(args(1)) + val clazz = Class.forName(mainClass, true, loader) val mainMethod = clazz.getMethod("main", classOf[Array[String]]) mainMethod.invoke(null, extraArgs.toArray[String]) actorSystem.shutdown() case _ => - System.err.println("Usage: DriverWrapper [options]") + System.err.println("Usage: DriverWrapper [options]") System.exit(-1) } } diff --git a/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala b/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala index 3a42f8b157977..dd19e4947db1e 100644 --- a/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala +++ b/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala @@ -17,8 +17,10 @@ package org.apache.spark.executor +import java.net.URL import java.nio.ByteBuffer +import scala.collection.mutable import scala.concurrent.Await import akka.actor.{Actor, ActorSelection, Props} @@ -38,6 +40,7 @@ private[spark] class CoarseGrainedExecutorBackend( executorId: String, hostPort: String, cores: Int, + userClassPath: Seq[URL], env: SparkEnv) extends Actor with ActorLogReceive with ExecutorBackend with Logging { @@ -63,7 +66,7 @@ private[spark] class CoarseGrainedExecutorBackend( case RegisteredExecutor => logInfo("Successfully registered with driver") val (hostname, _) = Utils.parseHostPort(hostPort) - executor = new Executor(executorId, hostname, env, isLocal = false) + executor = new Executor(executorId, hostname, env, userClassPath, isLocal = false) case RegisterExecutorFailed(message) => logError("Slave registration failed: " + message) @@ -117,7 +120,8 @@ private[spark] object CoarseGrainedExecutorBackend extends Logging { hostname: String, cores: Int, appId: String, - workerUrl: Option[String]) { + workerUrl: Option[String], + userClassPath: Seq[URL]) { SignalLogger.register(log) @@ -162,7 +166,7 @@ private[spark] object CoarseGrainedExecutorBackend extends Logging { val sparkHostPort = hostname + ":" + boundPort env.actorSystem.actorOf( Props(classOf[CoarseGrainedExecutorBackend], - driverUrl, executorId, sparkHostPort, cores, env), + driverUrl, executorId, sparkHostPort, cores, userClassPath, env), name = "Executor") workerUrl.foreach { url => env.actorSystem.actorOf(Props(classOf[WorkerWatcher], url), name = "WorkerWatcher") @@ -172,20 +176,69 @@ private[spark] object CoarseGrainedExecutorBackend extends Logging { } def main(args: Array[String]) { - args.length match { - case x if x < 5 => - System.err.println( + var driverUrl: String = null + var executorId: String = null + var hostname: String = null + var cores: Int = 0 + var appId: String = null + var workerUrl: Option[String] = None + val userClassPath = new mutable.ListBuffer[URL]() + + var argv = args.toList + while (!argv.isEmpty) { + argv match { + case ("--driver-url") :: value :: tail => + driverUrl = value + argv = tail + case ("--executor-id") :: value :: tail => + executorId = value + argv = tail + case ("--hostname") :: value :: tail => + hostname = value + argv = tail + case ("--cores") :: value :: tail => + cores = value.toInt + argv = tail + case ("--app-id") :: value :: tail => + appId = value + argv = tail + case ("--worker-url") :: value :: tail => // Worker url is used in spark standalone mode to enforce fate-sharing with worker - "Usage: CoarseGrainedExecutorBackend " + - " [] ") - System.exit(1) + workerUrl = Some(value) + argv = tail + case ("--user-class-path") :: value :: tail => + userClassPath += new URL(value) + argv = tail + case Nil => + case tail => + System.err.println(s"Unrecognized options: ${tail.mkString(" ")}") + printUsageAndExit() + } + } - // NB: These arguments are provided by SparkDeploySchedulerBackend (for standalone mode) - // and CoarseMesosSchedulerBackend (for mesos mode). - case 5 => - run(args(0), args(1), args(2), args(3).toInt, args(4), None) - case x if x > 5 => - run(args(0), args(1), args(2), args(3).toInt, args(4), Some(args(5))) + if (driverUrl == null || executorId == null || hostname == null || cores <= 0 || + appId == null) { + printUsageAndExit() } + + run(driverUrl, executorId, hostname, cores, appId, workerUrl, userClassPath) } + + private def printUsageAndExit() = { + System.err.println( + """ + |"Usage: CoarseGrainedExecutorBackend [options] + | + | Options are: + | --driver-url + | --executor-id + | --hostname + | --cores + | --app-id + | --worker-url + | --user-class-path + |""".stripMargin) + System.exit(1) + } + } diff --git a/core/src/main/scala/org/apache/spark/executor/Executor.scala b/core/src/main/scala/org/apache/spark/executor/Executor.scala index 5141483d1e745..6b22dcd6f5cbf 100644 --- a/core/src/main/scala/org/apache/spark/executor/Executor.scala +++ b/core/src/main/scala/org/apache/spark/executor/Executor.scala @@ -19,6 +19,7 @@ package org.apache.spark.executor import java.io.File import java.lang.management.ManagementFactory +import java.net.URL import java.nio.ByteBuffer import java.util.concurrent._ @@ -33,7 +34,8 @@ import org.apache.spark.deploy.SparkHadoopUtil import org.apache.spark.scheduler._ import org.apache.spark.shuffle.FetchFailedException import org.apache.spark.storage.{StorageLevel, TaskResultBlockId} -import org.apache.spark.util.{SparkUncaughtExceptionHandler, AkkaUtils, Utils} +import org.apache.spark.util.{ChildFirstURLClassLoader, MutableURLClassLoader, + SparkUncaughtExceptionHandler, AkkaUtils, Utils} /** * Spark executor used with Mesos, YARN, and the standalone scheduler. @@ -43,6 +45,7 @@ private[spark] class Executor( executorId: String, executorHostname: String, env: SparkEnv, + userClassPath: Seq[URL] = Nil, isLocal: Boolean = false) extends Logging { @@ -288,17 +291,23 @@ private[spark] class Executor( * created by the interpreter to the search path */ private def createClassLoader(): MutableURLClassLoader = { + // Bootstrap the list of jars with the user class path. + val now = System.currentTimeMillis() + userClassPath.foreach { url => + currentJars(url.getPath().split("/").last) = now + } + val currentLoader = Utils.getContextOrSparkClassLoader // For each of the jars in the jarSet, add them to the class loader. // We assume each of the files has already been fetched. - val urls = currentJars.keySet.map { uri => + val urls = userClassPath.toArray ++ currentJars.keySet.map { uri => new File(uri.split("/").last).toURI.toURL - }.toArray - val userClassPathFirst = conf.getBoolean("spark.files.userClassPathFirst", false) - userClassPathFirst match { - case true => new ChildExecutorURLClassLoader(urls, currentLoader) - case false => new ExecutorURLClassLoader(urls, currentLoader) + } + if (conf.getBoolean("spark.executor.userClassPathFirst", false)) { + new ChildFirstURLClassLoader(urls, currentLoader) + } else { + new MutableURLClassLoader(urls, currentLoader) } } @@ -311,7 +320,7 @@ private[spark] class Executor( if (classUri != null) { logInfo("Using REPL class URI: " + classUri) val userClassPathFirst: java.lang.Boolean = - conf.getBoolean("spark.files.userClassPathFirst", false) + conf.getBoolean("spark.executor.userClassPathFirst", false) try { val klass = Class.forName("org.apache.spark.repl.ExecutorClassLoader") .asInstanceOf[Class[_ <: ClassLoader]] @@ -344,18 +353,23 @@ private[spark] class Executor( env.securityManager, hadoopConf, timestamp, useCache = !isLocal) currentFiles(name) = timestamp } - for ((name, timestamp) <- newJars if currentJars.getOrElse(name, -1L) < timestamp) { - logInfo("Fetching " + name + " with timestamp " + timestamp) - // Fetch file with useCache mode, close cache for local mode. - Utils.fetchFile(name, new File(SparkFiles.getRootDirectory), conf, - env.securityManager, hadoopConf, timestamp, useCache = !isLocal) - currentJars(name) = timestamp - // Add it to our class loader + for ((name, timestamp) <- newJars) { val localName = name.split("/").last - val url = new File(SparkFiles.getRootDirectory, localName).toURI.toURL - if (!urlClassLoader.getURLs.contains(url)) { - logInfo("Adding " + url + " to class loader") - urlClassLoader.addURL(url) + val currentTimeStamp = currentJars.get(name) + .orElse(currentJars.get(localName)) + .getOrElse(-1L) + if (currentTimeStamp < timestamp) { + logInfo("Fetching " + name + " with timestamp " + timestamp) + // Fetch file with useCache mode, close cache for local mode. + Utils.fetchFile(name, new File(SparkFiles.getRootDirectory), conf, + env.securityManager, hadoopConf, timestamp, useCache = !isLocal) + currentJars(name) = timestamp + // Add it to our class loader + val url = new File(SparkFiles.getRootDirectory, localName).toURI.toURL + if (!urlClassLoader.getURLs.contains(url)) { + logInfo("Adding " + url + " to class loader") + urlClassLoader.addURL(url) + } } } } diff --git a/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala b/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala deleted file mode 100644 index 8011e75944aac..0000000000000 --- a/core/src/main/scala/org/apache/spark/executor/ExecutorURLClassLoader.scala +++ /dev/null @@ -1,84 +0,0 @@ -/* - * Licensed to the Apache Software Foundation (ASF) under one or more - * contributor license agreements. See the NOTICE file distributed with - * this work for additional information regarding copyright ownership. - * The ASF licenses this file to You under the Apache License, Version 2.0 - * (the "License"); you may not use this file except in compliance with - * the License. You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -package org.apache.spark.executor - -import java.net.{URLClassLoader, URL} - -import org.apache.spark.util.ParentClassLoader - -/** - * The addURL method in URLClassLoader is protected. We subclass it to make this accessible. - * We also make changes so user classes can come before the default classes. - */ - -private[spark] trait MutableURLClassLoader extends ClassLoader { - def addURL(url: URL) - def getURLs: Array[URL] -} - -private[spark] class ChildExecutorURLClassLoader(urls: Array[URL], parent: ClassLoader) - extends MutableURLClassLoader { - - private object userClassLoader extends URLClassLoader(urls, null){ - override def addURL(url: URL) { - super.addURL(url) - } - override def findClass(name: String): Class[_] = { - val loaded = super.findLoadedClass(name) - if (loaded != null) { - return loaded - } - try { - super.findClass(name) - } catch { - case e: ClassNotFoundException => { - parentClassLoader.loadClass(name) - } - } - } - } - - private val parentClassLoader = new ParentClassLoader(parent) - - override def findClass(name: String): Class[_] = { - try { - userClassLoader.findClass(name) - } catch { - case e: ClassNotFoundException => { - parentClassLoader.loadClass(name) - } - } - } - - def addURL(url: URL) { - userClassLoader.addURL(url) - } - - def getURLs() = { - userClassLoader.getURLs() - } -} - -private[spark] class ExecutorURLClassLoader(urls: Array[URL], parent: ClassLoader) - extends URLClassLoader(urls, parent) with MutableURLClassLoader { - - override def addURL(url: URL) { - super.addURL(url) - } -} - diff --git a/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala b/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala index 97912c68c5982..bf3f1e4fc7832 100644 --- a/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala +++ b/core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala @@ -177,8 +177,8 @@ class TaskMetrics extends Serializable { * Once https://issues.apache.org/jira/browse/SPARK-5225 is addressed, * we can store all the different inputMetrics (one per readMethod). */ - private[spark] def getInputMetricsForReadMethod(readMethod: DataReadMethod): - InputMetrics =synchronized { + private[spark] def getInputMetricsForReadMethod( + readMethod: DataReadMethod): InputMetrics = synchronized { _inputMetrics match { case None => val metrics = new InputMetrics(readMethod) @@ -194,18 +194,21 @@ class TaskMetrics extends Serializable { /** * Aggregates shuffle read metrics for all registered dependencies into shuffleReadMetrics. */ - private[spark] def updateShuffleReadMetrics() = synchronized { - val merged = new ShuffleReadMetrics() - for (depMetrics <- depsShuffleReadMetrics) { - merged.incFetchWaitTime(depMetrics.fetchWaitTime) - merged.incLocalBlocksFetched(depMetrics.localBlocksFetched) - merged.incRemoteBlocksFetched(depMetrics.remoteBlocksFetched) - merged.incRemoteBytesRead(depMetrics.remoteBytesRead) + private[spark] def updateShuffleReadMetrics(): Unit = synchronized { + if (!depsShuffleReadMetrics.isEmpty) { + val merged = new ShuffleReadMetrics() + for (depMetrics <- depsShuffleReadMetrics) { + merged.incFetchWaitTime(depMetrics.fetchWaitTime) + merged.incLocalBlocksFetched(depMetrics.localBlocksFetched) + merged.incRemoteBlocksFetched(depMetrics.remoteBlocksFetched) + merged.incRemoteBytesRead(depMetrics.remoteBytesRead) + merged.incRecordsRead(depMetrics.recordsRead) + } + _shuffleReadMetrics = Some(merged) } - _shuffleReadMetrics = Some(merged) } - private[spark] def updateInputMetrics() = synchronized { + private[spark] def updateInputMetrics(): Unit = synchronized { inputMetrics.foreach(_.updateBytesRead()) } } @@ -242,27 +245,31 @@ object DataWriteMethod extends Enumeration with Serializable { @DeveloperApi case class InputMetrics(readMethod: DataReadMethod.Value) { - private val _bytesRead: AtomicLong = new AtomicLong() + /** + * This is volatile so that it is visible to the updater thread. + */ + @volatile @transient var bytesReadCallback: Option[() => Long] = None /** * Total bytes read. */ - def bytesRead: Long = _bytesRead.get() - @volatile @transient var bytesReadCallback: Option[() => Long] = None + private var _bytesRead: Long = _ + def bytesRead: Long = _bytesRead + def incBytesRead(bytes: Long) = _bytesRead += bytes /** - * Adds additional bytes read for this read method. + * Total records read. */ - def addBytesRead(bytes: Long) = { - _bytesRead.addAndGet(bytes) - } + private var _recordsRead: Long = _ + def recordsRead: Long = _recordsRead + def incRecordsRead(records: Long) = _recordsRead += records /** * Invoke the bytesReadCallback and mutate bytesRead. */ def updateBytesRead() { bytesReadCallback.foreach { c => - _bytesRead.set(c()) + _bytesRead = c() } } @@ -287,6 +294,13 @@ case class OutputMetrics(writeMethod: DataWriteMethod.Value) { private var _bytesWritten: Long = _ def bytesWritten = _bytesWritten private[spark] def setBytesWritten(value : Long) = _bytesWritten = value + + /** + * Total records written + */ + private var _recordsWritten: Long = 0L + def recordsWritten = _recordsWritten + private[spark] def setRecordsWritten(value: Long) = _recordsWritten = value } /** @@ -301,7 +315,7 @@ class ShuffleReadMetrics extends Serializable { private var _remoteBlocksFetched: Int = _ def remoteBlocksFetched = _remoteBlocksFetched private[spark] def incRemoteBlocksFetched(value: Int) = _remoteBlocksFetched += value - private[spark] def defRemoteBlocksFetched(value: Int) = _remoteBlocksFetched -= value + private[spark] def decRemoteBlocksFetched(value: Int) = _remoteBlocksFetched -= value /** * Number of local blocks fetched in this shuffle by this task @@ -309,8 +323,7 @@ class ShuffleReadMetrics extends Serializable { private var _localBlocksFetched: Int = _ def localBlocksFetched = _localBlocksFetched private[spark] def incLocalBlocksFetched(value: Int) = _localBlocksFetched += value - private[spark] def defLocalBlocksFetched(value: Int) = _localBlocksFetched -= value - + private[spark] def decLocalBlocksFetched(value: Int) = _localBlocksFetched -= value /** * Time the task spent waiting for remote shuffle blocks. This only includes the time @@ -334,6 +347,14 @@ class ShuffleReadMetrics extends Serializable { * Number of blocks fetched in this shuffle by this task (remote or local) */ def totalBlocksFetched = _remoteBlocksFetched + _localBlocksFetched + + /** + * Total number of records read from the shuffle by this task + */ + private var _recordsRead: Long = _ + def recordsRead = _recordsRead + private[spark] def incRecordsRead(value: Long) = _recordsRead += value + private[spark] def decRecordsRead(value: Long) = _recordsRead -= value } /** @@ -358,5 +379,12 @@ class ShuffleWriteMetrics extends Serializable { private[spark] def incShuffleWriteTime(value: Long) = _shuffleWriteTime += value private[spark] def decShuffleWriteTime(value: Long) = _shuffleWriteTime -= value - + /** + * Total number of records written to the shuffle by this task + */ + @volatile private var _shuffleRecordsWritten: Long = _ + def shuffleRecordsWritten = _shuffleRecordsWritten + private[spark] def incShuffleRecordsWritten(value: Long) = _shuffleRecordsWritten += value + private[spark] def decShuffleRecordsWritten(value: Long) = _shuffleRecordsWritten -= value + private[spark] def setShuffleRecordsWritten(value: Long) = _shuffleRecordsWritten = value } diff --git a/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala b/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala index 89adddcf0ac36..486e86ce1bb19 100644 --- a/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala +++ b/core/src/main/scala/org/apache/spark/rdd/HadoopRDD.scala @@ -247,7 +247,9 @@ class HadoopRDD[K, V]( case eof: EOFException => finished = true } - + if (!finished) { + inputMetrics.incRecordsRead(1) + } (key, value) } @@ -261,7 +263,7 @@ class HadoopRDD[K, V]( // If we can't get the bytes read from the FS stats, fall back to the split size, // which may be inaccurate. try { - inputMetrics.addBytesRead(split.inputSplit.value.getLength) + inputMetrics.incBytesRead(split.inputSplit.value.getLength) } catch { case e: java.io.IOException => logWarning("Unable to get input size to set InputMetrics for task", e) diff --git a/core/src/main/scala/org/apache/spark/rdd/JdbcRDD.scala b/core/src/main/scala/org/apache/spark/rdd/JdbcRDD.scala index 642a12c1edf6c..4fe7622bda00f 100644 --- a/core/src/main/scala/org/apache/spark/rdd/JdbcRDD.scala +++ b/core/src/main/scala/org/apache/spark/rdd/JdbcRDD.scala @@ -99,21 +99,21 @@ class JdbcRDD[T: ClassTag]( override def close() { try { - if (null != rs && ! rs.isClosed()) { + if (null != rs) { rs.close() } } catch { case e: Exception => logWarning("Exception closing resultset", e) } try { - if (null != stmt && ! stmt.isClosed()) { + if (null != stmt) { stmt.close() } } catch { case e: Exception => logWarning("Exception closing statement", e) } try { - if (null != conn && ! conn.isClosed()) { + if (null != conn) { conn.close() } logInfo("closed connection") diff --git a/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala b/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala index 44b9ffd2a53fd..7fb94840df99c 100644 --- a/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala +++ b/core/src/main/scala/org/apache/spark/rdd/NewHadoopRDD.scala @@ -151,7 +151,9 @@ class NewHadoopRDD[K, V]( throw new java.util.NoSuchElementException("End of stream") } havePair = false - + if (!finished) { + inputMetrics.incRecordsRead(1) + } (reader.getCurrentKey, reader.getCurrentValue) } @@ -165,7 +167,7 @@ class NewHadoopRDD[K, V]( // If we can't get the bytes read from the FS stats, fall back to the split size, // which may be inaccurate. try { - inputMetrics.addBytesRead(split.serializableHadoopSplit.value.getLength) + inputMetrics.incBytesRead(split.serializableHadoopSplit.value.getLength) } catch { case e: java.io.IOException => logWarning("Unable to get input size to set InputMetrics for task", e) diff --git a/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala index 49b88a90ab5af..955b42c3baaa1 100644 --- a/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala +++ b/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala @@ -34,7 +34,7 @@ import org.apache.hadoop.io.SequenceFile.CompressionType import org.apache.hadoop.io.compress.CompressionCodec import org.apache.hadoop.mapred.{FileOutputCommitter, FileOutputFormat, JobConf, OutputFormat} import org.apache.hadoop.mapreduce.{Job => NewAPIHadoopJob, OutputFormat => NewOutputFormat, -RecordWriter => NewRecordWriter} + RecordWriter => NewRecordWriter} import org.apache.spark._ import org.apache.spark.Partitioner.defaultPartitioner @@ -993,8 +993,8 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) val (outputMetrics, bytesWrittenCallback) = initHadoopOutputMetrics(context) val writer = format.getRecordWriter(hadoopContext).asInstanceOf[NewRecordWriter[K,V]] + var recordsWritten = 0L try { - var recordsWritten = 0L while (iter.hasNext) { val pair = iter.next() writer.write(pair._1, pair._2) @@ -1008,6 +1008,7 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) } committer.commitTask(hadoopContext) bytesWrittenCallback.foreach { fn => outputMetrics.setBytesWritten(fn()) } + outputMetrics.setRecordsWritten(recordsWritten) 1 } : Int @@ -1065,8 +1066,8 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) writer.setup(context.stageId, context.partitionId, taskAttemptId) writer.open() + var recordsWritten = 0L try { - var recordsWritten = 0L while (iter.hasNext) { val record = iter.next() writer.write(record._1.asInstanceOf[AnyRef], record._2.asInstanceOf[AnyRef]) @@ -1080,6 +1081,7 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) } writer.commit() bytesWrittenCallback.foreach { fn => outputMetrics.setBytesWritten(fn()) } + outputMetrics.setRecordsWritten(recordsWritten) } self.context.runJob(self, writeToFile) @@ -1097,9 +1099,9 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)]) private def maybeUpdateOutputMetrics(bytesWrittenCallback: Option[() => Long], outputMetrics: OutputMetrics, recordsWritten: Long): Unit = { - if (recordsWritten % PairRDDFunctions.RECORDS_BETWEEN_BYTES_WRITTEN_METRIC_UPDATES == 0 - && bytesWrittenCallback.isDefined) { + if (recordsWritten % PairRDDFunctions.RECORDS_BETWEEN_BYTES_WRITTEN_METRIC_UPDATES == 0) { bytesWrittenCallback.foreach { fn => outputMetrics.setBytesWritten(fn()) } + outputMetrics.setRecordsWritten(recordsWritten) } } diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala index 9d2fb4f3b4729..6f77fa32ce37b 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala @@ -311,9 +311,14 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val actorSyste /** * Request an additional number of executors from the cluster manager. - * Return whether the request is acknowledged. + * @return whether the request is acknowledged. */ final override def requestExecutors(numAdditionalExecutors: Int): Boolean = synchronized { + if (numAdditionalExecutors < 0) { + throw new IllegalArgumentException( + "Attempted to request a negative number of additional executor(s) " + + s"$numAdditionalExecutors from the cluster manager. Please specify a positive number!") + } logInfo(s"Requesting $numAdditionalExecutors additional executor(s) from the cluster manager") logDebug(s"Number of pending executors is now $numPendingExecutors") numPendingExecutors += numAdditionalExecutors @@ -322,6 +327,22 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val actorSyste doRequestTotalExecutors(newTotal) } + /** + * Express a preference to the cluster manager for a given total number of executors. This can + * result in canceling pending requests or filing additional requests. + * @return whether the request is acknowledged. + */ + final override def requestTotalExecutors(numExecutors: Int): Boolean = synchronized { + if (numExecutors < 0) { + throw new IllegalArgumentException( + "Attempted to request a negative number of executor(s) " + + s"$numExecutors from the cluster manager. Please specify a positive number!") + } + numPendingExecutors = + math.max(numExecutors - numExistingExecutors + executorsPendingToRemove.size, 0) + doRequestTotalExecutors(numExecutors) + } + /** * Request executors from the cluster manager by specifying the total number desired, * including existing pending and running executors. @@ -332,7 +353,7 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val actorSyste * insufficient resources to satisfy the first request. We make the assumption here that the * cluster manager will eventually fulfill all requests when resources free up. * - * Return whether the request is acknowledged. + * @return whether the request is acknowledged. */ protected def doRequestTotalExecutors(requestedTotal: Int): Boolean = false diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/SparkDeploySchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/SparkDeploySchedulerBackend.scala index d2e1680a5fd1b..40fc6b59cdf7b 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/SparkDeploySchedulerBackend.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/SparkDeploySchedulerBackend.scala @@ -52,8 +52,13 @@ private[spark] class SparkDeploySchedulerBackend( conf.get("spark.driver.host"), conf.get("spark.driver.port"), CoarseGrainedSchedulerBackend.ACTOR_NAME) - val args = Seq(driverUrl, "{{EXECUTOR_ID}}", "{{HOSTNAME}}", "{{CORES}}", "{{APP_ID}}", - "{{WORKER_URL}}") + val args = Seq( + "--driver-url", driverUrl, + "--executor-id", "{{EXECUTOR_ID}}", + "--hostname", "{{HOSTNAME}}", + "--cores", "{{CORES}}", + "--app-id", "{{APP_ID}}", + "--worker-url", "{{WORKER_URL}}") val extraJavaOpts = sc.conf.getOption("spark.executor.extraJavaOptions") .map(Utils.splitCommandString).getOrElse(Seq.empty) val classPathEntries = sc.conf.getOption("spark.executor.extraClassPath") diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala index 0d1c2a916ca7f..90dfe14352a8e 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala @@ -154,18 +154,25 @@ private[spark] class CoarseMesosSchedulerBackend( if (uri == null) { val runScript = new File(executorSparkHome, "./bin/spark-class").getCanonicalPath command.setValue( - "%s \"%s\" org.apache.spark.executor.CoarseGrainedExecutorBackend %s %s %s %d %s".format( - prefixEnv, runScript, driverUrl, offer.getSlaveId.getValue, - offer.getHostname, numCores, appId)) + "%s \"%s\" org.apache.spark.executor.CoarseGrainedExecutorBackend" + .format(prefixEnv, runScript) + + s" --driver-url $driverUrl" + + s" --executor-id ${offer.getSlaveId.getValue}" + + s" --hostname ${offer.getHostname}" + + s" --cores $numCores" + + s" --app-id $appId") } else { // Grab everything to the first '.'. We'll use that and '*' to // glob the directory "correctly". val basename = uri.split('/').last.split('.').head command.setValue( - ("cd %s*; %s " + - "./bin/spark-class org.apache.spark.executor.CoarseGrainedExecutorBackend %s %s %s %d %s") - .format(basename, prefixEnv, driverUrl, offer.getSlaveId.getValue, - offer.getHostname, numCores, appId)) + s"cd $basename*; $prefixEnv " + + "./bin/spark-class org.apache.spark.executor.CoarseGrainedExecutorBackend" + + s" --driver-url $driverUrl" + + s" --executor-id ${offer.getSlaveId.getValue}" + + s" --hostname ${offer.getHostname}" + + s" --cores $numCores" + + s" --app-id $appId") command.addUris(CommandInfo.URI.newBuilder().setValue(uri)) } command.build() diff --git a/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala b/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala index e3e7434df45b0..7a2c5ae32d98b 100644 --- a/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala +++ b/core/src/main/scala/org/apache/spark/shuffle/hash/BlockStoreShuffleFetcher.scala @@ -86,6 +86,12 @@ private[hash] object BlockStoreShuffleFetcher extends Logging { context.taskMetrics.updateShuffleReadMetrics() }) - new InterruptibleIterator[T](context, completionIter) + new InterruptibleIterator[T](context, completionIter) { + val readMetrics = context.taskMetrics.createShuffleReadMetricsForDependency() + override def next(): T = { + readMetrics.incRecordsRead(1) + delegate.next() + } + } } } diff --git a/core/src/main/scala/org/apache/spark/storage/BlockManager.scala b/core/src/main/scala/org/apache/spark/storage/BlockManager.scala index 8bc5a1cd18b64..86dbd89f0ffb8 100644 --- a/core/src/main/scala/org/apache/spark/storage/BlockManager.scala +++ b/core/src/main/scala/org/apache/spark/storage/BlockManager.scala @@ -53,7 +53,7 @@ private[spark] class BlockResult( readMethod: DataReadMethod.Value, bytes: Long) { val inputMetrics = new InputMetrics(readMethod) - inputMetrics.addBytesRead(bytes) + inputMetrics.incBytesRead(bytes) } /** diff --git a/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala b/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala index 3198d766fca37..81164178b9e8e 100644 --- a/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala +++ b/core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala @@ -29,7 +29,8 @@ import org.apache.spark.executor.ShuffleWriteMetrics * appending data to an existing block, and can guarantee atomicity in the case of faults * as it allows the caller to revert partial writes. * - * This interface does not support concurrent writes. + * This interface does not support concurrent writes. Also, once the writer has + * been opened, it cannot be reopened again. */ private[spark] abstract class BlockObjectWriter(val blockId: BlockId) { @@ -95,6 +96,7 @@ private[spark] class DiskBlockObjectWriter( private var ts: TimeTrackingOutputStream = null private var objOut: SerializationStream = null private var initialized = false + private var hasBeenClosed = false /** * Cursors used to represent positions in the file. @@ -115,11 +117,16 @@ private[spark] class DiskBlockObjectWriter( private var finalPosition: Long = -1 private var reportedPosition = initialPosition - /** Calling channel.position() to update the write metrics can be a little bit expensive, so we - * only call it every N writes */ - private var writesSinceMetricsUpdate = 0 + /** + * Keep track of number of records written and also use this to periodically + * output bytes written since the latter is expensive to do for each record. + */ + private var numRecordsWritten = 0 override def open(): BlockObjectWriter = { + if (hasBeenClosed) { + throw new IllegalStateException("Writer already closed. Cannot be reopened.") + } fos = new FileOutputStream(file, true) ts = new TimeTrackingOutputStream(fos) channel = fos.getChannel() @@ -145,6 +152,7 @@ private[spark] class DiskBlockObjectWriter( ts = null objOut = null initialized = false + hasBeenClosed = true } } @@ -168,6 +176,7 @@ private[spark] class DiskBlockObjectWriter( override def revertPartialWritesAndClose() { try { writeMetrics.decShuffleBytesWritten(reportedPosition - initialPosition) + writeMetrics.decShuffleRecordsWritten(numRecordsWritten) if (initialized) { objOut.flush() @@ -193,12 +202,11 @@ private[spark] class DiskBlockObjectWriter( } objOut.writeObject(value) + numRecordsWritten += 1 + writeMetrics.incShuffleRecordsWritten(1) - if (writesSinceMetricsUpdate == 32) { - writesSinceMetricsUpdate = 0 + if (numRecordsWritten % 32 == 0) { updateBytesWritten() - } else { - writesSinceMetricsUpdate += 1 } } diff --git a/core/src/main/scala/org/apache/spark/ui/JettyUtils.scala b/core/src/main/scala/org/apache/spark/ui/JettyUtils.scala index 88fed833f922d..bf4b24e98b134 100644 --- a/core/src/main/scala/org/apache/spark/ui/JettyUtils.scala +++ b/core/src/main/scala/org/apache/spark/ui/JettyUtils.scala @@ -62,17 +62,22 @@ private[spark] object JettyUtils extends Logging { securityMgr: SecurityManager): HttpServlet = { new HttpServlet { override def doGet(request: HttpServletRequest, response: HttpServletResponse) { - if (securityMgr.checkUIViewPermissions(request.getRemoteUser)) { - response.setContentType("%s;charset=utf-8".format(servletParams.contentType)) - response.setStatus(HttpServletResponse.SC_OK) - val result = servletParams.responder(request) - response.setHeader("Cache-Control", "no-cache, no-store, must-revalidate") - response.getWriter.println(servletParams.extractFn(result)) - } else { - response.setStatus(HttpServletResponse.SC_UNAUTHORIZED) - response.setHeader("Cache-Control", "no-cache, no-store, must-revalidate") - response.sendError(HttpServletResponse.SC_UNAUTHORIZED, - "User is not authorized to access this page.") + try { + if (securityMgr.checkUIViewPermissions(request.getRemoteUser)) { + response.setContentType("%s;charset=utf-8".format(servletParams.contentType)) + response.setStatus(HttpServletResponse.SC_OK) + val result = servletParams.responder(request) + response.setHeader("Cache-Control", "no-cache, no-store, must-revalidate") + response.getWriter.println(servletParams.extractFn(result)) + } else { + response.setStatus(HttpServletResponse.SC_UNAUTHORIZED) + response.setHeader("Cache-Control", "no-cache, no-store, must-revalidate") + response.sendError(HttpServletResponse.SC_UNAUTHORIZED, + "User is not authorized to access this page.") + } + } catch { + case e: IllegalArgumentException => + response.sendError(HttpServletResponse.SC_BAD_REQUEST, e.getMessage) } } } diff --git a/core/src/main/scala/org/apache/spark/ui/ToolTips.scala b/core/src/main/scala/org/apache/spark/ui/ToolTips.scala index 4307029d44fbb..3a15e603b1969 100644 --- a/core/src/main/scala/org/apache/spark/ui/ToolTips.scala +++ b/core/src/main/scala/org/apache/spark/ui/ToolTips.scala @@ -29,14 +29,15 @@ private[spark] object ToolTips { val SHUFFLE_READ_BLOCKED_TIME = "Time that the task spent blocked waiting for shuffle data to be read from remote machines." - val INPUT = "Bytes read from Hadoop or from Spark storage." + val INPUT = "Bytes and records read from Hadoop or from Spark storage." - val OUTPUT = "Bytes written to Hadoop." + val OUTPUT = "Bytes and records written to Hadoop." - val SHUFFLE_WRITE = "Bytes written to disk in order to be read by a shuffle in a future stage." + val SHUFFLE_WRITE = + "Bytes and records written to disk in order to be read by a shuffle in a future stage." val SHUFFLE_READ = - """Bytes read from remote executors. Typically less than shuffle write bytes + """Bytes and records read from remote executors. Typically less than shuffle write bytes because this does not include shuffle data read locally.""" val GETTING_RESULT_TIME = diff --git a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala index c82730f524eb7..f0ae95bb8c812 100644 --- a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala +++ b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala @@ -43,7 +43,7 @@ private[ui] class ExecutorThreadDumpPage(parent: ExecutorsTab) extends WebUIPage } id }.getOrElse { - return Text(s"Missing executorId parameter") + throw new IllegalArgumentException(s"Missing executorId parameter") } val time = System.currentTimeMillis() val maybeThreadDump = sc.get.getExecutorThreadDump(executorId) diff --git a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala index a38cb75fdd8c6..3afd7ef07d7c9 100644 --- a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala +++ b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorsTab.scala @@ -48,7 +48,9 @@ class ExecutorsListener(storageStatusListener: StorageStatusListener) extends Sp val executorToTasksFailed = HashMap[String, Int]() val executorToDuration = HashMap[String, Long]() val executorToInputBytes = HashMap[String, Long]() + val executorToInputRecords = HashMap[String, Long]() val executorToOutputBytes = HashMap[String, Long]() + val executorToOutputRecords = HashMap[String, Long]() val executorToShuffleRead = HashMap[String, Long]() val executorToShuffleWrite = HashMap[String, Long]() val executorToLogUrls = HashMap[String, Map[String, String]]() @@ -84,10 +86,14 @@ class ExecutorsListener(storageStatusListener: StorageStatusListener) extends Sp metrics.inputMetrics.foreach { inputMetrics => executorToInputBytes(eid) = executorToInputBytes.getOrElse(eid, 0L) + inputMetrics.bytesRead + executorToInputRecords(eid) = + executorToInputRecords.getOrElse(eid, 0L) + inputMetrics.recordsRead } metrics.outputMetrics.foreach { outputMetrics => executorToOutputBytes(eid) = executorToOutputBytes.getOrElse(eid, 0L) + outputMetrics.bytesWritten + executorToOutputRecords(eid) = + executorToOutputRecords.getOrElse(eid, 0L) + outputMetrics.recordsWritten } metrics.shuffleReadMetrics.foreach { shuffleRead => executorToShuffleRead(eid) = diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/AllJobsPage.scala b/core/src/main/scala/org/apache/spark/ui/jobs/AllJobsPage.scala index 045c69da06feb..bd923d78a86ce 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/AllJobsPage.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/AllJobsPage.scala @@ -42,7 +42,9 @@ private[ui] class AllJobsPage(parent: JobsTab) extends WebUIPage("") { } def makeRow(job: JobUIData): Seq[Node] = { - val lastStageInfo = listener.stageIdToInfo.get(job.stageIds.max) + val lastStageInfo = Option(job.stageIds) + .filter(_.nonEmpty) + .flatMap { ids => listener.stageIdToInfo.get(ids.max) } val lastStageData = lastStageInfo.flatMap { s => listener.stageIdToData.get((s.stageId, s.attemptId)) } diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala b/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala index 9836d11a6d85f..1f8536d1b7195 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala @@ -36,6 +36,20 @@ private[ui] class ExecutorTable(stageId: Int, stageAttemptId: Int, parent: Stage /** Special table which merges two header cells. */ private def executorTable[T](): Seq[Node] = { + val stageData = listener.stageIdToData.get((stageId, stageAttemptId)) + var hasInput = false + var hasOutput = false + var hasShuffleWrite = false + var hasShuffleRead = false + var hasBytesSpilled = false + stageData.foreach(data => { + hasInput = data.hasInput + hasOutput = data.hasOutput + hasShuffleRead = data.hasShuffleRead + hasShuffleWrite = data.hasShuffleWrite + hasBytesSpilled = data.hasBytesSpilled + }) + @@ -44,12 +58,32 @@ private[ui] class ExecutorTable(stageId: Int, stageAttemptId: Int, parent: Stage - - - - - - + {if (hasInput) { + + }} + {if (hasOutput) { + + }} + {if (hasShuffleRead) { + + }} + {if (hasShuffleWrite) { + + }} + {if (hasBytesSpilled) { + + + }} {createExecutorTable()} @@ -76,18 +110,34 @@ private[ui] class ExecutorTable(stageId: Int, stageAttemptId: Int, parent: Stage - - - - - - + {if (stageData.hasInput) { + + }} + {if (stageData.hasOutput) { + + }} + {if (stageData.hasShuffleRead) { + + }} + {if (stageData.hasShuffleWrite) { + + }} + {if (stageData.hasBytesSpilled) { + + + }} } case None => diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/JobPage.scala b/core/src/main/scala/org/apache/spark/ui/jobs/JobPage.scala index 77d36209c6048..7541d3e9c72e7 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/JobPage.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/JobPage.scala @@ -32,7 +32,10 @@ private[ui] class JobPage(parent: JobsTab) extends WebUIPage("job") { def render(request: HttpServletRequest): Seq[Node] = { listener.synchronized { - val jobId = request.getParameter("id").toInt + val parameterId = request.getParameter("id") + require(parameterId != null && parameterId.nonEmpty, "Missing id parameter") + + val jobId = parameterId.toInt val jobDataOption = listener.jobIdToData.get(jobId) if (jobDataOption.isEmpty) { val content = diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala b/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala index 4d200eeda86b9..f463f8d7c7215 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala @@ -394,24 +394,48 @@ class JobProgressListener(conf: SparkConf) extends SparkListener with Logging { stageData.shuffleWriteBytes += shuffleWriteDelta execSummary.shuffleWrite += shuffleWriteDelta + val shuffleWriteRecordsDelta = + (taskMetrics.shuffleWriteMetrics.map(_.shuffleRecordsWritten).getOrElse(0L) + - oldMetrics.flatMap(_.shuffleWriteMetrics).map(_.shuffleRecordsWritten).getOrElse(0L)) + stageData.shuffleWriteRecords += shuffleWriteRecordsDelta + execSummary.shuffleWriteRecords += shuffleWriteRecordsDelta + val shuffleReadDelta = (taskMetrics.shuffleReadMetrics.map(_.remoteBytesRead).getOrElse(0L) - oldMetrics.flatMap(_.shuffleReadMetrics).map(_.remoteBytesRead).getOrElse(0L)) stageData.shuffleReadBytes += shuffleReadDelta execSummary.shuffleRead += shuffleReadDelta + val shuffleReadRecordsDelta = + (taskMetrics.shuffleReadMetrics.map(_.recordsRead).getOrElse(0L) + - oldMetrics.flatMap(_.shuffleReadMetrics).map(_.recordsRead).getOrElse(0L)) + stageData.shuffleReadRecords += shuffleReadRecordsDelta + execSummary.shuffleReadRecords += shuffleReadRecordsDelta + val inputBytesDelta = (taskMetrics.inputMetrics.map(_.bytesRead).getOrElse(0L) - oldMetrics.flatMap(_.inputMetrics).map(_.bytesRead).getOrElse(0L)) stageData.inputBytes += inputBytesDelta execSummary.inputBytes += inputBytesDelta + val inputRecordsDelta = + (taskMetrics.inputMetrics.map(_.recordsRead).getOrElse(0L) + - oldMetrics.flatMap(_.inputMetrics).map(_.recordsRead).getOrElse(0L)) + stageData.inputRecords += inputRecordsDelta + execSummary.inputRecords += inputRecordsDelta + val outputBytesDelta = (taskMetrics.outputMetrics.map(_.bytesWritten).getOrElse(0L) - oldMetrics.flatMap(_.outputMetrics).map(_.bytesWritten).getOrElse(0L)) stageData.outputBytes += outputBytesDelta execSummary.outputBytes += outputBytesDelta + val outputRecordsDelta = + (taskMetrics.outputMetrics.map(_.recordsWritten).getOrElse(0L) + - oldMetrics.flatMap(_.outputMetrics).map(_.recordsWritten).getOrElse(0L)) + stageData.outputRecords += outputRecordsDelta + execSummary.outputRecords += outputRecordsDelta + val diskSpillDelta = taskMetrics.diskBytesSpilled - oldMetrics.map(_.diskBytesSpilled).getOrElse(0L) stageData.diskBytesSpilled += diskSpillDelta diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/PoolPage.scala b/core/src/main/scala/org/apache/spark/ui/jobs/PoolPage.scala index 5fc6cc7533150..f47cdc935e539 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/PoolPage.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/PoolPage.scala @@ -32,6 +32,8 @@ private[ui] class PoolPage(parent: StagesTab) extends WebUIPage("pool") { def render(request: HttpServletRequest): Seq[Node] = { listener.synchronized { val poolName = request.getParameter("poolname") + require(poolName != null && poolName.nonEmpty, "Missing poolname parameter") + val poolToActiveStages = listener.poolToActiveStages val activeStages = poolToActiveStages.get(poolName) match { case Some(s) => s.values.toSeq diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala b/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala index d8be1b20b3acd..05ffd5bc58fbb 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala @@ -36,8 +36,14 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { def render(request: HttpServletRequest): Seq[Node] = { listener.synchronized { - val stageId = request.getParameter("id").toInt - val stageAttemptId = request.getParameter("attempt").toInt + val parameterId = request.getParameter("id") + require(parameterId != null && parameterId.nonEmpty, "Missing id parameter") + + val parameterAttempt = request.getParameter("attempt") + require(parameterAttempt != null && parameterAttempt.nonEmpty, "Missing attempt parameter") + + val stageId = parameterId.toInt + val stageAttemptId = parameterAttempt.toInt val stageDataOption = listener.stageIdToData.get((stageId, stageAttemptId)) if (stageDataOption.isEmpty || stageDataOption.get.taskData.isEmpty) { @@ -56,11 +62,6 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { val numCompleted = tasks.count(_.taskInfo.finished) val accumulables = listener.stageIdToData((stageId, stageAttemptId)).accumulables val hasAccumulators = accumulables.size > 0 - val hasInput = stageData.inputBytes > 0 - val hasOutput = stageData.outputBytes > 0 - val hasShuffleRead = stageData.shuffleReadBytes > 0 - val hasShuffleWrite = stageData.shuffleWriteBytes > 0 - val hasBytesSpilled = stageData.memoryBytesSpilled > 0 && stageData.diskBytesSpilled > 0 val summary =
    @@ -69,31 +70,33 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { Total task time across all tasks: {UIUtils.formatDuration(stageData.executorRunTime)} - {if (hasInput) { + {if (stageData.hasInput) {
  • - Input: - {Utils.bytesToString(stageData.inputBytes)} + Input Size / Records: + {s"${Utils.bytesToString(stageData.inputBytes)} / ${stageData.inputRecords}"}
    • }} - {if (hasOutput) { + {if (stageData.hasOutput) {
  • Output: - {Utils.bytesToString(stageData.outputBytes)} + {s"${Utils.bytesToString(stageData.outputBytes)} / ${stageData.outputRecords}"}
    • }} - {if (hasShuffleRead) { + {if (stageData.hasShuffleRead) {
  • Shuffle read: - {Utils.bytesToString(stageData.shuffleReadBytes)} + {s"${Utils.bytesToString(stageData.shuffleReadBytes)} / " + + s"${stageData.shuffleReadRecords}"}
    • }} - {if (hasShuffleWrite) { + {if (stageData.hasShuffleWrite) {
  • Shuffle write: - {Utils.bytesToString(stageData.shuffleWriteBytes)} + {s"${Utils.bytesToString(stageData.shuffleWriteBytes)} / " + + s"${stageData.shuffleWriteRecords}"}
    • }} - {if (hasBytesSpilled) { + {if (stageData.hasBytesSpilled) {
  • Shuffle spill (memory): {Utils.bytesToString(stageData.memoryBytesSpilled)} @@ -132,7 +135,7 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { Task Deserialization Time
    • - {if (hasShuffleRead) { + {if (stageData.hasShuffleRead) {
  • @@ -174,25 +177,32 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { ("Result Serialization Time", TaskDetailsClassNames.RESULT_SERIALIZATION_TIME), ("Getting Result Time", TaskDetailsClassNames.GETTING_RESULT_TIME)) ++ {if (hasAccumulators) Seq(("Accumulators", "")) else Nil} ++ - {if (hasInput) Seq(("Input", "")) else Nil} ++ - {if (hasOutput) Seq(("Output", "")) else Nil} ++ - {if (hasShuffleRead) { + {if (stageData.hasInput) Seq(("Input Size / Records", "")) else Nil} ++ + {if (stageData.hasOutput) Seq(("Output Size / Records", "")) else Nil} ++ + {if (stageData.hasShuffleRead) { Seq(("Shuffle Read Blocked Time", TaskDetailsClassNames.SHUFFLE_READ_BLOCKED_TIME), - ("Shuffle Read", "")) + ("Shuffle Read Size / Records", "")) + } else { + Nil + }} ++ + {if (stageData.hasShuffleWrite) { + Seq(("Write Time", ""), ("Shuffle Write Size / Records", "")) + } else { + Nil + }} ++ + {if (stageData.hasBytesSpilled) { + Seq(("Shuffle Spill (Memory)", ""), ("Shuffle Spill (Disk)", "")) } else { Nil }} ++ - {if (hasShuffleWrite) Seq(("Write Time", ""), ("Shuffle Write", "")) else Nil} ++ - {if (hasBytesSpilled) Seq(("Shuffle Spill (Memory)", ""), ("Shuffle Spill (Disk)", "")) - else Nil} ++ Seq(("Errors", "")) val unzipped = taskHeadersAndCssClasses.unzip val taskTable = UIUtils.listingTable( unzipped._1, - taskRow(hasAccumulators, hasInput, hasOutput, hasShuffleRead, hasShuffleWrite, - hasBytesSpilled), + taskRow(hasAccumulators, stageData.hasInput, stageData.hasOutput, + stageData.hasShuffleRead, stageData.hasShuffleWrite, stageData.hasBytesSpilled), tasks, headerClasses = unzipped._2) // Excludes tasks which failed and have incomplete metrics @@ -203,8 +213,11 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { None } else { + def getDistributionQuantiles(data: Seq[Double]): IndexedSeq[Double] = + Distribution(data).get.getQuantiles() + def getFormattedTimeQuantiles(times: Seq[Double]): Seq[Node] = { - Distribution(times).get.getQuantiles().map { millis => + getDistributionQuantiles(times).map { millis =>
    • } } @@ -273,17 +286,36 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { getFormattedTimeQuantiles(schedulerDelays) def getFormattedSizeQuantiles(data: Seq[Double]) = - Distribution(data).get.getQuantiles().map(d => ) + getDistributionQuantiles(data).map(d => ) + + def getFormattedSizeQuantilesWithRecords(data: Seq[Double], records: Seq[Double]) = { + val recordDist = getDistributionQuantiles(records).iterator + getDistributionQuantiles(data).map(d => + + ) + } val inputSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.inputMetrics.map(_.bytesRead).getOrElse(0L).toDouble } - val inputQuantiles = +: getFormattedSizeQuantiles(inputSizes) + + val inputRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.inputMetrics.map(_.recordsRead).getOrElse(0L).toDouble + } + + val inputQuantiles = +: + getFormattedSizeQuantilesWithRecords(inputSizes, inputRecords) val outputSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.outputMetrics.map(_.bytesWritten).getOrElse(0L).toDouble } - val outputQuantiles = +: getFormattedSizeQuantiles(outputSizes) + + val outputRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.outputMetrics.map(_.recordsWritten).getOrElse(0L).toDouble + } + + val outputQuantiles = +: + getFormattedSizeQuantilesWithRecords(outputSizes, outputRecords) val shuffleReadBlockedTimes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.shuffleReadMetrics.map(_.fetchWaitTime).getOrElse(0L).toDouble @@ -294,14 +326,24 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { val shuffleReadSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.shuffleReadMetrics.map(_.remoteBytesRead).getOrElse(0L).toDouble } - val shuffleReadQuantiles = +: - getFormattedSizeQuantiles(shuffleReadSizes) + + val shuffleReadRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.shuffleReadMetrics.map(_.recordsRead).getOrElse(0L).toDouble + } + + val shuffleReadQuantiles = +: + getFormattedSizeQuantilesWithRecords(shuffleReadSizes, shuffleReadRecords) val shuffleWriteSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.shuffleWriteMetrics.map(_.shuffleBytesWritten).getOrElse(0L).toDouble } - val shuffleWriteQuantiles = +: - getFormattedSizeQuantiles(shuffleWriteSizes) + + val shuffleWriteRecords = validTasks.map { case TaskUIData(_, metrics, _) => + metrics.get.shuffleWriteMetrics.map(_.shuffleRecordsWritten).getOrElse(0L).toDouble + } + + val shuffleWriteQuantiles = +: + getFormattedSizeQuantilesWithRecords(shuffleWriteSizes, shuffleWriteRecords) val memoryBytesSpilledSizes = validTasks.map { case TaskUIData(_, metrics, _) => metrics.get.memoryBytesSpilled.toDouble @@ -326,9 +368,9 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { {serializationQuantiles} , {gettingResultQuantiles}, - if (hasInput) {inputQuantiles} else Nil, - if (hasOutput) {outputQuantiles} else Nil, - if (hasShuffleRead) { + if (stageData.hasInput) {inputQuantiles} else Nil, + if (stageData.hasOutput) {outputQuantiles} else Nil, + if (stageData.hasShuffleRead) { {shuffleReadBlockedQuantiles} @@ -336,9 +378,9 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { } else { Nil }, - if (hasShuffleWrite) {shuffleWriteQuantiles} else Nil, - if (hasBytesSpilled) {memoryBytesSpilledQuantiles} else Nil, - if (hasBytesSpilled) {diskBytesSpilledQuantiles} else Nil) + if (stageData.hasShuffleWrite) {shuffleWriteQuantiles} else Nil, + if (stageData.hasBytesSpilled) {memoryBytesSpilledQuantiles} else Nil, + if (stageData.hasBytesSpilled) {diskBytesSpilledQuantiles} else Nil) val quantileHeaders = Seq("Metric", "Min", "25th percentile", "Median", "75th percentile", "Max") @@ -397,26 +439,32 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { val inputReadable = maybeInput .map(m => s"${Utils.bytesToString(m.bytesRead)} (${m.readMethod.toString.toLowerCase()})") .getOrElse("") + val inputRecords = maybeInput.map(_.recordsRead.toString).getOrElse("") val maybeOutput = metrics.flatMap(_.outputMetrics) val outputSortable = maybeOutput.map(_.bytesWritten.toString).getOrElse("") val outputReadable = maybeOutput .map(m => s"${Utils.bytesToString(m.bytesWritten)}") .getOrElse("") + val outputRecords = maybeOutput.map(_.recordsWritten.toString).getOrElse("") - val maybeShuffleReadBlockedTime = metrics.flatMap(_.shuffleReadMetrics).map(_.fetchWaitTime) - val shuffleReadBlockedTimeSortable = maybeShuffleReadBlockedTime.map(_.toString).getOrElse("") + val maybeShuffleRead = metrics.flatMap(_.shuffleReadMetrics) + val shuffleReadBlockedTimeSortable = maybeShuffleRead + .map(_.fetchWaitTime.toString).getOrElse("") val shuffleReadBlockedTimeReadable = - maybeShuffleReadBlockedTime.map(ms => UIUtils.formatDuration(ms)).getOrElse("") + maybeShuffleRead.map(ms => UIUtils.formatDuration(ms.fetchWaitTime)).getOrElse("") - val maybeShuffleRead = metrics.flatMap(_.shuffleReadMetrics).map(_.remoteBytesRead) - val shuffleReadSortable = maybeShuffleRead.map(_.toString).getOrElse("") - val shuffleReadReadable = maybeShuffleRead.map(Utils.bytesToString).getOrElse("") + val shuffleReadSortable = maybeShuffleRead.map(_.remoteBytesRead.toString).getOrElse("") + val shuffleReadReadable = maybeShuffleRead + .map(m => s"${Utils.bytesToString(m.remoteBytesRead)}").getOrElse("") + val shuffleReadRecords = maybeShuffleRead.map(_.recordsRead.toString).getOrElse("") - val maybeShuffleWrite = - metrics.flatMap(_.shuffleWriteMetrics).map(_.shuffleBytesWritten) - val shuffleWriteSortable = maybeShuffleWrite.map(_.toString).getOrElse("") - val shuffleWriteReadable = maybeShuffleWrite.map(Utils.bytesToString).getOrElse("") + val maybeShuffleWrite = metrics.flatMap(_.shuffleWriteMetrics) + val shuffleWriteSortable = maybeShuffleWrite.map(_.shuffleBytesWritten.toString).getOrElse("") + val shuffleWriteReadable = maybeShuffleWrite + .map(m => s"${Utils.bytesToString(m.shuffleBytesWritten)}").getOrElse("") + val shuffleWriteRecords = maybeShuffleWrite + .map(_.shuffleRecordsWritten.toString).getOrElse("") val maybeWriteTime = metrics.flatMap(_.shuffleWriteMetrics).map(_.shuffleWriteTime) val writeTimeSortable = maybeWriteTime.map(_.toString).getOrElse("") @@ -472,12 +520,12 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { }} {if (hasInput) { }} {if (hasOutput) { }} {if (hasShuffleRead) { @@ -486,7 +534,7 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { {shuffleReadBlockedTimeReadable} }} {if (hasShuffleWrite) { @@ -494,7 +542,7 @@ private[ui] class StagePage(parent: StagesTab) extends WebUIPage("stage") { {writeTimeReadable} }} {if (hasBytesSpilled) { diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala b/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala index 01f7e23212c3d..69aac6c862de5 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala @@ -31,9 +31,13 @@ private[jobs] object UIData { var failedTasks : Int = 0 var succeededTasks : Int = 0 var inputBytes : Long = 0 + var inputRecords : Long = 0 var outputBytes : Long = 0 + var outputRecords : Long = 0 var shuffleRead : Long = 0 + var shuffleReadRecords : Long = 0 var shuffleWrite : Long = 0 + var shuffleWriteRecords : Long = 0 var memoryBytesSpilled : Long = 0 var diskBytesSpilled : Long = 0 } @@ -73,9 +77,13 @@ private[jobs] object UIData { var executorRunTime: Long = _ var inputBytes: Long = _ + var inputRecords: Long = _ var outputBytes: Long = _ + var outputRecords: Long = _ var shuffleReadBytes: Long = _ + var shuffleReadRecords : Long = _ var shuffleWriteBytes: Long = _ + var shuffleWriteRecords: Long = _ var memoryBytesSpilled: Long = _ var diskBytesSpilled: Long = _ @@ -85,6 +93,12 @@ private[jobs] object UIData { var accumulables = new HashMap[Long, AccumulableInfo] var taskData = new HashMap[Long, TaskUIData] var executorSummary = new HashMap[String, ExecutorSummary] + + def hasInput = inputBytes > 0 + def hasOutput = outputBytes > 0 + def hasShuffleRead = shuffleReadBytes > 0 + def hasShuffleWrite = shuffleWriteBytes > 0 + def hasBytesSpilled = memoryBytesSpilled > 0 && diskBytesSpilled > 0 } /** diff --git a/core/src/main/scala/org/apache/spark/ui/storage/RDDPage.scala b/core/src/main/scala/org/apache/spark/ui/storage/RDDPage.scala index 12d23a92878cf..199f731b92bcc 100644 --- a/core/src/main/scala/org/apache/spark/ui/storage/RDDPage.scala +++ b/core/src/main/scala/org/apache/spark/ui/storage/RDDPage.scala @@ -30,7 +30,10 @@ private[ui] class RDDPage(parent: StorageTab) extends WebUIPage("rdd") { private val listener = parent.listener def render(request: HttpServletRequest): Seq[Node] = { - val rddId = request.getParameter("id").toInt + val parameterId = request.getParameter("id") + require(parameterId != null && parameterId.nonEmpty, "Missing id parameter") + + val rddId = parameterId.toInt val storageStatusList = listener.storageStatusList val rddInfo = listener.rddInfoList.find(_.id == rddId).getOrElse { // Rather than crashing, render an "RDD Not Found" page diff --git a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala index c8407bbcb780b..b0b545640f5aa 100644 --- a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala +++ b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala @@ -293,22 +293,26 @@ private[spark] object JsonProtocol { ("Remote Blocks Fetched" -> shuffleReadMetrics.remoteBlocksFetched) ~ ("Local Blocks Fetched" -> shuffleReadMetrics.localBlocksFetched) ~ ("Fetch Wait Time" -> shuffleReadMetrics.fetchWaitTime) ~ - ("Remote Bytes Read" -> shuffleReadMetrics.remoteBytesRead) + ("Remote Bytes Read" -> shuffleReadMetrics.remoteBytesRead) ~ + ("Total Records Read" -> shuffleReadMetrics.recordsRead) } def shuffleWriteMetricsToJson(shuffleWriteMetrics: ShuffleWriteMetrics): JValue = { ("Shuffle Bytes Written" -> shuffleWriteMetrics.shuffleBytesWritten) ~ - ("Shuffle Write Time" -> shuffleWriteMetrics.shuffleWriteTime) + ("Shuffle Write Time" -> shuffleWriteMetrics.shuffleWriteTime) ~ + ("Shuffle Records Written" -> shuffleWriteMetrics.shuffleRecordsWritten) } def inputMetricsToJson(inputMetrics: InputMetrics): JValue = { ("Data Read Method" -> inputMetrics.readMethod.toString) ~ - ("Bytes Read" -> inputMetrics.bytesRead) + ("Bytes Read" -> inputMetrics.bytesRead) ~ + ("Records Read" -> inputMetrics.recordsRead) } def outputMetricsToJson(outputMetrics: OutputMetrics): JValue = { ("Data Write Method" -> outputMetrics.writeMethod.toString) ~ - ("Bytes Written" -> outputMetrics.bytesWritten) + ("Bytes Written" -> outputMetrics.bytesWritten) ~ + ("Records Written" -> outputMetrics.recordsWritten) } def taskEndReasonToJson(taskEndReason: TaskEndReason): JValue = { @@ -670,6 +674,7 @@ private[spark] object JsonProtocol { metrics.incLocalBlocksFetched((json \ "Local Blocks Fetched").extract[Int]) metrics.incFetchWaitTime((json \ "Fetch Wait Time").extract[Long]) metrics.incRemoteBytesRead((json \ "Remote Bytes Read").extract[Long]) + metrics.incRecordsRead((json \ "Total Records Read").extractOpt[Long].getOrElse(0)) metrics } @@ -677,13 +682,16 @@ private[spark] object JsonProtocol { val metrics = new ShuffleWriteMetrics metrics.incShuffleBytesWritten((json \ "Shuffle Bytes Written").extract[Long]) metrics.incShuffleWriteTime((json \ "Shuffle Write Time").extract[Long]) + metrics.setShuffleRecordsWritten((json \ "Shuffle Records Written") + .extractOpt[Long].getOrElse(0)) metrics } def inputMetricsFromJson(json: JValue): InputMetrics = { val metrics = new InputMetrics( DataReadMethod.withName((json \ "Data Read Method").extract[String])) - metrics.addBytesRead((json \ "Bytes Read").extract[Long]) + metrics.incBytesRead((json \ "Bytes Read").extract[Long]) + metrics.incRecordsRead((json \ "Records Read").extractOpt[Long].getOrElse(0)) metrics } @@ -691,6 +699,7 @@ private[spark] object JsonProtocol { val metrics = new OutputMetrics( DataWriteMethod.withName((json \ "Data Write Method").extract[String])) metrics.setBytesWritten((json \ "Bytes Written").extract[Long]) + metrics.setRecordsWritten((json \ "Records Written").extractOpt[Long].getOrElse(0)) metrics } diff --git a/core/src/main/scala/org/apache/spark/util/MutableURLClassLoader.scala b/core/src/main/scala/org/apache/spark/util/MutableURLClassLoader.scala new file mode 100644 index 0000000000000..d9c7103b2f3bf --- /dev/null +++ b/core/src/main/scala/org/apache/spark/util/MutableURLClassLoader.scala @@ -0,0 +1,103 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.util + +import java.net.{URLClassLoader, URL} +import java.util.Enumeration +import java.util.concurrent.ConcurrentHashMap + +import scala.collection.JavaConversions._ + +import org.apache.spark.util.ParentClassLoader + +/** + * URL class loader that exposes the `addURL` and `getURLs` methods in URLClassLoader. + */ +private[spark] class MutableURLClassLoader(urls: Array[URL], parent: ClassLoader) + extends URLClassLoader(urls, parent) { + + override def addURL(url: URL): Unit = { + super.addURL(url) + } + + override def getURLs(): Array[URL] = { + super.getURLs() + } + +} + +/** + * A mutable class loader that gives preference to its own URLs over the parent class loader + * when loading classes and resources. + */ +private[spark] class ChildFirstURLClassLoader(urls: Array[URL], parent: ClassLoader) + extends MutableURLClassLoader(urls, null) { + + private val parentClassLoader = new ParentClassLoader(parent) + + /** + * Used to implement fine-grained class loading locks similar to what is done by Java 7. This + * prevents deadlock issues when using non-hierarchical class loaders. + * + * Note that due to Java 6 compatibility (and some issues with implementing class loaders in + * Scala), Java 7's `ClassLoader.registerAsParallelCapable` method is not called. + */ + private val locks = new ConcurrentHashMap[String, Object]() + + override def loadClass(name: String, resolve: Boolean): Class[_] = { + var lock = locks.get(name) + if (lock == null) { + val newLock = new Object() + lock = locks.putIfAbsent(name, newLock) + if (lock == null) { + lock = newLock + } + } + + lock.synchronized { + try { + super.loadClass(name, resolve) + } catch { + case e: ClassNotFoundException => + parentClassLoader.loadClass(name, resolve) + } + } + } + + override def getResource(name: String): URL = { + val url = super.findResource(name) + val res = if (url != null) url else parentClassLoader.getResource(name) + res + } + + override def getResources(name: String): Enumeration[URL] = { + val urls = super.findResources(name) + val res = + if (urls != null && urls.hasMoreElements()) { + urls + } else { + parentClassLoader.getResources(name) + } + res + } + + override def addURL(url: URL) { + super.addURL(url) + } + +} diff --git a/core/src/main/scala/org/apache/spark/util/ParentClassLoader.scala b/core/src/main/scala/org/apache/spark/util/ParentClassLoader.scala index 3abc12681fe9a..6d8d9e8da3678 100644 --- a/core/src/main/scala/org/apache/spark/util/ParentClassLoader.scala +++ b/core/src/main/scala/org/apache/spark/util/ParentClassLoader.scala @@ -18,7 +18,7 @@ package org.apache.spark.util /** - * A class loader which makes findClass accesible to the child + * A class loader which makes some protected methods in ClassLoader accesible. */ private[spark] class ParentClassLoader(parent: ClassLoader) extends ClassLoader(parent) { @@ -29,4 +29,9 @@ private[spark] class ParentClassLoader(parent: ClassLoader) extends ClassLoader( override def loadClass(name: String): Class[_] = { super.loadClass(name) } + + override def loadClass(name: String, resolve: Boolean): Class[_] = { + super.loadClass(name, resolve) + } + } diff --git a/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala b/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala index 6ba03841f746b..eaec5a71e6819 100644 --- a/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala +++ b/core/src/main/scala/org/apache/spark/util/collection/ExternalSorter.scala @@ -763,6 +763,7 @@ private[spark] class ExternalSorter[K, V, C]( if (curWriteMetrics != null) { m.incShuffleBytesWritten(curWriteMetrics.shuffleBytesWritten) m.incShuffleWriteTime(curWriteMetrics.shuffleWriteTime) + m.incShuffleRecordsWritten(curWriteMetrics.shuffleRecordsWritten) } } diff --git a/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala b/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala index 9eb87f016068d..d3123e854016b 100644 --- a/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala +++ b/core/src/test/scala/org/apache/spark/ExecutorAllocationManagerSuite.scala @@ -175,6 +175,33 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { assert(numExecutorsPending(manager) === 9) } + test("cancel pending executors when no longer needed") { + sc = createSparkContext(1, 10) + val manager = sc.executorAllocationManager.get + sc.listenerBus.postToAll(SparkListenerStageSubmitted(createStageInfo(2, 5))) + + assert(numExecutorsPending(manager) === 0) + assert(numExecutorsToAdd(manager) === 1) + assert(addExecutors(manager) === 1) + assert(numExecutorsPending(manager) === 1) + assert(numExecutorsToAdd(manager) === 2) + assert(addExecutors(manager) === 2) + assert(numExecutorsPending(manager) === 3) + + val task1Info = createTaskInfo(0, 0, "executor-1") + sc.listenerBus.postToAll(SparkListenerTaskStart(2, 0, task1Info)) + + assert(numExecutorsToAdd(manager) === 4) + assert(addExecutors(manager) === 2) + + val task2Info = createTaskInfo(1, 0, "executor-1") + sc.listenerBus.postToAll(SparkListenerTaskStart(2, 0, task2Info)) + sc.listenerBus.postToAll(SparkListenerTaskEnd(2, 0, null, null, task1Info, null)) + sc.listenerBus.postToAll(SparkListenerTaskEnd(2, 0, null, null, task2Info, null)) + + assert(adjustRequestedExecutors(manager) === -1) + } + test("remove executors") { sc = createSparkContext(5, 10) val manager = sc.executorAllocationManager.get @@ -270,15 +297,15 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { assert(removeExecutor(manager, "5")) assert(removeExecutor(manager, "6")) assert(executorIds(manager).size === 10) - assert(addExecutors(manager) === 0) // still at upper limit + assert(addExecutors(manager) === 1) onExecutorRemoved(manager, "3") onExecutorRemoved(manager, "4") assert(executorIds(manager).size === 8) // Add succeeds again, now that we are no longer at the upper limit // Number of executors added restarts at 1 - assert(addExecutors(manager) === 1) - assert(addExecutors(manager) === 1) // upper limit reached again + assert(addExecutors(manager) === 2) + assert(addExecutors(manager) === 1) // upper limit reached assert(addExecutors(manager) === 0) assert(executorIds(manager).size === 8) onExecutorRemoved(manager, "5") @@ -286,9 +313,7 @@ class ExecutorAllocationManagerSuite extends FunSuite with LocalSparkContext { onExecutorAdded(manager, "13") onExecutorAdded(manager, "14") assert(executorIds(manager).size === 8) - assert(addExecutors(manager) === 1) - assert(addExecutors(manager) === 1) // upper limit reached again - assert(addExecutors(manager) === 0) + assert(addExecutors(manager) === 0) // still at upper limit onExecutorAdded(manager, "15") onExecutorAdded(manager, "16") assert(executorIds(manager).size === 10) @@ -679,6 +704,7 @@ private object ExecutorAllocationManagerSuite extends PrivateMethodTester { private val _numExecutorsToAdd = PrivateMethod[Int]('numExecutorsToAdd) private val _numExecutorsPending = PrivateMethod[Int]('numExecutorsPending) + private val _maxNumExecutorsNeeded = PrivateMethod[Int]('maxNumExecutorsNeeded) private val _executorsPendingToRemove = PrivateMethod[collection.Set[String]]('executorsPendingToRemove) private val _executorIds = PrivateMethod[collection.Set[String]]('executorIds) @@ -686,6 +712,7 @@ private object ExecutorAllocationManagerSuite extends PrivateMethodTester { private val _removeTimes = PrivateMethod[collection.Map[String, Long]]('removeTimes) private val _schedule = PrivateMethod[Unit]('schedule) private val _addExecutors = PrivateMethod[Int]('addExecutors) + private val _addOrCancelExecutorRequests = PrivateMethod[Int]('addOrCancelExecutorRequests) private val _removeExecutor = PrivateMethod[Boolean]('removeExecutor) private val _onExecutorAdded = PrivateMethod[Unit]('onExecutorAdded) private val _onExecutorRemoved = PrivateMethod[Unit]('onExecutorRemoved) @@ -724,7 +751,12 @@ private object ExecutorAllocationManagerSuite extends PrivateMethodTester { } private def addExecutors(manager: ExecutorAllocationManager): Int = { - manager invokePrivate _addExecutors() + val maxNumExecutorsNeeded = manager invokePrivate _maxNumExecutorsNeeded() + manager invokePrivate _addExecutors(maxNumExecutorsNeeded) + } + + private def adjustRequestedExecutors(manager: ExecutorAllocationManager): Int = { + manager invokePrivate _addOrCancelExecutorRequests(0L) } private def removeExecutor(manager: ExecutorAllocationManager, id: String): Boolean = { diff --git a/core/src/test/scala/org/apache/spark/SparkConfSuite.scala b/core/src/test/scala/org/apache/spark/SparkConfSuite.scala index e08210ae60d17..ea6b73bc68b34 100644 --- a/core/src/test/scala/org/apache/spark/SparkConfSuite.scala +++ b/core/src/test/scala/org/apache/spark/SparkConfSuite.scala @@ -197,6 +197,18 @@ class SparkConfSuite extends FunSuite with LocalSparkContext with ResetSystemPro serializer.newInstance().serialize(new StringBuffer()) } + test("deprecated config keys") { + val conf = new SparkConf() + .set("spark.files.userClassPathFirst", "true") + .set("spark.yarn.user.classpath.first", "true") + assert(conf.contains("spark.files.userClassPathFirst")) + assert(conf.contains("spark.executor.userClassPathFirst")) + assert(conf.contains("spark.yarn.user.classpath.first")) + assert(conf.getBoolean("spark.files.userClassPathFirst", false)) + assert(conf.getBoolean("spark.executor.userClassPathFirst", false)) + assert(conf.getBoolean("spark.yarn.user.classpath.first", false)) + } + } class Class1 {} diff --git a/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala index ed02ca81e405c..e955636cf5b59 100644 --- a/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/JsonProtocolSuite.scala @@ -68,7 +68,8 @@ class JsonProtocolSuite extends FunSuite { val completedApps = Array[ApplicationInfo]() val activeDrivers = Array(createDriverInfo()) val completedDrivers = Array(createDriverInfo()) - val stateResponse = new MasterStateResponse("host", 8080, workers, activeApps, completedApps, + val stateResponse = new MasterStateResponse( + "host", 8080, None, workers, activeApps, completedApps, activeDrivers, completedDrivers, RecoveryState.ALIVE) val output = JsonProtocol.writeMasterState(stateResponse) assertValidJson(output) diff --git a/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala b/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala index 3f1355f82893e..46d745c4ecbfa 100644 --- a/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala @@ -21,6 +21,8 @@ import java.io._ import scala.collection.mutable.ArrayBuffer +import com.google.common.base.Charsets.UTF_8 +import com.google.common.io.ByteStreams import org.scalatest.FunSuite import org.scalatest.Matchers import org.scalatest.concurrent.Timeouts @@ -141,7 +143,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) val childArgsStr = childArgs.mkString(" ") childArgsStr should include ("--class org.SomeClass") childArgsStr should include ("--executor-memory 5g") @@ -180,7 +182,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) childArgs.mkString(" ") should be ("arg1 arg2") mainClass should be ("org.SomeClass") classpath should have length (4) @@ -201,6 +203,18 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties } test("handles standalone cluster mode") { + testStandaloneCluster(useRest = true) + } + + test("handles legacy standalone cluster mode") { + testStandaloneCluster(useRest = false) + } + + /** + * Test whether the launch environment is correctly set up in standalone cluster mode. + * @param useRest whether to use the REST submission gateway introduced in Spark 1.3 + */ + private def testStandaloneCluster(useRest: Boolean): Unit = { val clArgs = Seq( "--deploy-mode", "cluster", "--master", "spark://h:p", @@ -212,17 +226,26 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + appArgs.useRest = useRest + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) val childArgsStr = childArgs.mkString(" ") - childArgsStr should startWith ("--memory 4g --cores 5 --supervise") - childArgsStr should include regex ("launch spark://h:p .*thejar.jar org.SomeClass arg1 arg2") - mainClass should be ("org.apache.spark.deploy.Client") - classpath should have size (0) - sysProps should have size (5) + if (useRest) { + childArgsStr should endWith ("thejar.jar org.SomeClass arg1 arg2") + mainClass should be ("org.apache.spark.deploy.rest.StandaloneRestClient") + } else { + childArgsStr should startWith ("--supervise --memory 4g --cores 5") + childArgsStr should include regex "launch spark://h:p .*thejar.jar org.SomeClass arg1 arg2" + mainClass should be ("org.apache.spark.deploy.Client") + } + classpath should have size 0 + sysProps should have size 8 sysProps.keys should contain ("SPARK_SUBMIT") sysProps.keys should contain ("spark.master") sysProps.keys should contain ("spark.app.name") sysProps.keys should contain ("spark.jars") + sysProps.keys should contain ("spark.driver.memory") + sysProps.keys should contain ("spark.driver.cores") + sysProps.keys should contain ("spark.driver.supervise") sysProps.keys should contain ("spark.shuffle.spill") sysProps("spark.shuffle.spill") should be ("false") } @@ -239,7 +262,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) childArgs.mkString(" ") should be ("arg1 arg2") mainClass should be ("org.SomeClass") classpath should have length (1) @@ -261,7 +284,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (childArgs, classpath, sysProps, mainClass) = createLaunchEnv(appArgs) + val (childArgs, classpath, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) childArgs.mkString(" ") should be ("arg1 arg2") mainClass should be ("org.SomeClass") classpath should have length (1) @@ -281,7 +304,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar", "arg1", "arg2") val appArgs = new SparkSubmitArguments(clArgs) - val (_, _, sysProps, mainClass) = createLaunchEnv(appArgs) + val (_, _, sysProps, mainClass) = prepareSubmitEnvironment(appArgs) sysProps("spark.executor.memory") should be ("5g") sysProps("spark.master") should be ("yarn-cluster") mainClass should be ("org.apache.spark.deploy.yarn.Client") @@ -339,7 +362,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "--files", files, "thejar.jar") val appArgs = new SparkSubmitArguments(clArgs) - val sysProps = SparkSubmit.createLaunchEnv(appArgs)._3 + val sysProps = SparkSubmit.prepareSubmitEnvironment(appArgs)._3 appArgs.jars should be (Utils.resolveURIs(jars)) appArgs.files should be (Utils.resolveURIs(files)) sysProps("spark.jars") should be (Utils.resolveURIs(jars + ",thejar.jar")) @@ -354,7 +377,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar" ) val appArgs2 = new SparkSubmitArguments(clArgs2) - val sysProps2 = SparkSubmit.createLaunchEnv(appArgs2)._3 + val sysProps2 = SparkSubmit.prepareSubmitEnvironment(appArgs2)._3 appArgs2.files should be (Utils.resolveURIs(files)) appArgs2.archives should be (Utils.resolveURIs(archives)) sysProps2("spark.yarn.dist.files") should be (Utils.resolveURIs(files)) @@ -367,7 +390,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "mister.py" ) val appArgs3 = new SparkSubmitArguments(clArgs3) - val sysProps3 = SparkSubmit.createLaunchEnv(appArgs3)._3 + val sysProps3 = SparkSubmit.prepareSubmitEnvironment(appArgs3)._3 appArgs3.pyFiles should be (Utils.resolveURIs(pyFiles)) sysProps3("spark.submit.pyFiles") should be ( PythonRunner.formatPaths(Utils.resolveURIs(pyFiles)).mkString(",")) @@ -392,7 +415,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar" ) val appArgs = new SparkSubmitArguments(clArgs) - val sysProps = SparkSubmit.createLaunchEnv(appArgs)._3 + val sysProps = SparkSubmit.prepareSubmitEnvironment(appArgs)._3 sysProps("spark.jars") should be(Utils.resolveURIs(jars + ",thejar.jar")) sysProps("spark.files") should be(Utils.resolveURIs(files)) @@ -409,7 +432,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "thejar.jar" ) val appArgs2 = new SparkSubmitArguments(clArgs2) - val sysProps2 = SparkSubmit.createLaunchEnv(appArgs2)._3 + val sysProps2 = SparkSubmit.prepareSubmitEnvironment(appArgs2)._3 sysProps2("spark.yarn.dist.files") should be(Utils.resolveURIs(files)) sysProps2("spark.yarn.dist.archives") should be(Utils.resolveURIs(archives)) @@ -424,11 +447,24 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties "mister.py" ) val appArgs3 = new SparkSubmitArguments(clArgs3) - val sysProps3 = SparkSubmit.createLaunchEnv(appArgs3)._3 + val sysProps3 = SparkSubmit.prepareSubmitEnvironment(appArgs3)._3 sysProps3("spark.submit.pyFiles") should be( PythonRunner.formatPaths(Utils.resolveURIs(pyFiles)).mkString(",")) } + test("user classpath first in driver") { + val systemJar = TestUtils.createJarWithFiles(Map("test.resource" -> "SYSTEM")) + val userJar = TestUtils.createJarWithFiles(Map("test.resource" -> "USER")) + val args = Seq( + "--class", UserClasspathFirstTest.getClass.getName.stripSuffix("$"), + "--name", "testApp", + "--master", "local", + "--conf", "spark.driver.extraClassPath=" + systemJar, + "--conf", "spark.driver.userClassPathFirst=true", + userJar.toString) + runSparkSubmit(args) + } + test("SPARK_CONF_DIR overrides spark-defaults.conf") { forConfDir(Map("spark.executor.memory" -> "2.3g")) { path => val unusedJar = TestUtils.createJarWithClasses(Seq.empty) @@ -440,7 +476,7 @@ class SparkSubmitSuite extends FunSuite with Matchers with ResetSystemProperties val appArgs = new SparkSubmitArguments(args, Map("SPARK_CONF_DIR" -> path)) assert(appArgs.propertiesFile != null) assert(appArgs.propertiesFile.startsWith(path)) - appArgs.executorMemory should be ("2.3g") + appArgs.executorMemory should be ("2.3g") } } @@ -520,3 +556,15 @@ object SimpleApplicationTest { } } } + +object UserClasspathFirstTest { + def main(args: Array[String]) { + val ccl = Thread.currentThread().getContextClassLoader() + val resource = ccl.getResourceAsStream("test.resource") + val bytes = ByteStreams.toByteArray(resource) + val contents = new String(bytes, 0, bytes.length, UTF_8) + if (contents != "USER") { + throw new SparkException("Should have read user resource, but instead read: " + contents) + } + } +} diff --git a/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala b/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala index 1d95432258111..85939eaadccc7 100644 --- a/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala +++ b/core/src/test/scala/org/apache/spark/deploy/history/FsHistoryProviderSuite.scala @@ -37,13 +37,8 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers private var testDir: File = null - private var provider: FsHistoryProvider = null - before { testDir = Utils.createTempDir() - provider = new FsHistoryProvider(new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.history.fs.updateInterval", "0")) } after { @@ -51,40 +46,41 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("Parse new and old application logs") { - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.history.fs.updateInterval", "0") - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) // Write a new-style application log. - val logFile1 = new File(testDir, "new1") - writeFile(logFile1, true, None, - SparkListenerApplicationStart("app1-1", None, 1L, "test"), - SparkListenerApplicationEnd(2L) + val newAppComplete = new File(testDir, "new1") + writeFile(newAppComplete, true, None, + SparkListenerApplicationStart("new-app-complete", None, 1L, "test"), + SparkListenerApplicationEnd(4L) ) // Write an unfinished app, new-style. - val logFile2 = new File(testDir, "new2" + EventLoggingListener.IN_PROGRESS) - writeFile(logFile2, true, None, - SparkListenerApplicationStart("app2-2", None, 1L, "test") + val newAppIncomplete = new File(testDir, "new2" + EventLoggingListener.IN_PROGRESS) + writeFile(newAppIncomplete, true, None, + SparkListenerApplicationStart("new-app-incomplete", None, 1L, "test") ) // Write an old-style application log. - val oldLog = new File(testDir, "old1") - oldLog.mkdir() - createEmptyFile(new File(oldLog, provider.SPARK_VERSION_PREFIX + "1.0")) - writeFile(new File(oldLog, provider.LOG_PREFIX + "1"), false, None, - SparkListenerApplicationStart("app3", None, 2L, "test"), + val oldAppComplete = new File(testDir, "old1") + oldAppComplete.mkdir() + createEmptyFile(new File(oldAppComplete, provider.SPARK_VERSION_PREFIX + "1.0")) + writeFile(new File(oldAppComplete, provider.LOG_PREFIX + "1"), false, None, + SparkListenerApplicationStart("old-app-complete", None, 2L, "test"), SparkListenerApplicationEnd(3L) ) - createEmptyFile(new File(oldLog, provider.APPLICATION_COMPLETE)) + createEmptyFile(new File(oldAppComplete, provider.APPLICATION_COMPLETE)) + + // Check for logs so that we force the older unfinished app to be loaded, to make + // sure unfinished apps are also sorted correctly. + provider.checkForLogs() // Write an unfinished app, old-style. - val oldLog2 = new File(testDir, "old2") - oldLog2.mkdir() - createEmptyFile(new File(oldLog2, provider.SPARK_VERSION_PREFIX + "1.0")) - writeFile(new File(oldLog2, provider.LOG_PREFIX + "1"), false, None, - SparkListenerApplicationStart("app4", None, 2L, "test") + val oldAppIncomplete = new File(testDir, "old2") + oldAppIncomplete.mkdir() + createEmptyFile(new File(oldAppIncomplete, provider.SPARK_VERSION_PREFIX + "1.0")) + writeFile(new File(oldAppIncomplete, provider.LOG_PREFIX + "1"), false, None, + SparkListenerApplicationStart("old-app-incomplete", None, 2L, "test") ) // Force a reload of data from the log directory, and check that both logs are loaded. @@ -96,14 +92,14 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers list.size should be (4) list.count(e => e.completed) should be (2) - list(0) should be (ApplicationHistoryInfo(oldLog.getName(), "app3", 2L, 3L, - oldLog.lastModified(), "test", true)) - list(1) should be (ApplicationHistoryInfo(logFile1.getName(), "app1-1", 1L, 2L, - logFile1.lastModified(), "test", true)) - list(2) should be (ApplicationHistoryInfo(oldLog2.getName(), "app4", 2L, -1L, - oldLog2.lastModified(), "test", false)) - list(3) should be (ApplicationHistoryInfo(logFile2.getName(), "app2-2", 1L, -1L, - logFile2.lastModified(), "test", false)) + list(0) should be (ApplicationHistoryInfo(newAppComplete.getName(), "new-app-complete", 1L, 4L, + newAppComplete.lastModified(), "test", true)) + list(1) should be (ApplicationHistoryInfo(oldAppComplete.getName(), "old-app-complete", 2L, 3L, + oldAppComplete.lastModified(), "test", true)) + list(2) should be (ApplicationHistoryInfo(oldAppIncomplete.getName(), "old-app-incomplete", 2L, + -1L, oldAppIncomplete.lastModified(), "test", false)) + list(3) should be (ApplicationHistoryInfo(newAppIncomplete.getName(), "new-app-incomplete", 1L, + -1L, newAppIncomplete.lastModified(), "test", false)) // Make sure the UI can be rendered. list.foreach { case info => @@ -113,6 +109,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("Parse legacy logs with compression codec set") { + val provider = new FsHistoryProvider(createTestConf()) val testCodecs = List((classOf[LZFCompressionCodec].getName(), true), (classOf[SnappyCompressionCodec].getName(), true), ("invalid.codec", false)) @@ -156,10 +153,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers ) logFile2.setReadable(false, false) - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.history.fs.updateInterval", "0") - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) provider.checkForLogs() val list = provider.getListing().toSeq @@ -168,10 +162,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("history file is renamed from inprogress to completed") { - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - .set("spark.testing", "true") - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) val logFile1 = new File(testDir, "app1" + EventLoggingListener.IN_PROGRESS) writeFile(logFile1, true, None, @@ -191,9 +182,7 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers } test("SPARK-5582: empty log directory") { - val conf = new SparkConf() - .set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) - val provider = new FsHistoryProvider(conf) + val provider = new FsHistoryProvider(createTestConf()) val logFile1 = new File(testDir, "app1" + EventLoggingListener.IN_PROGRESS) writeFile(logFile1, true, None, @@ -229,4 +218,8 @@ class FsHistoryProviderSuite extends FunSuite with BeforeAndAfter with Matchers new FileOutputStream(file).close() } + private def createTestConf(): SparkConf = { + new SparkConf().set("spark.history.fs.logDirectory", testDir.getAbsolutePath()) + } + } diff --git a/core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala b/core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala new file mode 100644 index 0000000000000..29aed89b67aa7 --- /dev/null +++ b/core/src/test/scala/org/apache/spark/deploy/rest/StandaloneRestSubmitSuite.scala @@ -0,0 +1,265 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.io.{File, FileInputStream, FileOutputStream, PrintWriter} +import java.util.jar.{JarEntry, JarOutputStream} +import java.util.zip.ZipEntry + +import scala.collection.mutable.ArrayBuffer +import scala.io.Source + +import akka.actor.ActorSystem +import com.google.common.io.ByteStreams +import org.scalatest.{BeforeAndAfterAll, BeforeAndAfterEach, FunSuite} +import org.scalatest.exceptions.TestFailedException + +import org.apache.spark._ +import org.apache.spark.util.Utils +import org.apache.spark.deploy.{SparkSubmit, SparkSubmitArguments} +import org.apache.spark.deploy.master.{DriverState, Master} +import org.apache.spark.deploy.worker.Worker + +/** + * End-to-end tests for the REST application submission protocol in standalone mode. + */ +class StandaloneRestSubmitSuite extends FunSuite with BeforeAndAfterAll with BeforeAndAfterEach { + private val systemsToStop = new ArrayBuffer[ActorSystem] + private val masterRestUrl = startLocalCluster() + private val client = new StandaloneRestClient + private val mainJar = StandaloneRestSubmitSuite.createJar() + private val mainClass = StandaloneRestApp.getClass.getName.stripSuffix("$") + + override def afterAll() { + systemsToStop.foreach(_.shutdown()) + } + + test("simple submit until completion") { + val resultsFile = File.createTempFile("test-submit", ".txt") + val numbers = Seq(1, 2, 3) + val size = 500 + val submissionId = submitApplication(resultsFile, numbers, size) + waitUntilFinished(submissionId) + validateResult(resultsFile, numbers, size) + } + + test("kill empty submission") { + val response = client.killSubmission(masterRestUrl, "submission-that-does-not-exist") + val killResponse = getKillResponse(response) + val killSuccess = killResponse.success + assert(!killSuccess) + } + + test("kill running submission") { + val resultsFile = File.createTempFile("test-kill", ".txt") + val numbers = Seq(1, 2, 3) + val size = 500 + val submissionId = submitApplication(resultsFile, numbers, size) + val response = client.killSubmission(masterRestUrl, submissionId) + val killResponse = getKillResponse(response) + val killSuccess = killResponse.success + waitUntilFinished(submissionId) + val response2 = client.requestSubmissionStatus(masterRestUrl, submissionId) + val statusResponse = getStatusResponse(response2) + val statusSuccess = statusResponse.success + val driverState = statusResponse.driverState + assert(killSuccess) + assert(statusSuccess) + assert(driverState === DriverState.KILLED.toString) + // we should not see the expected results because we killed the submission + intercept[TestFailedException] { validateResult(resultsFile, numbers, size) } + } + + test("request status for empty submission") { + val response = client.requestSubmissionStatus(masterRestUrl, "submission-that-does-not-exist") + val statusResponse = getStatusResponse(response) + val statusSuccess = statusResponse.success + assert(!statusSuccess) + } + + /** + * Start a local cluster containing one Master and a few Workers. + * Do not use [[org.apache.spark.deploy.LocalSparkCluster]] here because we want the REST URL. + * Return the Master's REST URL to which applications should be submitted. + */ + private def startLocalCluster(): String = { + val conf = new SparkConf(false) + .set("spark.master.rest.enabled", "true") + .set("spark.master.rest.port", "0") + val (numWorkers, coresPerWorker, memPerWorker) = (2, 1, 512) + val localHostName = Utils.localHostName() + val (masterSystem, masterPort, _, _masterRestPort) = + Master.startSystemAndActor(localHostName, 0, 0, conf) + val masterRestPort = _masterRestPort.getOrElse { fail("REST server not started on Master!") } + val masterUrl = "spark://" + localHostName + ":" + masterPort + val masterRestUrl = "spark://" + localHostName + ":" + masterRestPort + (1 to numWorkers).foreach { n => + val (workerSystem, _) = Worker.startSystemAndActor( + localHostName, 0, 0, coresPerWorker, memPerWorker, Array(masterUrl), null, Some(n)) + systemsToStop.append(workerSystem) + } + systemsToStop.append(masterSystem) + masterRestUrl + } + + /** Submit the [[StandaloneRestApp]] and return the corresponding submission ID. */ + private def submitApplication(resultsFile: File, numbers: Seq[Int], size: Int): String = { + val appArgs = Seq(resultsFile.getAbsolutePath) ++ numbers.map(_.toString) ++ Seq(size.toString) + val commandLineArgs = Array( + "--deploy-mode", "cluster", + "--master", masterRestUrl, + "--name", mainClass, + "--class", mainClass, + mainJar) ++ appArgs + val args = new SparkSubmitArguments(commandLineArgs) + val (_, _, sparkProperties, _) = SparkSubmit.prepareSubmitEnvironment(args) + val request = client.constructSubmitRequest( + mainJar, mainClass, appArgs.toArray, sparkProperties.toMap, Map.empty) + val response = client.createSubmission(masterRestUrl, request) + val submitResponse = getSubmitResponse(response) + val submissionId = submitResponse.submissionId + assert(submissionId != null, "Application submission was unsuccessful!") + submissionId + } + + /** Wait until the given submission has finished running up to the specified timeout. */ + private def waitUntilFinished(submissionId: String, maxSeconds: Int = 30): Unit = { + var finished = false + val expireTime = System.currentTimeMillis + maxSeconds * 1000 + while (!finished) { + val response = client.requestSubmissionStatus(masterRestUrl, submissionId) + val statusResponse = getStatusResponse(response) + val driverState = statusResponse.driverState + finished = + driverState != DriverState.SUBMITTED.toString && + driverState != DriverState.RUNNING.toString + if (System.currentTimeMillis > expireTime) { + fail(s"Driver $submissionId did not finish within $maxSeconds seconds.") + } + } + } + + /** Return the response as a submit response, or fail with error otherwise. */ + private def getSubmitResponse(response: SubmitRestProtocolResponse): CreateSubmissionResponse = { + response match { + case s: CreateSubmissionResponse => s + case e: ErrorResponse => fail(s"Server returned error: ${e.message}") + case r => fail(s"Expected submit response. Actual: ${r.toJson}") + } + } + + /** Return the response as a kill response, or fail with error otherwise. */ + private def getKillResponse(response: SubmitRestProtocolResponse): KillSubmissionResponse = { + response match { + case k: KillSubmissionResponse => k + case e: ErrorResponse => fail(s"Server returned error: ${e.message}") + case r => fail(s"Expected kill response. Actual: ${r.toJson}") + } + } + + /** Return the response as a status response, or fail with error otherwise. */ + private def getStatusResponse(response: SubmitRestProtocolResponse): SubmissionStatusResponse = { + response match { + case s: SubmissionStatusResponse => s + case e: ErrorResponse => fail(s"Server returned error: ${e.message}") + case r => fail(s"Expected status response. Actual: ${r.toJson}") + } + } + + /** Validate whether the application produced the corrupt output. */ + private def validateResult(resultsFile: File, numbers: Seq[Int], size: Int): Unit = { + val lines = Source.fromFile(resultsFile.getAbsolutePath).getLines().toSeq + val unexpectedContent = + if (lines.nonEmpty) { + "[\n" + lines.map { l => " " + l }.mkString("\n") + "\n]" + } else { + "[EMPTY]" + } + assert(lines.size === 2, s"Unexpected content in file: $unexpectedContent") + assert(lines(0).toInt === numbers.sum, s"Sum of ${numbers.mkString(",")} is incorrect") + assert(lines(1).toInt === (size / 2) + 1, "Result of Spark job is incorrect") + } +} + +private object StandaloneRestSubmitSuite { + private val pathPrefix = this.getClass.getPackage.getName.replaceAll("\\.", "/") + + /** + * Create a jar that contains all the class files needed for running the [[StandaloneRestApp]]. + * Return the absolute path to that jar. + */ + def createJar(): String = { + val jarFile = File.createTempFile("test-standalone-rest-protocol", ".jar") + val jarFileStream = new FileOutputStream(jarFile) + val jarStream = new JarOutputStream(jarFileStream, new java.util.jar.Manifest) + jarStream.putNextEntry(new ZipEntry(pathPrefix)) + getClassFiles.foreach { cf => + jarStream.putNextEntry(new JarEntry(pathPrefix + "/" + cf.getName)) + val in = new FileInputStream(cf) + ByteStreams.copy(in, jarStream) + in.close() + } + jarStream.close() + jarFileStream.close() + jarFile.getAbsolutePath + } + + /** + * Return a list of class files compiled for [[StandaloneRestApp]]. + * This includes all the anonymous classes used in the application. + */ + private def getClassFiles: Seq[File] = { + val className = Utils.getFormattedClassName(StandaloneRestApp) + val clazz = StandaloneRestApp.getClass + val basePath = clazz.getProtectionDomain.getCodeSource.getLocation.toURI.getPath + val baseDir = new File(basePath + "/" + pathPrefix) + baseDir.listFiles().filter(_.getName.contains(className)) + } +} + +/** + * Sample application to be submitted to the cluster using the REST gateway. + * All relevant classes will be packaged into a jar at run time. + */ +object StandaloneRestApp { + // Usage: [path to results file] [num1] [num2] [num3] [rddSize] + // The first line of the results file should be (num1 + num2 + num3) + // The second line should be (rddSize / 2) + 1 + def main(args: Array[String]) { + assert(args.size == 5, s"Expected exactly 5 arguments: ${args.mkString(",")}") + val resultFile = new File(args(0)) + val writer = new PrintWriter(resultFile) + try { + val conf = new SparkConf() + val sc = new SparkContext(conf) + val firstLine = args(1).toInt + args(2).toInt + args(3).toInt + val secondLine = sc.parallelize(1 to args(4).toInt) + .map { i => (i / 2, i) } + .reduceByKey(_ + _) + .count() + writer.println(firstLine) + writer.println(secondLine) + } catch { + case e: Exception => + writer.println(e) + e.getStackTrace.foreach { l => writer.println(" " + l) } + } finally { + writer.close() + } + } +} diff --git a/core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala b/core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala new file mode 100644 index 0000000000000..1d64ec201e647 --- /dev/null +++ b/core/src/test/scala/org/apache/spark/deploy/rest/SubmitRestProtocolSuite.scala @@ -0,0 +1,324 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.deploy.rest + +import java.lang.Boolean +import java.lang.Integer + +import org.json4s.jackson.JsonMethods._ +import org.scalatest.FunSuite + +import org.apache.spark.SparkConf + +/** + * Tests for the REST application submission protocol. + */ +class SubmitRestProtocolSuite extends FunSuite { + + test("validate") { + val request = new DummyRequest + intercept[SubmitRestProtocolException] { request.validate() } // missing everything + request.clientSparkVersion = "1.2.3" + intercept[SubmitRestProtocolException] { request.validate() } // missing name and age + request.name = "something" + intercept[SubmitRestProtocolException] { request.validate() } // missing only age + request.age = 2 + intercept[SubmitRestProtocolException] { request.validate() } // age too low + request.age = 10 + request.validate() // everything is set properly + request.clientSparkVersion = null + intercept[SubmitRestProtocolException] { request.validate() } // missing only Spark version + request.clientSparkVersion = "1.2.3" + request.name = null + intercept[SubmitRestProtocolException] { request.validate() } // missing only name + request.message = "not-setting-name" + intercept[SubmitRestProtocolException] { request.validate() } // still missing name + } + + test("request to and from JSON") { + val request = new DummyRequest + intercept[SubmitRestProtocolException] { request.toJson } // implicit validation + request.clientSparkVersion = "1.2.3" + request.active = true + request.age = 25 + request.name = "jung" + val json = request.toJson + assertJsonEquals(json, dummyRequestJson) + val newRequest = SubmitRestProtocolMessage.fromJson(json, classOf[DummyRequest]) + assert(newRequest.clientSparkVersion === "1.2.3") + assert(newRequest.clientSparkVersion === "1.2.3") + assert(newRequest.active) + assert(newRequest.age === 25) + assert(newRequest.name === "jung") + assert(newRequest.message === null) + } + + test("response to and from JSON") { + val response = new DummyResponse + response.serverSparkVersion = "3.3.4" + response.success = true + val json = response.toJson + assertJsonEquals(json, dummyResponseJson) + val newResponse = SubmitRestProtocolMessage.fromJson(json, classOf[DummyResponse]) + assert(newResponse.serverSparkVersion === "3.3.4") + assert(newResponse.serverSparkVersion === "3.3.4") + assert(newResponse.success) + assert(newResponse.message === null) + } + + test("CreateSubmissionRequest") { + val message = new CreateSubmissionRequest + intercept[SubmitRestProtocolException] { message.validate() } + message.clientSparkVersion = "1.2.3" + message.appResource = "honey-walnut-cherry.jar" + message.mainClass = "org.apache.spark.examples.SparkPie" + val conf = new SparkConf(false) + conf.set("spark.app.name", "SparkPie") + message.sparkProperties = conf.getAll.toMap + message.validate() + // optional fields + conf.set("spark.jars", "mayonnaise.jar,ketchup.jar") + conf.set("spark.files", "fireball.png") + conf.set("spark.driver.memory", "512m") + conf.set("spark.driver.cores", "180") + conf.set("spark.driver.extraJavaOptions", " -Dslices=5 -Dcolor=mostly_red") + conf.set("spark.driver.extraClassPath", "food-coloring.jar") + conf.set("spark.driver.extraLibraryPath", "pickle.jar") + conf.set("spark.driver.supervise", "false") + conf.set("spark.executor.memory", "256m") + conf.set("spark.cores.max", "10000") + message.sparkProperties = conf.getAll.toMap + message.appArgs = Array("two slices", "a hint of cinnamon") + message.environmentVariables = Map("PATH" -> "/dev/null") + message.validate() + // bad fields + var badConf = conf.clone().set("spark.driver.cores", "one hundred feet") + message.sparkProperties = badConf.getAll.toMap + intercept[SubmitRestProtocolException] { message.validate() } + badConf = conf.clone().set("spark.driver.supervise", "nope, never") + message.sparkProperties = badConf.getAll.toMap + intercept[SubmitRestProtocolException] { message.validate() } + badConf = conf.clone().set("spark.cores.max", "two men") + message.sparkProperties = badConf.getAll.toMap + intercept[SubmitRestProtocolException] { message.validate() } + message.sparkProperties = conf.getAll.toMap + // test JSON + val json = message.toJson + assertJsonEquals(json, submitDriverRequestJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[CreateSubmissionRequest]) + assert(newMessage.clientSparkVersion === "1.2.3") + assert(newMessage.appResource === "honey-walnut-cherry.jar") + assert(newMessage.mainClass === "org.apache.spark.examples.SparkPie") + assert(newMessage.sparkProperties("spark.app.name") === "SparkPie") + assert(newMessage.sparkProperties("spark.jars") === "mayonnaise.jar,ketchup.jar") + assert(newMessage.sparkProperties("spark.files") === "fireball.png") + assert(newMessage.sparkProperties("spark.driver.memory") === "512m") + assert(newMessage.sparkProperties("spark.driver.cores") === "180") + assert(newMessage.sparkProperties("spark.driver.extraJavaOptions") === " -Dslices=5 -Dcolor=mostly_red") + assert(newMessage.sparkProperties("spark.driver.extraClassPath") === "food-coloring.jar") + assert(newMessage.sparkProperties("spark.driver.extraLibraryPath") === "pickle.jar") + assert(newMessage.sparkProperties("spark.driver.supervise") === "false") + assert(newMessage.sparkProperties("spark.executor.memory") === "256m") + assert(newMessage.sparkProperties("spark.cores.max") === "10000") + assert(newMessage.appArgs === message.appArgs) + assert(newMessage.sparkProperties === message.sparkProperties) + assert(newMessage.environmentVariables === message.environmentVariables) + } + + test("CreateSubmissionResponse") { + val message = new CreateSubmissionResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.submissionId = "driver_123" + message.success = true + message.validate() + // test JSON + val json = message.toJson + assertJsonEquals(json, submitDriverResponseJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[CreateSubmissionResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.submissionId === "driver_123") + assert(newMessage.success) + } + + test("KillSubmissionResponse") { + val message = new KillSubmissionResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.submissionId = "driver_123" + message.success = true + message.validate() + // test JSON + val json = message.toJson + assertJsonEquals(json, killDriverResponseJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[KillSubmissionResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.submissionId === "driver_123") + assert(newMessage.success) + } + + test("SubmissionStatusResponse") { + val message = new SubmissionStatusResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.submissionId = "driver_123" + message.success = true + message.validate() + // optional fields + message.driverState = "RUNNING" + message.workerId = "worker_123" + message.workerHostPort = "1.2.3.4:7780" + // test JSON + val json = message.toJson + assertJsonEquals(json, driverStatusResponseJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[SubmissionStatusResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.submissionId === "driver_123") + assert(newMessage.driverState === "RUNNING") + assert(newMessage.success) + assert(newMessage.workerId === "worker_123") + assert(newMessage.workerHostPort === "1.2.3.4:7780") + } + + test("ErrorResponse") { + val message = new ErrorResponse + intercept[SubmitRestProtocolException] { message.validate() } + message.serverSparkVersion = "1.2.3" + message.message = "Field not found in submit request: X" + message.validate() + // test JSON + val json = message.toJson + assertJsonEquals(json, errorJson) + val newMessage = SubmitRestProtocolMessage.fromJson(json, classOf[ErrorResponse]) + assert(newMessage.serverSparkVersion === "1.2.3") + assert(newMessage.message === "Field not found in submit request: X") + } + + private val dummyRequestJson = + """ + |{ + | "action" : "DummyRequest", + | "active" : true, + | "age" : 25, + | "clientSparkVersion" : "1.2.3", + | "name" : "jung" + |} + """.stripMargin + + private val dummyResponseJson = + """ + |{ + | "action" : "DummyResponse", + | "serverSparkVersion" : "3.3.4", + | "success": true + |} + """.stripMargin + + private val submitDriverRequestJson = + """ + |{ + | "action" : "CreateSubmissionRequest", + | "appArgs" : [ "two slices", "a hint of cinnamon" ], + | "appResource" : "honey-walnut-cherry.jar", + | "clientSparkVersion" : "1.2.3", + | "environmentVariables" : { + | "PATH" : "/dev/null" + | }, + | "mainClass" : "org.apache.spark.examples.SparkPie", + | "sparkProperties" : { + | "spark.driver.extraLibraryPath" : "pickle.jar", + | "spark.jars" : "mayonnaise.jar,ketchup.jar", + | "spark.driver.supervise" : "false", + | "spark.app.name" : "SparkPie", + | "spark.cores.max" : "10000", + | "spark.driver.memory" : "512m", + | "spark.files" : "fireball.png", + | "spark.driver.cores" : "180", + | "spark.driver.extraJavaOptions" : " -Dslices=5 -Dcolor=mostly_red", + | "spark.executor.memory" : "256m", + | "spark.driver.extraClassPath" : "food-coloring.jar" + | } + |} + """.stripMargin + + private val submitDriverResponseJson = + """ + |{ + | "action" : "CreateSubmissionResponse", + | "serverSparkVersion" : "1.2.3", + | "submissionId" : "driver_123", + | "success" : true + |} + """.stripMargin + + private val killDriverResponseJson = + """ + |{ + | "action" : "KillSubmissionResponse", + | "serverSparkVersion" : "1.2.3", + | "submissionId" : "driver_123", + | "success" : true + |} + """.stripMargin + + private val driverStatusResponseJson = + """ + |{ + | "action" : "SubmissionStatusResponse", + | "driverState" : "RUNNING", + | "serverSparkVersion" : "1.2.3", + | "submissionId" : "driver_123", + | "success" : true, + | "workerHostPort" : "1.2.3.4:7780", + | "workerId" : "worker_123" + |} + """.stripMargin + + private val errorJson = + """ + |{ + | "action" : "ErrorResponse", + | "message" : "Field not found in submit request: X", + | "serverSparkVersion" : "1.2.3" + |} + """.stripMargin + + /** Assert that the contents in the two JSON strings are equal after ignoring whitespace. */ + private def assertJsonEquals(jsonString1: String, jsonString2: String): Unit = { + val trimmedJson1 = jsonString1.trim + val trimmedJson2 = jsonString2.trim + val json1 = compact(render(parse(trimmedJson1))) + val json2 = compact(render(parse(trimmedJson2))) + // Put this on a separate line to avoid printing comparison twice when test fails + val equals = json1 == json2 + assert(equals, "\"[%s]\" did not equal \"[%s]\"".format(trimmedJson1, trimmedJson2)) + } +} + +private class DummyResponse extends SubmitRestProtocolResponse +private class DummyRequest extends SubmitRestProtocolRequest { + var active: Boolean = null + var age: Integer = null + var name: String = null + protected override def doValidate(): Unit = { + super.doValidate() + assertFieldIsSet(name, "name") + assertFieldIsSet(age, "age") + assert(age > 5, "Not old enough!") + } +} diff --git a/core/src/test/scala/org/apache/spark/executor/TaskMetricsSuite.scala b/core/src/test/scala/org/apache/spark/executor/TaskMetricsSuite.scala new file mode 100644 index 0000000000000..326e203afe136 --- /dev/null +++ b/core/src/test/scala/org/apache/spark/executor/TaskMetricsSuite.scala @@ -0,0 +1,28 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.executor + +import org.scalatest.FunSuite + +class TaskMetricsSuite extends FunSuite { + test("[SPARK-5701] updateShuffleReadMetrics: ShuffleReadMetrics not added when no shuffle deps") { + val taskMetrics = new TaskMetrics() + taskMetrics.updateShuffleReadMetrics() + assert(taskMetrics.shuffleReadMetrics.isEmpty) + } +} diff --git a/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala b/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala index 81db66ae17464..78fa98a3b9065 100644 --- a/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala +++ b/core/src/test/scala/org/apache/spark/metrics/InputOutputMetricsSuite.scala @@ -21,44 +21,46 @@ import java.io.{File, FileWriter, PrintWriter} import scala.collection.mutable.ArrayBuffer -import org.scalatest.FunSuite - +import org.apache.commons.lang.math.RandomUtils import org.apache.hadoop.conf.Configuration import org.apache.hadoop.fs.{FileSystem, Path} import org.apache.hadoop.io.{LongWritable, Text} -import org.apache.hadoop.mapred.{FileSplit => OldFileSplit, InputSplit => OldInputSplit, JobConf, - LineRecordReader => OldLineRecordReader, RecordReader => OldRecordReader, Reporter, - TextInputFormat => OldTextInputFormat} import org.apache.hadoop.mapred.lib.{CombineFileInputFormat => OldCombineFileInputFormat, - CombineFileSplit => OldCombineFileSplit, CombineFileRecordReader => OldCombineFileRecordReader} -import org.apache.hadoop.mapreduce.{InputSplit => NewInputSplit, RecordReader => NewRecordReader, - TaskAttemptContext} + CombineFileRecordReader => OldCombineFileRecordReader, CombineFileSplit => OldCombineFileSplit} +import org.apache.hadoop.mapred.{JobConf, Reporter, FileSplit => OldFileSplit, + InputSplit => OldInputSplit, LineRecordReader => OldLineRecordReader, + RecordReader => OldRecordReader, TextInputFormat => OldTextInputFormat} import org.apache.hadoop.mapreduce.lib.input.{CombineFileInputFormat => NewCombineFileInputFormat, CombineFileRecordReader => NewCombineFileRecordReader, CombineFileSplit => NewCombineFileSplit, FileSplit => NewFileSplit, TextInputFormat => NewTextInputFormat} +import org.apache.hadoop.mapreduce.lib.output.{TextOutputFormat => NewTextOutputFormat} +import org.apache.hadoop.mapreduce.{TaskAttemptContext, InputSplit => NewInputSplit, + RecordReader => NewRecordReader} +import org.scalatest.{BeforeAndAfter, FunSuite} import org.apache.spark.SharedSparkContext import org.apache.spark.deploy.SparkHadoopUtil import org.apache.spark.scheduler.{SparkListener, SparkListenerTaskEnd} import org.apache.spark.util.Utils -class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { +class InputOutputMetricsSuite extends FunSuite with SharedSparkContext + with BeforeAndAfter { @transient var tmpDir: File = _ @transient var tmpFile: File = _ @transient var tmpFilePath: String = _ + @transient val numRecords: Int = 100000 + @transient val numBuckets: Int = 10 - override def beforeAll() { - super.beforeAll() - + before { tmpDir = Utils.createTempDir() val testTempDir = new File(tmpDir, "test") testTempDir.mkdir() tmpFile = new File(testTempDir, getClass.getSimpleName + ".txt") val pw = new PrintWriter(new FileWriter(tmpFile)) - for (x <- 1 to 1000000) { - pw.println("s") + for (x <- 1 to numRecords) { + pw.println(RandomUtils.nextInt(numBuckets)) } pw.close() @@ -66,8 +68,7 @@ class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { tmpFilePath = "file://" + tmpFile.getAbsolutePath } - override def afterAll() { - super.afterAll() + after { Utils.deleteRecursively(tmpDir) } @@ -155,6 +156,101 @@ class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { assert(bytesRead >= tmpFile.length()) } + test("input metrics on records read - simple") { + val records = runAndReturnRecordsRead { + sc.textFile(tmpFilePath, 4).count() + } + assert(records == numRecords) + } + + test("input metrics on records read - more stages") { + val records = runAndReturnRecordsRead { + sc.textFile(tmpFilePath, 4) + .map(key => (key.length, 1)) + .reduceByKey(_ + _) + .count() + } + assert(records == numRecords) + } + + test("input metrics on records - New Hadoop API") { + val records = runAndReturnRecordsRead { + sc.newAPIHadoopFile(tmpFilePath, classOf[NewTextInputFormat], classOf[LongWritable], + classOf[Text]).count() + } + assert(records == numRecords) + } + + test("input metrics on recordsd read with cache") { + // prime the cache manager + val rdd = sc.textFile(tmpFilePath, 4).cache() + rdd.collect() + + val records = runAndReturnRecordsRead { + rdd.count() + } + + assert(records == numRecords) + } + + test("shuffle records read metrics") { + val recordsRead = runAndReturnShuffleRecordsRead { + sc.textFile(tmpFilePath, 4) + .map(key => (key, 1)) + .groupByKey() + .collect() + } + assert(recordsRead == numRecords) + } + + test("shuffle records written metrics") { + val recordsWritten = runAndReturnShuffleRecordsWritten { + sc.textFile(tmpFilePath, 4) + .map(key => (key, 1)) + .groupByKey() + .collect() + } + assert(recordsWritten == numRecords) + } + + /** + * Tests the metrics from end to end. + * 1) reading a hadoop file + * 2) shuffle and writing to a hadoop file. + * 3) writing to hadoop file. + */ + test("input read/write and shuffle read/write metrics all line up") { + var inputRead = 0L + var outputWritten = 0L + var shuffleRead = 0L + var shuffleWritten = 0L + sc.addSparkListener(new SparkListener() { + override def onTaskEnd(taskEnd: SparkListenerTaskEnd) { + val metrics = taskEnd.taskMetrics + metrics.inputMetrics.foreach(inputRead += _.recordsRead) + metrics.outputMetrics.foreach(outputWritten += _.recordsWritten) + metrics.shuffleReadMetrics.foreach(shuffleRead += _.recordsRead) + metrics.shuffleWriteMetrics.foreach(shuffleWritten += _.shuffleRecordsWritten) + } + }) + + val tmpFile = new File(tmpDir, getClass.getSimpleName) + + sc.textFile(tmpFilePath, 4) + .map(key => (key, 1)) + .reduceByKey(_+_) + .saveAsTextFile("file://" + tmpFile.getAbsolutePath) + + sc.listenerBus.waitUntilEmpty(500) + assert(inputRead == numRecords) + + // Only supported on newer Hadoop + if (SparkHadoopUtil.get.getFSBytesWrittenOnThreadCallback().isDefined) { + assert(outputWritten == numBuckets) + } + assert(shuffleRead == shuffleWritten) + } + test("input metrics with interleaved reads") { val numPartitions = 2 val cartVector = 0 to 9 @@ -193,18 +289,66 @@ class InputOutputMetricsSuite extends FunSuite with SharedSparkContext { assert(cartesianBytes == firstSize * numPartitions + (cartVector.length * secondSize)) } - private def runAndReturnBytesRead(job : => Unit): Long = { - val taskBytesRead = new ArrayBuffer[Long]() + private def runAndReturnBytesRead(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.inputMetrics.map(_.bytesRead)) + } + + private def runAndReturnRecordsRead(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.inputMetrics.map(_.recordsRead)) + } + + private def runAndReturnRecordsWritten(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.outputMetrics.map(_.recordsWritten)) + } + + private def runAndReturnShuffleRecordsRead(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.shuffleReadMetrics.map(_.recordsRead)) + } + + private def runAndReturnShuffleRecordsWritten(job: => Unit): Long = { + runAndReturnMetrics(job, _.taskMetrics.shuffleWriteMetrics.map(_.shuffleRecordsWritten)) + } + + private def runAndReturnMetrics(job: => Unit, + collector: (SparkListenerTaskEnd) => Option[Long]): Long = { + val taskMetrics = new ArrayBuffer[Long]() sc.addSparkListener(new SparkListener() { override def onTaskEnd(taskEnd: SparkListenerTaskEnd) { - taskBytesRead += taskEnd.taskMetrics.inputMetrics.get.bytesRead + collector(taskEnd).foreach(taskMetrics += _) } }) job sc.listenerBus.waitUntilEmpty(500) - taskBytesRead.sum + taskMetrics.sum + } + + test("output metrics on records written") { + // Only supported on newer Hadoop + if (SparkHadoopUtil.get.getFSBytesWrittenOnThreadCallback().isDefined) { + val file = new File(tmpDir, getClass.getSimpleName) + val filePath = "file://" + file.getAbsolutePath + + val records = runAndReturnRecordsWritten { + sc.parallelize(1 to numRecords).saveAsTextFile(filePath) + } + assert(records == numRecords) + } + } + + test("output metrics on records written - new Hadoop API") { + // Only supported on newer Hadoop + if (SparkHadoopUtil.get.getFSBytesWrittenOnThreadCallback().isDefined) { + val file = new File(tmpDir, getClass.getSimpleName) + val filePath = "file://" + file.getAbsolutePath + + val records = runAndReturnRecordsWritten { + sc.parallelize(1 to numRecords).map(key => (key.toString, key.toString)) + .saveAsNewAPIHadoopFile[NewTextOutputFormat[String, String]](filePath) + } + assert(records == numRecords) + } } test("output metrics when writing text file") { @@ -318,4 +462,4 @@ class NewCombineTextRecordReaderWrapper( override def getCurrentValue(): Text = delegate.getCurrentValue override def getProgress(): Float = delegate.getProgress override def close(): Unit = delegate.close() -} \ No newline at end of file +} diff --git a/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala b/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala index 855f1b6276089..054a4c64897a9 100644 --- a/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala +++ b/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala @@ -29,9 +29,9 @@ class KryoSerializerDistributedSuite extends FunSuite { test("kryo objects are serialised consistently in different processes") { val conf = new SparkConf(false) - conf.set("spark.serializer", "org.apache.spark.serializer.KryoSerializer") - conf.set("spark.kryo.registrator", classOf[AppJarRegistrator].getName) - conf.set("spark.task.maxFailures", "1") + .set("spark.serializer", "org.apache.spark.serializer.KryoSerializer") + .set("spark.kryo.registrator", classOf[AppJarRegistrator].getName) + .set("spark.task.maxFailures", "1") val jar = TestUtils.createJarWithClasses(List(AppJarRegistrator.customClassName)) conf.setJars(List(jar.getPath)) diff --git a/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala b/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala index bbc7e1357b90d..c21c92b63ad13 100644 --- a/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala +++ b/core/src/test/scala/org/apache/spark/storage/BlockObjectWriterSuite.scala @@ -31,6 +31,8 @@ class BlockObjectWriterSuite extends FunSuite { new JavaSerializer(new SparkConf()), 1024, os => os, true, writeMetrics) writer.write(Long.box(20)) + // Record metrics update on every write + assert(writeMetrics.shuffleRecordsWritten === 1) // Metrics don't update on every write assert(writeMetrics.shuffleBytesWritten == 0) // After 32 writes, metrics should update @@ -39,6 +41,7 @@ class BlockObjectWriterSuite extends FunSuite { writer.write(Long.box(i)) } assert(writeMetrics.shuffleBytesWritten > 0) + assert(writeMetrics.shuffleRecordsWritten === 33) writer.commitAndClose() assert(file.length() == writeMetrics.shuffleBytesWritten) } @@ -51,6 +54,8 @@ class BlockObjectWriterSuite extends FunSuite { new JavaSerializer(new SparkConf()), 1024, os => os, true, writeMetrics) writer.write(Long.box(20)) + // Record metrics update on every write + assert(writeMetrics.shuffleRecordsWritten === 1) // Metrics don't update on every write assert(writeMetrics.shuffleBytesWritten == 0) // After 32 writes, metrics should update @@ -59,7 +64,23 @@ class BlockObjectWriterSuite extends FunSuite { writer.write(Long.box(i)) } assert(writeMetrics.shuffleBytesWritten > 0) + assert(writeMetrics.shuffleRecordsWritten === 33) writer.revertPartialWritesAndClose() assert(writeMetrics.shuffleBytesWritten == 0) + assert(writeMetrics.shuffleRecordsWritten == 0) + } + + test("Reopening a closed block writer") { + val file = new File("somefile") + file.deleteOnExit() + val writeMetrics = new ShuffleWriteMetrics() + val writer = new DiskBlockObjectWriter(new TestBlockId("0"), file, + new JavaSerializer(new SparkConf()), 1024, os => os, true, writeMetrics) + + writer.open() + writer.close() + intercept[IllegalStateException] { + writer.open() + } } } diff --git a/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala b/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala index 68074ae32a672..e8405baa8e3ea 100644 --- a/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala +++ b/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala @@ -234,7 +234,7 @@ class JobProgressListenerSuite extends FunSuite with LocalSparkContext with Matc taskMetrics.incMemoryBytesSpilled(base + 6) val inputMetrics = new InputMetrics(DataReadMethod.Hadoop) taskMetrics.setInputMetrics(Some(inputMetrics)) - inputMetrics.addBytesRead(base + 7) + inputMetrics.incBytesRead(base + 7) val outputMetrics = new OutputMetrics(DataWriteMethod.Hadoop) taskMetrics.outputMetrics = Some(outputMetrics) outputMetrics.setBytesWritten(base + 8) diff --git a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala index 842f54529baf0..f3017dc42cd5c 100644 --- a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala @@ -189,6 +189,34 @@ class JsonProtocolSuite extends FunSuite { assert(newMetrics.inputMetrics.isEmpty) } + test("Input/Output records backwards compatibility") { + // records read were added after 1.2 + val metrics = makeTaskMetrics(1L, 2L, 3L, 4L, 5, 6, + hasHadoopInput = true, hasOutput = true, hasRecords = false) + assert(metrics.inputMetrics.nonEmpty) + assert(metrics.outputMetrics.nonEmpty) + val newJson = JsonProtocol.taskMetricsToJson(metrics) + val oldJson = newJson.removeField { case (field, _) => field == "Records Read" } + .removeField { case (field, _) => field == "Records Written" } + val newMetrics = JsonProtocol.taskMetricsFromJson(oldJson) + assert(newMetrics.inputMetrics.get.recordsRead == 0) + assert(newMetrics.outputMetrics.get.recordsWritten == 0) + } + + test("Shuffle Read/Write records backwards compatibility") { + // records read were added after 1.2 + val metrics = makeTaskMetrics(1L, 2L, 3L, 4L, 5, 6, + hasHadoopInput = false, hasOutput = false, hasRecords = false) + assert(metrics.shuffleReadMetrics.nonEmpty) + assert(metrics.shuffleWriteMetrics.nonEmpty) + val newJson = JsonProtocol.taskMetricsToJson(metrics) + val oldJson = newJson.removeField { case (field, _) => field == "Total Records Read" } + .removeField { case (field, _) => field == "Shuffle Records Written" } + val newMetrics = JsonProtocol.taskMetricsFromJson(oldJson) + assert(newMetrics.shuffleReadMetrics.get.recordsRead == 0) + assert(newMetrics.shuffleWriteMetrics.get.shuffleRecordsWritten == 0) + } + test("OutputMetrics backward compatibility") { // OutputMetrics were added after 1.1 val metrics = makeTaskMetrics(1L, 2L, 3L, 4L, 5, 6, hasHadoopInput = false, hasOutput = true) @@ -644,7 +672,8 @@ class JsonProtocolSuite extends FunSuite { e: Int, f: Int, hasHadoopInput: Boolean, - hasOutput: Boolean) = { + hasOutput: Boolean, + hasRecords: Boolean = true) = { val t = new TaskMetrics t.setHostname("localhost") t.setExecutorDeserializeTime(a) @@ -656,7 +685,8 @@ class JsonProtocolSuite extends FunSuite { if (hasHadoopInput) { val inputMetrics = new InputMetrics(DataReadMethod.Hadoop) - inputMetrics.addBytesRead(d + e + f) + inputMetrics.incBytesRead(d + e + f) + inputMetrics.incRecordsRead(if (hasRecords) (d + e + f) / 100 else -1) t.setInputMetrics(Some(inputMetrics)) } else { val sr = new ShuffleReadMetrics @@ -664,16 +694,19 @@ class JsonProtocolSuite extends FunSuite { sr.incLocalBlocksFetched(e) sr.incFetchWaitTime(a + d) sr.incRemoteBlocksFetched(f) + sr.incRecordsRead(if (hasRecords) (b + d) / 100 else -1) t.setShuffleReadMetrics(Some(sr)) } if (hasOutput) { val outputMetrics = new OutputMetrics(DataWriteMethod.Hadoop) outputMetrics.setBytesWritten(a + b + c) + outputMetrics.setRecordsWritten(if (hasRecords) (a + b + c)/100 else -1) t.outputMetrics = Some(outputMetrics) } else { val sw = new ShuffleWriteMetrics sw.incShuffleBytesWritten(a + b + c) sw.incShuffleWriteTime(b + c + d) + sw.setShuffleRecordsWritten(if (hasRecords) (a + b + c) / 100 else -1) t.shuffleWriteMetrics = Some(sw) } // Make at most 6 blocks @@ -907,11 +940,13 @@ class JsonProtocolSuite extends FunSuite { | "Remote Blocks Fetched": 800, | "Local Blocks Fetched": 700, | "Fetch Wait Time": 900, - | "Remote Bytes Read": 1000 + | "Remote Bytes Read": 1000, + | "Total Records Read" : 10 | }, | "Shuffle Write Metrics": { | "Shuffle Bytes Written": 1200, - | "Shuffle Write Time": 1500 + | "Shuffle Write Time": 1500, + | "Shuffle Records Written": 12 | }, | "Updated Blocks": [ | { @@ -988,11 +1023,13 @@ class JsonProtocolSuite extends FunSuite { | "Disk Bytes Spilled": 0, | "Shuffle Write Metrics": { | "Shuffle Bytes Written": 1200, - | "Shuffle Write Time": 1500 + | "Shuffle Write Time": 1500, + | "Shuffle Records Written": 12 | }, | "Input Metrics": { | "Data Read Method": "Hadoop", - | "Bytes Read": 2100 + | "Bytes Read": 2100, + | "Records Read": 21 | }, | "Updated Blocks": [ | { @@ -1069,11 +1106,13 @@ class JsonProtocolSuite extends FunSuite { | "Disk Bytes Spilled": 0, | "Input Metrics": { | "Data Read Method": "Hadoop", - | "Bytes Read": 2100 + | "Bytes Read": 2100, + | "Records Read": 21 | }, | "Output Metrics": { | "Data Write Method": "Hadoop", - | "Bytes Written": 1200 + | "Bytes Written": 1200, + | "Records Written": 12 | }, | "Updated Blocks": [ | { diff --git a/core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala b/core/src/test/scala/org/apache/spark/util/MutableURLClassLoaderSuite.scala similarity index 90% rename from core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala rename to core/src/test/scala/org/apache/spark/util/MutableURLClassLoaderSuite.scala index b7912c09d1410..31e3b7e7bb71b 100644 --- a/core/src/test/scala/org/apache/spark/executor/ExecutorURLClassLoaderSuite.scala +++ b/core/src/test/scala/org/apache/spark/util/MutableURLClassLoaderSuite.scala @@ -15,7 +15,7 @@ * limitations under the License. */ -package org.apache.spark.executor +package org.apache.spark.util import java.net.URLClassLoader @@ -24,7 +24,7 @@ import org.scalatest.FunSuite import org.apache.spark.{LocalSparkContext, SparkContext, SparkException, TestUtils} import org.apache.spark.util.Utils -class ExecutorURLClassLoaderSuite extends FunSuite { +class MutableURLClassLoaderSuite extends FunSuite { val urls2 = List(TestUtils.createJarWithClasses( classNames = Seq("FakeClass1", "FakeClass2", "FakeClass3"), @@ -37,7 +37,7 @@ class ExecutorURLClassLoaderSuite extends FunSuite { test("child first") { val parentLoader = new URLClassLoader(urls2, null) - val classLoader = new ChildExecutorURLClassLoader(urls, parentLoader) + val classLoader = new ChildFirstURLClassLoader(urls, parentLoader) val fakeClass = classLoader.loadClass("FakeClass2").newInstance() val fakeClassVersion = fakeClass.toString assert(fakeClassVersion === "1") @@ -47,7 +47,7 @@ class ExecutorURLClassLoaderSuite extends FunSuite { test("parent first") { val parentLoader = new URLClassLoader(urls2, null) - val classLoader = new ExecutorURLClassLoader(urls, parentLoader) + val classLoader = new MutableURLClassLoader(urls, parentLoader) val fakeClass = classLoader.loadClass("FakeClass1").newInstance() val fakeClassVersion = fakeClass.toString assert(fakeClassVersion === "2") @@ -57,7 +57,7 @@ class ExecutorURLClassLoaderSuite extends FunSuite { test("child first can fall back") { val parentLoader = new URLClassLoader(urls2, null) - val classLoader = new ChildExecutorURLClassLoader(urls, parentLoader) + val classLoader = new ChildFirstURLClassLoader(urls, parentLoader) val fakeClass = classLoader.loadClass("FakeClass3").newInstance() val fakeClassVersion = fakeClass.toString assert(fakeClassVersion === "2") @@ -65,7 +65,7 @@ class ExecutorURLClassLoaderSuite extends FunSuite { test("child first can fail") { val parentLoader = new URLClassLoader(urls2, null) - val classLoader = new ChildExecutorURLClassLoader(urls, parentLoader) + val classLoader = new ChildFirstURLClassLoader(urls, parentLoader) intercept[java.lang.ClassNotFoundException] { classLoader.loadClass("FakeClassDoesNotExist").newInstance() } diff --git a/data/mllib/sample_lda_data.txt b/data/mllib/sample_lda_data.txt new file mode 100644 index 0000000000000..2e76702ca9d67 --- /dev/null +++ b/data/mllib/sample_lda_data.txt @@ -0,0 +1,12 @@ +1 2 6 0 2 3 1 1 0 0 3 +1 3 0 1 3 0 0 2 0 0 1 +1 4 1 0 0 4 9 0 1 2 0 +2 1 0 3 0 0 5 0 2 3 9 +3 1 1 9 3 0 2 0 0 1 3 +4 2 0 3 4 5 1 1 1 4 0 +2 1 0 3 0 0 5 0 2 2 9 +1 1 1 9 2 1 2 0 0 1 3 +4 4 0 3 4 2 1 3 0 0 0 +2 8 2 0 3 0 2 0 2 7 2 +1 1 1 9 0 2 2 0 0 3 3 +4 1 0 0 4 5 1 3 0 1 0 diff --git a/dev/run-tests b/dev/run-tests index 2257a566bb1bb..483958757a2dd 100755 --- a/dev/run-tests +++ b/dev/run-tests @@ -36,7 +36,7 @@ function handle_error () { } -# Build against the right verison of Hadoop. +# Build against the right version of Hadoop. { if [ -n "$AMPLAB_JENKINS_BUILD_PROFILE" ]; then if [ "$AMPLAB_JENKINS_BUILD_PROFILE" = "hadoop1.0" ]; then @@ -77,7 +77,7 @@ export SBT_MAVEN_PROFILES_ARGS="$SBT_MAVEN_PROFILES_ARGS -Pkinesis-asl" fi } -# Only run Hive tests if there are sql changes. +# Only run Hive tests if there are SQL changes. # Partial solution for SPARK-1455. if [ -n "$AMPLAB_JENKINS" ]; then git fetch origin master:master @@ -183,7 +183,7 @@ CURRENT_BLOCK=$BLOCK_SPARK_UNIT_TESTS if [ -n "$_SQL_TESTS_ONLY" ]; then # This must be an array of individual arguments. Otherwise, having one long string # will be interpreted as a single test, which doesn't work. - SBT_MAVEN_TEST_ARGS=("catalyst/test" "sql/test" "hive/test" "mllib/test") + SBT_MAVEN_TEST_ARGS=("catalyst/test" "sql/test" "hive/test" "hive-thriftserver/test" "mllib/test") else SBT_MAVEN_TEST_ARGS=("test") fi diff --git a/docs/configuration.md b/docs/configuration.md index 00e973c245005..eb0d6d33c97d9 100644 --- a/docs/configuration.md +++ b/docs/configuration.md @@ -230,6 +230,15 @@ Apart from these, the following properties are also available, and may be useful Set a special library path to use when launching the driver JVM. + + + + + @@ -297,13 +306,11 @@ Apart from these, the following properties are also available, and may be useful - + @@ -865,8 +872,8 @@ Apart from these, the following properties are also available, and may be useful @@ -911,8 +918,8 @@ Apart from these, the following properties are also available, and may be useful @@ -920,7 +927,7 @@ Apart from these, the following properties are also available, and may be useful @@ -930,7 +937,7 @@ Apart from these, the following properties are also available, and may be useful @@ -939,7 +946,7 @@ Apart from these, the following properties are also available, and may be useful
    Executor IDTotal Tasks Failed Tasks Succeeded TasksInputOutputShuffle ReadShuffle WriteShuffle Spill (Memory)Shuffle Spill (Disk) + Input Size / Records + + Output Size / Records + + + Shuffle Read Size / Records + + + Shuffle Write Size / Records + Shuffle Spill (Memory)Shuffle Spill (Disk)
    {v.failedTasks + v.succeededTasks} {v.failedTasks} {v.succeededTasks} - {Utils.bytesToString(v.inputBytes)} - {Utils.bytesToString(v.outputBytes)} - {Utils.bytesToString(v.shuffleRead)} - {Utils.bytesToString(v.shuffleWrite)} - {Utils.bytesToString(v.memoryBytesSpilled)} - {Utils.bytesToString(v.diskBytesSpilled)} + {s"${Utils.bytesToString(v.inputBytes)} / ${v.inputRecords}"} + + {s"${Utils.bytesToString(v.outputBytes)} / ${v.outputRecords}"} + + {s"${Utils.bytesToString(v.shuffleRead)} / ${v.shuffleReadRecords}"} + + {s"${Utils.bytesToString(v.shuffleWrite)} / ${v.shuffleWriteRecords}"} + + {Utils.bytesToString(v.memoryBytesSpilled)} + + {Utils.bytesToString(v.diskBytesSpilled)} +
    {UIUtils.formatDuration(millis.toLong)}{Utils.bytesToString(d.toLong)}{Utils.bytesToString(d.toLong)}{s"${Utils.bytesToString(d.toLong)} / ${recordDist.next().toLong}"}InputInput Size / RecordsOutputOutput Size / RecordsShuffle Read (Remote)Shuffle Read Size / Records (Remote)Shuffle WriteShuffle Write Size / Records
    - {inputReadable} + {s"$inputReadable / $inputRecords"} - {outputReadable} + {s"$outputReadable / $outputRecords"} - {shuffleReadReadable} + {s"$shuffleReadReadable / $shuffleReadRecords"} - {shuffleWriteReadable} + {s"$shuffleWriteReadable / $shuffleWriteRecords"}
    spark.driver.userClassPathFirstfalse + (Experimental) Whether to give user-added jars precedence over Spark's own jars when loading + classes in the the driver. This feature can be used to mitigate conflicts between Spark's + dependencies and user dependencies. It is currently an experimental feature. +
    spark.executor.extraJavaOptions (none)
    spark.files.userClassPathFirstspark.executor.userClassPathFirst false - (Experimental) Whether to give user-added jars precedence over Spark's own jars when - loading classes in Executors. This feature can be used to mitigate conflicts between - Spark's dependencies and user dependencies. It is currently an experimental feature. - (Currently, this setting does not work for YARN, see SPARK-2996 for more details). + (Experimental) Same functionality as spark.driver.userClassPathFirst, but + applied to executor instances.
    spark.network.timeout 120 - Default timeout for all network interactions, in seconds. This config will be used in - place of spark.core.connection.ack.wait.timeout, spark.akka.timeout, + Default timeout for all network interactions, in seconds. This config will be used in + place of spark.core.connection.ack.wait.timeout, spark.akka.timeout, spark.storage.blockManagerSlaveTimeoutMs or spark.shuffle.io.connectionTimeout, if they are not configured. spark.shuffle.io.preferDirectBufs true - (Netty only) Off-heap buffers are used to reduce garbage collection during shuffle and cache - block transfer. For environments where off-heap memory is tightly limited, users may wish to + (Netty only) Off-heap buffers are used to reduce garbage collection during shuffle and cache + block transfer. For environments where off-heap memory is tightly limited, users may wish to turn this off to force all allocations from Netty to be on-heap.
    spark.shuffle.io.numConnectionsPerPeer 1 - (Netty only) Connections between hosts are reused in order to reduce connection buildup for + (Netty only) Connections between hosts are reused in order to reduce connection buildup for large clusters. For clusters with many hard disks and few hosts, this may result in insufficient concurrency to saturate all disks, and so users may consider increasing this value. 3 (Netty only) Fetches that fail due to IO-related exceptions are automatically retried if this is - set to a non-zero value. This retry logic helps stabilize large shuffles in the face of long GC + set to a non-zero value. This retry logic helps stabilize large shuffles in the face of long GC pauses or transient network connectivity issues.
    5 (Netty only) Seconds to wait between retries of fetches. The maximum delay caused by retrying - is simply maxRetries * retryWait, by default 15 seconds. + is simply maxRetries * retryWait, by default 15 seconds.
    diff --git a/docs/mllib-clustering.md b/docs/mllib-clustering.md index 1e9ef345b7435..99ed6b60e3f00 100644 --- a/docs/mllib-clustering.md +++ b/docs/mllib-clustering.md @@ -55,7 +55,7 @@ has the following parameters: Power iteration clustering is a scalable and efficient algorithm for clustering points given pointwise mutual affinity values. Internally the algorithm: -* accepts a [Graph](https://spark.apache.org/docs/0.9.2/api/graphx/index.html#org.apache.spark.graphx.Graph) that represents a normalized pairwise affinity between all input points. +* accepts a [Graph](api/graphx/index.html#org.apache.spark.graphx.Graph) that represents a normalized pairwise affinity between all input points. * calculates the principal eigenvalue and eigenvector * Clusters each of the input points according to their principal eigenvector component value @@ -71,6 +71,35 @@ Example outputs for a dataset inspired by the paper - but with five clusters ins

    +### Latent Dirichlet Allocation (LDA) + +[Latent Dirichlet Allocation (LDA)](http://en.wikipedia.org/wiki/Latent_Dirichlet_allocation) +is a topic model which infers topics from a collection of text documents. +LDA can be thought of as a clustering algorithm as follows: + +* Topics correspond to cluster centers, and documents correspond to examples (rows) in a dataset. +* Topics and documents both exist in a feature space, where feature vectors are vectors of word counts. +* Rather than estimating a clustering using a traditional distance, LDA uses a function based + on a statistical model of how text documents are generated. + +LDA takes in a collection of documents as vectors of word counts. +It learns clustering using [expectation-maximization](http://en.wikipedia.org/wiki/Expectation%E2%80%93maximization_algorithm) +on the likelihood function. After fitting on the documents, LDA provides: + +* Topics: Inferred topics, each of which is a probability distribution over terms (words). +* Topic distributions for documents: For each document in the training set, LDA gives a probability distribution over topics. + +LDA takes the following parameters: + +* `k`: Number of topics (i.e., cluster centers) +* `maxIterations`: Limit on the number of iterations of EM used for learning +* `docConcentration`: Hyperparameter for prior over documents' distributions over topics. Currently must be > 1, where larger values encourage smoother inferred distributions. +* `topicConcentration`: Hyperparameter for prior over topics' distributions over terms (words). Currently must be > 1, where larger values encourage smoother inferred distributions. +* `checkpointInterval`: If using checkpointing (set in the Spark configuration), this parameter specifies the frequency with which checkpoints will be created. If `maxIterations` is large, using checkpointing can help reduce shuffle file sizes on disk and help with failure recovery. + +*Note*: LDA is a new feature with some missing functionality. In particular, it does not yet +support prediction on new documents, and it does not have a Python API. These will be added in the future. + ### Examples #### k-means @@ -293,6 +322,104 @@ for i in range(2):
+#### Latent Dirichlet Allocation (LDA) Example + +In the following example, we load word count vectors representing a corpus of documents. +We then use [LDA](api/scala/index.html#org.apache.spark.mllib.clustering.LDA) +to infer three topics from the documents. The number of desired clusters is passed +to the algorithm. We then output the topics, represented as probability distributions over words. + +
+
+ +{% highlight scala %} +import org.apache.spark.mllib.clustering.LDA +import org.apache.spark.mllib.linalg.Vectors + +// Load and parse the data +val data = sc.textFile("data/mllib/sample_lda_data.txt") +val parsedData = data.map(s => Vectors.dense(s.trim.split(' ').map(_.toDouble))) +// Index documents with unique IDs +val corpus = parsedData.zipWithIndex.map(_.swap).cache() + +// Cluster the documents into three topics using LDA +val ldaModel = new LDA().setK(3).run(corpus) + +// Output topics. Each is a distribution over words (matching word count vectors) +println("Learned topics (as distributions over vocab of " + ldaModel.vocabSize + " words):") +val topics = ldaModel.topicsMatrix +for (topic <- Range(0, 3)) { + print("Topic " + topic + ":") + for (word <- Range(0, ldaModel.vocabSize)) { print(" " + topics(word, topic)); } + println() +} +{% endhighlight %} +
+ +
+{% highlight java %} +import scala.Tuple2; + +import org.apache.spark.api.java.*; +import org.apache.spark.api.java.function.Function; +import org.apache.spark.mllib.clustering.DistributedLDAModel; +import org.apache.spark.mllib.clustering.LDA; +import org.apache.spark.mllib.linalg.Matrix; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.SparkConf; + +public class JavaLDAExample { + public static void main(String[] args) { + SparkConf conf = new SparkConf().setAppName("LDA Example"); + JavaSparkContext sc = new JavaSparkContext(conf); + + // Load and parse the data + String path = "data/mllib/sample_lda_data.txt"; + JavaRDD data = sc.textFile(path); + JavaRDD parsedData = data.map( + new Function() { + public Vector call(String s) { + String[] sarray = s.trim().split(" "); + double[] values = new double[sarray.length]; + for (int i = 0; i < sarray.length; i++) + values[i] = Double.parseDouble(sarray[i]); + return Vectors.dense(values); + } + } + ); + // Index documents with unique IDs + JavaPairRDD corpus = JavaPairRDD.fromJavaRDD(parsedData.zipWithIndex().map( + new Function, Tuple2>() { + public Tuple2 call(Tuple2 doc_id) { + return doc_id.swap(); + } + } + )); + corpus.cache(); + + // Cluster the documents into three topics using LDA + DistributedLDAModel ldaModel = new LDA().setK(3).run(corpus); + + // Output topics. Each is a distribution over words (matching word count vectors) + System.out.println("Learned topics (as distributions over vocab of " + ldaModel.vocabSize() + + " words):"); + Matrix topics = ldaModel.topicsMatrix(); + for (int topic = 0; topic < 3; topic++) { + System.out.print("Topic " + topic + ":"); + for (int word = 0; word < ldaModel.vocabSize(); word++) { + System.out.print(" " + topics.apply(word, topic)); + } + System.out.println(); + } + } +} +{% endhighlight %} +
+ +
+ + In order to run the above application, follow the instructions provided in the [Self-Contained Applications](quick-start.html#self-contained-applications) section of the Spark diff --git a/docs/mllib-guide.md b/docs/mllib-guide.md index 7779fbc9c49e4..3d32d03e35c62 100644 --- a/docs/mllib-guide.md +++ b/docs/mllib-guide.md @@ -56,25 +56,32 @@ See the **[spark.ml programming guide](ml-guide.html)** for more information on # Dependencies -MLlib uses the linear algebra package [Breeze](http://www.scalanlp.org/), -which depends on [netlib-java](https://github.com/fommil/netlib-java), -and [jblas](https://github.com/mikiobraun/jblas). -`netlib-java` and `jblas` depend on native Fortran routines. -You need to install the +MLlib uses the linear algebra package +[Breeze](http://www.scalanlp.org/), which depends on +[netlib-java](https://github.com/fommil/netlib-java) for optimised +numerical processing. If natives are not available at runtime, you +will see a warning message and a pure JVM implementation will be used +instead. + +To learn more about the benefits and background of system optimised +natives, you may wish to watch Sam Halliday's ScalaX talk on +[High Performance Linear Algebra in Scala](http://fommil.github.io/scalax14/#/)). + +Due to licensing issues with runtime proprietary binaries, we do not +include `netlib-java`'s native proxies by default. To configure +`netlib-java` / Breeze to use system optimised binaries, include +`com.github.fommil.netlib:all:1.1.2` (or build Spark with +`-Pnetlib-lgpl`) as a dependency of your project and read the +[netlib-java](https://github.com/fommil/netlib-java) documentation for +your platform's additional installation instructions. + +MLlib also uses [jblas](https://github.com/mikiobraun/jblas) which +will require you to install the [gfortran runtime library](https://github.com/mikiobraun/jblas/wiki/Missing-Libraries) if it is not already present on your nodes. -MLlib will throw a linking error if it cannot detect these libraries automatically. -Due to license issues, we do not include `netlib-java`'s native libraries in MLlib's -dependency set under default settings. -If no native library is available at runtime, you will see a warning message. -To use native libraries from `netlib-java`, please build Spark with `-Pnetlib-lgpl` or -include `com.github.fommil.netlib:all:1.1.2` as a dependency of your project. -If you want to use optimized BLAS/LAPACK libraries such as -[OpenBLAS](http://www.openblas.net/), please link its shared libraries to -`/usr/lib/libblas.so.3` and `/usr/lib/liblapack.so.3`, respectively. -BLAS/LAPACK libraries on worker nodes should be built without multithreading. - -To use MLlib in Python, you will need [NumPy](http://www.numpy.org) version 1.4 or newer. + +To use MLlib in Python, you will need [NumPy](http://www.numpy.org) +version 1.4 or newer. --- diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py index 0de4a62e203fd..0ea7365d75b83 100755 --- a/ec2/spark_ec2.py +++ b/ec2/spark_ec2.py @@ -24,14 +24,17 @@ import hashlib import logging import os +import os.path import pipes import random import shutil import string +from stat import S_IRUSR import subprocess import sys import tarfile import tempfile +import textwrap import time import urllib2 import warnings @@ -59,10 +62,10 @@ DEFAULT_SPARK_VERSION = SPARK_EC2_VERSION DEFAULT_SPARK_GITHUB_REPO = "https://github.com/apache/spark" -MESOS_SPARK_EC2_BRANCH = "branch-1.3" -# A URL prefix from which to fetch AMI information -AMI_PREFIX = "https://raw.github.com/mesos/spark-ec2/{b}/ami-list".format(b=MESOS_SPARK_EC2_BRANCH) +# Default location to get the spark-ec2 scripts (and ami-list) from +DEFAULT_SPARK_EC2_GITHUB_REPO = "https://github.com/mesos/spark-ec2" +DEFAULT_SPARK_EC2_BRANCH = "branch-1.3" def setup_boto(): @@ -109,6 +112,7 @@ def parse_args(): version="%prog {v}".format(v=SPARK_EC2_VERSION), usage="%prog [options] \n\n" + " can be: launch, destroy, login, stop, start, get-master, reboot-slaves") + parser.add_option( "-s", "--slaves", type="int", default=1, help="Number of slaves to launch (default: %default)") @@ -130,13 +134,15 @@ def parse_args(): help="Master instance type (leave empty for same as instance-type)") parser.add_option( "-r", "--region", default="us-east-1", - help="EC2 region zone to launch instances in") + help="EC2 region used to launch instances in, or to find them in") parser.add_option( "-z", "--zone", default="", help="Availability zone to launch instances in, or 'all' to spread " + "slaves across multiple (an additional $0.01/Gb for bandwidth" + "between zones applies) (default: a single zone chosen at random)") - parser.add_option("-a", "--ami", help="Amazon Machine Image ID to use") + parser.add_option( + "-a", "--ami", + help="Amazon Machine Image ID to use") parser.add_option( "-v", "--spark-version", default=DEFAULT_SPARK_VERSION, help="Version of Spark to use: 'X.Y.Z' or a specific git hash (default: %default)") @@ -144,6 +150,14 @@ def parse_args(): "--spark-git-repo", default=DEFAULT_SPARK_GITHUB_REPO, help="Github repo from which to checkout supplied commit hash (default: %default)") + parser.add_option( + "--spark-ec2-git-repo", + default=DEFAULT_SPARK_EC2_GITHUB_REPO, + help="Github repo from which to checkout spark-ec2 (default: %default)") + parser.add_option( + "--spark-ec2-git-branch", + default=DEFAULT_SPARK_EC2_BRANCH, + help="Github repo branch of spark-ec2 to use (default: %default)") parser.add_option( "--hadoop-major-version", default="1", help="Major version of Hadoop (default: %default)") @@ -168,10 +182,11 @@ def parse_args(): "Only possible on EBS-backed AMIs. " + "EBS volumes are only attached if --ebs-vol-size > 0." + "Only support up to 8 EBS volumes.") - parser.add_option("--placement-group", type="string", default=None, - help="Which placement group to try and launch " + - "instances into. Assumes placement group is already " + - "created.") + parser.add_option( + "--placement-group", type="string", default=None, + help="Which placement group to try and launch " + + "instances into. Assumes placement group is already " + + "created.") parser.add_option( "--swap", metavar="SWAP", type="int", default=1024, help="Swap space to set up per node, in MB (default: %default)") @@ -215,9 +230,11 @@ def parse_args(): "--copy-aws-credentials", action="store_true", default=False, help="Add AWS credentials to hadoop configuration to allow Spark to access S3") parser.add_option( - "--subnet-id", default=None, help="VPC subnet to launch instances in") + "--subnet-id", default=None, + help="VPC subnet to launch instances in") parser.add_option( - "--vpc-id", default=None, help="VPC to launch instances in") + "--vpc-id", default=None, + help="VPC to launch instances in") (opts, args) = parser.parse_args() if len(args) != 2: @@ -279,58 +296,65 @@ def is_active(instance): return (instance.state in ['pending', 'running', 'stopping', 'stopped']) -# Attempt to resolve an appropriate AMI given the architecture and region of the request. # Source: http://aws.amazon.com/amazon-linux-ami/instance-type-matrix/ # Last Updated: 2014-06-20 # For easy maintainability, please keep this manually-inputted dictionary sorted by key. +EC2_INSTANCE_TYPES = { + "c1.medium": "pvm", + "c1.xlarge": "pvm", + "c3.2xlarge": "pvm", + "c3.4xlarge": "pvm", + "c3.8xlarge": "pvm", + "c3.large": "pvm", + "c3.xlarge": "pvm", + "cc1.4xlarge": "hvm", + "cc2.8xlarge": "hvm", + "cg1.4xlarge": "hvm", + "cr1.8xlarge": "hvm", + "hi1.4xlarge": "pvm", + "hs1.8xlarge": "pvm", + "i2.2xlarge": "hvm", + "i2.4xlarge": "hvm", + "i2.8xlarge": "hvm", + "i2.xlarge": "hvm", + "m1.large": "pvm", + "m1.medium": "pvm", + "m1.small": "pvm", + "m1.xlarge": "pvm", + "m2.2xlarge": "pvm", + "m2.4xlarge": "pvm", + "m2.xlarge": "pvm", + "m3.2xlarge": "hvm", + "m3.large": "hvm", + "m3.medium": "hvm", + "m3.xlarge": "hvm", + "r3.2xlarge": "hvm", + "r3.4xlarge": "hvm", + "r3.8xlarge": "hvm", + "r3.large": "hvm", + "r3.xlarge": "hvm", + "t1.micro": "pvm", + "t2.medium": "hvm", + "t2.micro": "hvm", + "t2.small": "hvm", +} + + +# Attempt to resolve an appropriate AMI given the architecture and region of the request. def get_spark_ami(opts): - instance_types = { - "c1.medium": "pvm", - "c1.xlarge": "pvm", - "c3.2xlarge": "pvm", - "c3.4xlarge": "pvm", - "c3.8xlarge": "pvm", - "c3.large": "pvm", - "c3.xlarge": "pvm", - "cc1.4xlarge": "hvm", - "cc2.8xlarge": "hvm", - "cg1.4xlarge": "hvm", - "cr1.8xlarge": "hvm", - "hi1.4xlarge": "pvm", - "hs1.8xlarge": "pvm", - "i2.2xlarge": "hvm", - "i2.4xlarge": "hvm", - "i2.8xlarge": "hvm", - "i2.xlarge": "hvm", - "m1.large": "pvm", - "m1.medium": "pvm", - "m1.small": "pvm", - "m1.xlarge": "pvm", - "m2.2xlarge": "pvm", - "m2.4xlarge": "pvm", - "m2.xlarge": "pvm", - "m3.2xlarge": "hvm", - "m3.large": "hvm", - "m3.medium": "hvm", - "m3.xlarge": "hvm", - "r3.2xlarge": "hvm", - "r3.4xlarge": "hvm", - "r3.8xlarge": "hvm", - "r3.large": "hvm", - "r3.xlarge": "hvm", - "t1.micro": "pvm", - "t2.medium": "hvm", - "t2.micro": "hvm", - "t2.small": "hvm", - } - if opts.instance_type in instance_types: - instance_type = instance_types[opts.instance_type] + if opts.instance_type in EC2_INSTANCE_TYPES: + instance_type = EC2_INSTANCE_TYPES[opts.instance_type] else: instance_type = "pvm" print >> stderr,\ "Don't recognize %s, assuming type is pvm" % opts.instance_type - ami_path = "%s/%s/%s" % (AMI_PREFIX, opts.region, instance_type) + # URL prefix from which to fetch AMI information + ami_prefix = "{r}/{b}/ami-list".format( + r=opts.spark_ec2_git_repo.replace("https://github.com", "https://raw.github.com", 1), + b=opts.spark_ec2_git_branch) + + ami_path = "%s/%s/%s" % (ami_prefix, opts.region, instance_type) try: ami = urllib2.urlopen(ami_path).read().strip() print "Spark AMI: " + ami @@ -349,6 +373,7 @@ def launch_cluster(conn, opts, cluster_name): if opts.identity_file is None: print >> stderr, "ERROR: Must provide an identity file (-i) for ssh connections." sys.exit(1) + if opts.key_pair is None: print >> stderr, "ERROR: Must provide a key pair name (-k) to use on instances." sys.exit(1) @@ -569,6 +594,9 @@ def launch_cluster(conn, opts, cluster_name): master_nodes = master_res.instances print "Launched master in %s, regid = %s" % (zone, master_res.id) + # This wait time corresponds to SPARK-4983 + print "Waiting for AWS to propagate instance metadata..." + time.sleep(5) # Give the instances descriptive names for master in master_nodes: master.add_tag( @@ -585,10 +613,9 @@ def launch_cluster(conn, opts, cluster_name): # Get the EC2 instances in an existing cluster if available. # Returns a tuple of lists of EC2 instance objects for the masters and slaves - - def get_existing_cluster(conn, opts, cluster_name, die_on_error=True): - print "Searching for existing cluster " + cluster_name + "..." + print "Searching for existing cluster " + cluster_name + " in region " \ + + opts.region + "..." reservations = conn.get_all_reservations() master_nodes = [] slave_nodes = [] @@ -606,9 +633,11 @@ def get_existing_cluster(conn, opts, cluster_name, die_on_error=True): return (master_nodes, slave_nodes) else: if master_nodes == [] and slave_nodes != []: - print >> sys.stderr, "ERROR: Could not find master in group " + cluster_name + "-master" + print >> sys.stderr, "ERROR: Could not find master in group " + cluster_name \ + + "-master" + " in region " + opts.region else: - print >> sys.stderr, "ERROR: Could not find any existing cluster" + print >> sys.stderr, "ERROR: Could not find any existing cluster" \ + + " in region " + opts.region sys.exit(1) @@ -643,12 +672,15 @@ def setup_cluster(conn, master_nodes, slave_nodes, opts, deploy_ssh_key): # NOTE: We should clone the repository before running deploy_files to # prevent ec2-variables.sh from being overwritten + print "Cloning spark-ec2 scripts from {r}/tree/{b} on master...".format( + r=opts.spark_ec2_git_repo, b=opts.spark_ec2_git_branch) ssh( host=master, opts=opts, command="rm -rf spark-ec2" + " && " - + "git clone https://github.com/mesos/spark-ec2.git -b {b}".format(b=MESOS_SPARK_EC2_BRANCH) + + "git clone {r} -b {b} spark-ec2".format(r=opts.spark_ec2_git_repo, + b=opts.spark_ec2_git_branch) ) print "Deploying files to master..." @@ -675,21 +707,32 @@ def setup_spark_cluster(master, opts): print "Ganglia started at http://%s:5080/ganglia" % master -def is_ssh_available(host, opts): +def is_ssh_available(host, opts, print_ssh_output=True): """ Check if SSH is available on a host. """ - try: - with open(os.devnull, 'w') as devnull: - ret = subprocess.check_call( - ssh_command(opts) + ['-t', '-t', '-o', 'ConnectTimeout=3', - '%s@%s' % (opts.user, host), stringify_command('true')], - stdout=devnull, - stderr=devnull - ) - return ret == 0 - except subprocess.CalledProcessError as e: - return False + s = subprocess.Popen( + ssh_command(opts) + ['-t', '-t', '-o', 'ConnectTimeout=3', + '%s@%s' % (opts.user, host), stringify_command('true')], + stdout=subprocess.PIPE, + stderr=subprocess.STDOUT # we pipe stderr through stdout to preserve output order + ) + cmd_output = s.communicate()[0] # [1] is stderr, which we redirected to stdout + + if s.returncode != 0 and print_ssh_output: + # extra leading newline is for spacing in wait_for_cluster_state() + print textwrap.dedent("""\n + Warning: SSH connection error. (This could be temporary.) + Host: {h} + SSH return code: {r} + SSH output: {o} + """).format( + h=host, + r=s.returncode, + o=cmd_output.strip() + ) + + return s.returncode == 0 def is_cluster_ssh_available(cluster_instances, opts): @@ -896,6 +939,7 @@ def stringify_command(parts): def ssh_args(opts): parts = ['-o', 'StrictHostKeyChecking=no'] + parts += ['-o', 'UserKnownHostsFile=/dev/null'] if opts.identity_file is not None: parts += ['-i', opts.identity_file] return parts @@ -1003,10 +1047,57 @@ def real_main(): DeprecationWarning ) + if opts.identity_file is not None: + if not os.path.exists(opts.identity_file): + print >> stderr,\ + "ERROR: The identity file '{f}' doesn't exist.".format(f=opts.identity_file) + sys.exit(1) + + file_mode = os.stat(opts.identity_file).st_mode + if not (file_mode & S_IRUSR) or not oct(file_mode)[-2:] == '00': + print >> stderr, "ERROR: The identity file must be accessible only by you." + print >> stderr, 'You can fix this with: chmod 400 "{f}"'.format(f=opts.identity_file) + sys.exit(1) + + if opts.instance_type not in EC2_INSTANCE_TYPES: + print >> stderr, "Warning: Unrecognized EC2 instance type for instance-type: {t}".format( + t=opts.instance_type) + + if opts.master_instance_type != "": + if opts.master_instance_type not in EC2_INSTANCE_TYPES: + print >> stderr, \ + "Warning: Unrecognized EC2 instance type for master-instance-type: {t}".format( + t=opts.master_instance_type) + # Since we try instance types even if we can't resolve them, we check if they resolve first + # and, if they do, see if they resolve to the same virtualization type. + if opts.instance_type in EC2_INSTANCE_TYPES and \ + opts.master_instance_type in EC2_INSTANCE_TYPES: + if EC2_INSTANCE_TYPES[opts.instance_type] != \ + EC2_INSTANCE_TYPES[opts.master_instance_type]: + print >> stderr, \ + "Error: spark-ec2 currently does not support having a master and slaves with " + \ + "different AMI virtualization types." + print >> stderr, "master instance virtualization type: {t}".format( + t=EC2_INSTANCE_TYPES[opts.master_instance_type]) + print >> stderr, "slave instance virtualization type: {t}".format( + t=EC2_INSTANCE_TYPES[opts.instance_type]) + sys.exit(1) + if opts.ebs_vol_num > 8: print >> stderr, "ebs-vol-num cannot be greater than 8" sys.exit(1) + # Prevent breaking ami_prefix (/, .git and startswith checks) + # Prevent forks with non spark-ec2 names for now. + if opts.spark_ec2_git_repo.endswith("/") or \ + opts.spark_ec2_git_repo.endswith(".git") or \ + not opts.spark_ec2_git_repo.startswith("https://github.com") or \ + not opts.spark_ec2_git_repo.endswith("spark-ec2"): + print >> stderr, "spark-ec2-git-repo must be a github repo and it must not have a " \ + "trailing / or .git. " \ + "Furthermore, we currently only support forks named spark-ec2." + sys.exit(1) + try: conn = ec2.connect_to_region(opts.region) except Exception as e: diff --git a/examples/scala-2.10/src/main/java/org/apache/spark/examples/streaming/JavaDirectKafkaWordCount.java b/examples/scala-2.10/src/main/java/org/apache/spark/examples/streaming/JavaDirectKafkaWordCount.java new file mode 100644 index 0000000000000..bab9f2478e779 --- /dev/null +++ b/examples/scala-2.10/src/main/java/org/apache/spark/examples/streaming/JavaDirectKafkaWordCount.java @@ -0,0 +1,113 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.examples.streaming; + +import java.util.HashMap; +import java.util.HashSet; +import java.util.Arrays; +import java.util.regex.Pattern; + +import scala.Tuple2; + +import com.google.common.collect.Lists; +import kafka.serializer.StringDecoder; + +import org.apache.spark.SparkConf; +import org.apache.spark.api.java.function.*; +import org.apache.spark.streaming.api.java.*; +import org.apache.spark.streaming.kafka.KafkaUtils; +import org.apache.spark.streaming.Durations; + +/** + * Consumes messages from one or more topics in Kafka and does wordcount. + * Usage: DirectKafkaWordCount + * is a list of one or more Kafka brokers + * is a list of one or more kafka topics to consume from + * + * Example: + * $ bin/run-example streaming.KafkaWordCount broker1-host:port,broker2-host:port topic1,topic2 + */ + +public final class JavaDirectKafkaWordCount { + private static final Pattern SPACE = Pattern.compile(" "); + + public static void main(String[] args) { + if (args.length < 2) { + System.err.println("Usage: DirectKafkaWordCount \n" + + " is a list of one or more Kafka brokers\n" + + " is a list of one or more kafka topics to consume from\n\n"); + System.exit(1); + } + + StreamingExamples.setStreamingLogLevels(); + + String brokers = args[0]; + String topics = args[1]; + + // Create context with 2 second batch interval + SparkConf sparkConf = new SparkConf().setAppName("JavaDirectKafkaWordCount"); + JavaStreamingContext jssc = new JavaStreamingContext(sparkConf, Durations.seconds(2)); + + HashSet topicsSet = new HashSet(Arrays.asList(topics.split(","))); + HashMap kafkaParams = new HashMap(); + kafkaParams.put("metadata.broker.list", brokers); + + // Create direct kafka stream with brokers and topics + JavaPairInputDStream messages = KafkaUtils.createDirectStream( + jssc, + String.class, + String.class, + StringDecoder.class, + StringDecoder.class, + kafkaParams, + topicsSet + ); + + // Get the lines, split them into words, count the words and print + JavaDStream lines = messages.map(new Function, String>() { + @Override + public String call(Tuple2 tuple2) { + return tuple2._2(); + } + }); + JavaDStream words = lines.flatMap(new FlatMapFunction() { + @Override + public Iterable call(String x) { + return Lists.newArrayList(SPACE.split(x)); + } + }); + JavaPairDStream wordCounts = words.mapToPair( + new PairFunction() { + @Override + public Tuple2 call(String s) { + return new Tuple2(s, 1); + } + }).reduceByKey( + new Function2() { + @Override + public Integer call(Integer i1, Integer i2) { + return i1 + i2; + } + }); + wordCounts.print(); + + // Start the computation + jssc.start(); + jssc.awaitTermination(); + } +} diff --git a/examples/scala-2.10/src/main/scala/org/apache/spark/examples/streaming/DirectKafkaWordCount.scala b/examples/scala-2.10/src/main/scala/org/apache/spark/examples/streaming/DirectKafkaWordCount.scala new file mode 100644 index 0000000000000..deb08fd57b8c7 --- /dev/null +++ b/examples/scala-2.10/src/main/scala/org/apache/spark/examples/streaming/DirectKafkaWordCount.scala @@ -0,0 +1,71 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.examples.streaming + +import kafka.serializer.StringDecoder + +import org.apache.spark.streaming._ +import org.apache.spark.streaming.kafka._ +import org.apache.spark.SparkConf + +/** + * Consumes messages from one or more topics in Kafka and does wordcount. + * Usage: DirectKafkaWordCount + * is a list of one or more Kafka brokers + * is a list of one or more kafka topics to consume from + * + * Example: + * $ bin/run-example streaming.KafkaWordCount broker1-host:port,broker2-host:port topic1,topic2 + */ +object DirectKafkaWordCount { + def main(args: Array[String]) { + if (args.length < 2) { + System.err.println(s""" + |Usage: DirectKafkaWordCount + | is a list of one or more Kafka brokers + | is a list of one or more kafka topics to consume from + | + """".stripMargin) + System.exit(1) + } + + StreamingExamples.setStreamingLogLevels() + + val Array(brokers, topics) = args + + // Create context with 2 second batch interval + val sparkConf = new SparkConf().setAppName("DirectKafkaWordCount") + val ssc = new StreamingContext(sparkConf, Seconds(2)) + + // Create direct kafka stream with brokers and topics + val topicsSet = topics.split(",").toSet + val kafkaParams = Map[String, String]("metadata.broker.list" -> brokers) + val messages = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder]( + ssc, kafkaParams, topicsSet) + + // Get the lines, split them into words, count the words and print + val lines = messages.map(_._2) + val words = lines.flatMap(_.split(" ")) + val wordCounts = words.map(x => (x, 1L)).reduceByKey(_ + _) + wordCounts.print() + + // Start the computation + ssc.start() + ssc.awaitTermination() + } +} diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java index 42d4d7d0bef26..4d9dad9f23038 100644 --- a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java +++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java @@ -116,7 +116,7 @@ class MyJavaLogisticRegression */ IntParam maxIter = new IntParam(this, "maxIter", "max number of iterations"); - int getMaxIter() { return (int)get(maxIter); } + int getMaxIter() { return (Integer) get(maxIter); } public MyJavaLogisticRegression() { setMaxIter(100); @@ -124,7 +124,7 @@ public MyJavaLogisticRegression() { // The parameter setter is in this class since it should return type MyJavaLogisticRegression. MyJavaLogisticRegression setMaxIter(int value) { - return (MyJavaLogisticRegression)set(maxIter, value); + return (MyJavaLogisticRegression) set(maxIter, value); } // This method is used by fit(). diff --git a/examples/src/main/java/org/apache/spark/examples/mllib/JavaLDAExample.java b/examples/src/main/java/org/apache/spark/examples/mllib/JavaLDAExample.java new file mode 100644 index 0000000000000..36207ae38d9a9 --- /dev/null +++ b/examples/src/main/java/org/apache/spark/examples/mllib/JavaLDAExample.java @@ -0,0 +1,76 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.examples.mllib; + +import scala.Tuple2; + +import org.apache.spark.api.java.*; +import org.apache.spark.api.java.function.Function; +import org.apache.spark.mllib.clustering.DistributedLDAModel; +import org.apache.spark.mllib.clustering.LDA; +import org.apache.spark.mllib.linalg.Matrix; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.SparkConf; + +public class JavaLDAExample { + public static void main(String[] args) { + SparkConf conf = new SparkConf().setAppName("LDA Example"); + JavaSparkContext sc = new JavaSparkContext(conf); + + // Load and parse the data + String path = "data/mllib/sample_lda_data.txt"; + JavaRDD data = sc.textFile(path); + JavaRDD parsedData = data.map( + new Function() { + public Vector call(String s) { + String[] sarray = s.trim().split(" "); + double[] values = new double[sarray.length]; + for (int i = 0; i < sarray.length; i++) + values[i] = Double.parseDouble(sarray[i]); + return Vectors.dense(values); + } + } + ); + // Index documents with unique IDs + JavaPairRDD corpus = JavaPairRDD.fromJavaRDD(parsedData.zipWithIndex().map( + new Function, Tuple2>() { + public Tuple2 call(Tuple2 doc_id) { + return doc_id.swap(); + } + } + )); + corpus.cache(); + + // Cluster the documents into three topics using LDA + DistributedLDAModel ldaModel = new LDA().setK(3).run(corpus); + + // Output topics. Each is a distribution over words (matching word count vectors) + System.out.println("Learned topics (as distributions over vocab of " + ldaModel.vocabSize() + + " words):"); + Matrix topics = ldaModel.topicsMatrix(); + for (int topic = 0; topic < 3; topic++) { + System.out.print("Topic " + topic + ":"); + for (int word = 0; word < ldaModel.vocabSize(); word++) { + System.out.print(" " + topics.apply(word, topic)); + } + System.out.println(); + } + sc.stop(); + } +} diff --git a/examples/src/main/scala/org/apache/spark/examples/mllib/LDAExample.scala b/examples/src/main/scala/org/apache/spark/examples/mllib/LDAExample.scala index 0e1b27a8bd2ee..11399a7633638 100644 --- a/examples/src/main/scala/org/apache/spark/examples/mllib/LDAExample.scala +++ b/examples/src/main/scala/org/apache/spark/examples/mllib/LDAExample.scala @@ -159,7 +159,7 @@ object LDAExample { } println() } - + sc.stop() } /** diff --git a/external/flume/src/test/scala/org/apache/spark/streaming/flume/FlumeStreamSuite.scala b/external/flume/src/test/scala/org/apache/spark/streaming/flume/FlumeStreamSuite.scala index f333e3891b5f0..322de7bf2fed8 100644 --- a/external/flume/src/test/scala/org/apache/spark/streaming/flume/FlumeStreamSuite.scala +++ b/external/flume/src/test/scala/org/apache/spark/streaming/flume/FlumeStreamSuite.scala @@ -19,13 +19,13 @@ package org.apache.spark.streaming.flume import java.net.{InetSocketAddress, ServerSocket} import java.nio.ByteBuffer -import java.nio.charset.Charset import scala.collection.JavaConversions._ import scala.collection.mutable.{ArrayBuffer, SynchronizedBuffer} import scala.concurrent.duration._ import scala.language.postfixOps +import com.google.common.base.Charsets import org.apache.avro.ipc.NettyTransceiver import org.apache.avro.ipc.specific.SpecificRequestor import org.apache.flume.source.avro @@ -108,7 +108,7 @@ class FlumeStreamSuite extends FunSuite with BeforeAndAfter with Matchers with L val inputEvents = input.map { item => val event = new AvroFlumeEvent - event.setBody(ByteBuffer.wrap(item.getBytes("UTF-8"))) + event.setBody(ByteBuffer.wrap(item.getBytes(Charsets.UTF_8))) event.setHeaders(Map[CharSequence, CharSequence]("test" -> "header")) event } @@ -138,14 +138,13 @@ class FlumeStreamSuite extends FunSuite with BeforeAndAfter with Matchers with L status should be (avro.Status.OK) } - val decoder = Charset.forName("UTF-8").newDecoder() eventually(timeout(10 seconds), interval(100 milliseconds)) { val outputEvents = outputBuffer.flatten.map { _.event } outputEvents.foreach { event => event.getHeaders.get("test") should be("header") } - val output = outputEvents.map(event => decoder.decode(event.getBody()).toString) + val output = outputEvents.map(event => new String(event.getBody.array(), Charsets.UTF_8)) output should be (input) } } diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/DirectKafkaInputDStream.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/DirectKafkaInputDStream.scala index c7bca43eb889d..04e65cb3d708c 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/DirectKafkaInputDStream.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/DirectKafkaInputDStream.scala @@ -50,14 +50,13 @@ import org.apache.spark.streaming.dstream._ * @param fromOffsets per-topic/partition Kafka offsets defining the (inclusive) * starting point of the stream * @param messageHandler function for translating each message into the desired type - * @param maxRetries maximum number of times in a row to retry getting leaders' offsets */ private[streaming] class DirectKafkaInputDStream[ K: ClassTag, V: ClassTag, - U <: Decoder[_]: ClassTag, - T <: Decoder[_]: ClassTag, + U <: Decoder[K]: ClassTag, + T <: Decoder[V]: ClassTag, R: ClassTag]( @transient ssc_ : StreamingContext, val kafkaParams: Map[String, String], diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaCluster.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaCluster.scala index ccc62bfe8f057..2f7e0ab39fefd 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaCluster.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaCluster.scala @@ -332,6 +332,9 @@ object KafkaCluster { extends ConsumerConfig(originalProps) { val seedBrokers: Array[(String, Int)] = brokers.split(",").map { hp => val hpa = hp.split(":") + if (hpa.size == 1) { + throw new SparkException(s"Broker not the in correct format of : [$brokers]") + } (hpa(0), hpa(1).toInt) } } diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDD.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDD.scala index 50bf7cbdb8dbf..d56cc01be9514 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDD.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDD.scala @@ -36,14 +36,12 @@ import kafka.utils.VerifiableProperties * Starting and ending offsets are specified in advance, * so that you can control exactly-once semantics. * @param kafkaParams Kafka - * configuration parameters. - * Requires "metadata.broker.list" or "bootstrap.servers" to be set with Kafka broker(s), - * NOT zookeeper servers, specified in host1:port1,host2:port2 form. - * @param batch Each KafkaRDDPartition in the batch corresponds to a - * range of offsets for a given Kafka topic/partition + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" to be set + * with Kafka broker(s) specified in host1:port1,host2:port2 form. + * @param offsetRanges offset ranges that define the Kafka data belonging to this RDD * @param messageHandler function for translating each message into the desired type */ -private[spark] +private[kafka] class KafkaRDD[ K: ClassTag, V: ClassTag, @@ -183,7 +181,7 @@ class KafkaRDD[ } } -private[spark] +private[kafka] object KafkaRDD { import KafkaCluster.LeaderOffset diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala index 36372e08f65f6..a842a6f17766f 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala @@ -26,7 +26,7 @@ import org.apache.spark.Partition * @param host preferred kafka host, i.e. the leader at the time the rdd was created * @param port preferred kafka host's port */ -private[spark] +private[kafka] class KafkaRDDPartition( val index: Int, val topic: String, @@ -36,24 +36,3 @@ class KafkaRDDPartition( val host: String, val port: Int ) extends Partition - -private[spark] -object KafkaRDDPartition { - def apply( - index: Int, - topic: String, - partition: Int, - fromOffset: Long, - untilOffset: Long, - host: String, - port: Int - ): KafkaRDDPartition = new KafkaRDDPartition( - index, - topic, - partition, - fromOffset, - untilOffset, - host, - port - ) -} diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaUtils.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaUtils.scala index f8aa6c5c6263c..7a2c3abdcc24b 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaUtils.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaUtils.scala @@ -18,7 +18,9 @@ package org.apache.spark.streaming.kafka import java.lang.{Integer => JInt} +import java.lang.{Long => JLong} import java.util.{Map => JMap} +import java.util.{Set => JSet} import scala.reflect.ClassTag import scala.collection.JavaConversions._ @@ -27,18 +29,19 @@ import kafka.common.TopicAndPartition import kafka.message.MessageAndMetadata import kafka.serializer.{Decoder, StringDecoder} - +import org.apache.spark.api.java.function.{Function => JFunction} import org.apache.spark.{SparkContext, SparkException} import org.apache.spark.annotation.Experimental import org.apache.spark.rdd.RDD import org.apache.spark.storage.StorageLevel import org.apache.spark.streaming.StreamingContext -import org.apache.spark.streaming.api.java.{JavaPairReceiverInputDStream, JavaStreamingContext} +import org.apache.spark.streaming.api.java.{JavaPairInputDStream, JavaInputDStream, JavaPairReceiverInputDStream, JavaStreamingContext} import org.apache.spark.streaming.dstream.{InputDStream, ReceiverInputDStream} +import org.apache.spark.api.java.{JavaSparkContext, JavaPairRDD, JavaRDD} object KafkaUtils { /** - * Create an input stream that pulls messages from a Kafka Broker. + * Create an input stream that pulls messages from Kafka Brokers. * @param ssc StreamingContext object * @param zkQuorum Zookeeper quorum (hostname:port,hostname:port,..) * @param groupId The group id for this consumer @@ -62,7 +65,7 @@ object KafkaUtils { } /** - * Create an input stream that pulls messages from a Kafka Broker. + * Create an input stream that pulls messages from Kafka Brokers. * @param ssc StreamingContext object * @param kafkaParams Map of kafka configuration parameters, * see http://kafka.apache.org/08/configuration.html @@ -81,7 +84,7 @@ object KafkaUtils { } /** - * Create an input stream that pulls messages from a Kafka Broker. + * Create an input stream that pulls messages from Kafka Brokers. * Storage level of the data will be the default StorageLevel.MEMORY_AND_DISK_SER_2. * @param jssc JavaStreamingContext object * @param zkQuorum Zookeeper quorum (hostname:port,hostname:port,..) @@ -99,7 +102,7 @@ object KafkaUtils { } /** - * Create an input stream that pulls messages from a Kafka Broker. + * Create an input stream that pulls messages from Kafka Brokers. * @param jssc JavaStreamingContext object * @param zkQuorum Zookeeper quorum (hostname:port,hostname:port,..). * @param groupId The group id for this consumer. @@ -119,10 +122,10 @@ object KafkaUtils { } /** - * Create an input stream that pulls messages from a Kafka Broker. + * Create an input stream that pulls messages from Kafka Brokers. * @param jssc JavaStreamingContext object - * @param keyTypeClass Key type of RDD - * @param valueTypeClass value type of RDD + * @param keyTypeClass Key type of DStream + * @param valueTypeClass value type of Dstream * @param keyDecoderClass Type of kafka key decoder * @param valueDecoderClass Type of kafka value decoder * @param kafkaParams Map of kafka configuration parameters, @@ -151,14 +154,14 @@ object KafkaUtils { jssc.ssc, kafkaParams.toMap, Map(topics.mapValues(_.intValue()).toSeq: _*), storageLevel) } - /** A batch-oriented interface for consuming from Kafka. - * Starting and ending offsets are specified in advance, - * so that you can control exactly-once semantics. + /** + * Create a RDD from Kafka using offset ranges for each topic and partition. + * * @param sc SparkContext object * @param kafkaParams Kafka - * configuration parameters. - * Requires "metadata.broker.list" or "bootstrap.servers" to be set with Kafka broker(s), - * NOT zookeeper servers, specified in host1:port1,host2:port2 form. + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers) specified in + * host1:port1,host2:port2 form. * @param offsetRanges Each OffsetRange in the batch corresponds to a * range of offsets for a given Kafka topic/partition */ @@ -166,12 +169,12 @@ object KafkaUtils { def createRDD[ K: ClassTag, V: ClassTag, - U <: Decoder[_]: ClassTag, - T <: Decoder[_]: ClassTag] ( + KD <: Decoder[K]: ClassTag, + VD <: Decoder[V]: ClassTag]( sc: SparkContext, kafkaParams: Map[String, String], offsetRanges: Array[OffsetRange] - ): RDD[(K, V)] = { + ): RDD[(K, V)] = { val messageHandler = (mmd: MessageAndMetadata[K, V]) => (mmd.key, mmd.message) val kc = new KafkaCluster(kafkaParams) val topics = offsetRanges.map(o => TopicAndPartition(o.topic, o.partition)).toSet @@ -179,121 +182,196 @@ object KafkaUtils { errs => throw new SparkException(errs.mkString("\n")), ok => ok ) - new KafkaRDD[K, V, U, T, (K, V)](sc, kafkaParams, offsetRanges, leaders, messageHandler) + new KafkaRDD[K, V, KD, VD, (K, V)](sc, kafkaParams, offsetRanges, leaders, messageHandler) } - /** A batch-oriented interface for consuming from Kafka. - * Starting and ending offsets are specified in advance, - * so that you can control exactly-once semantics. + /** + * :: Experimental :: + * Create a RDD from Kafka using offset ranges for each topic and partition. This allows you + * specify the Kafka leader to connect to (to optimize fetching) and access the message as well + * as the metadata. + * * @param sc SparkContext object * @param kafkaParams Kafka - * configuration parameters. - * Requires "metadata.broker.list" or "bootstrap.servers" to be set with Kafka broker(s), - * NOT zookeeper servers, specified in host1:port1,host2:port2 form. + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers) specified in + * host1:port1,host2:port2 form. * @param offsetRanges Each OffsetRange in the batch corresponds to a * range of offsets for a given Kafka topic/partition * @param leaders Kafka leaders for each offset range in batch - * @param messageHandler function for translating each message into the desired type + * @param messageHandler Function for translating each message and metadata into the desired type */ @Experimental def createRDD[ K: ClassTag, V: ClassTag, - U <: Decoder[_]: ClassTag, - T <: Decoder[_]: ClassTag, - R: ClassTag] ( + KD <: Decoder[K]: ClassTag, + VD <: Decoder[V]: ClassTag, + R: ClassTag]( sc: SparkContext, kafkaParams: Map[String, String], offsetRanges: Array[OffsetRange], leaders: Array[Leader], messageHandler: MessageAndMetadata[K, V] => R - ): RDD[R] = { - + ): RDD[R] = { val leaderMap = leaders .map(l => TopicAndPartition(l.topic, l.partition) -> (l.host, l.port)) .toMap - new KafkaRDD[K, V, U, T, R](sc, kafkaParams, offsetRanges, leaderMap, messageHandler) + new KafkaRDD[K, V, KD, VD, R](sc, kafkaParams, offsetRanges, leaderMap, messageHandler) } + /** - * This stream can guarantee that each message from Kafka is included in transformations - * (as opposed to output actions) exactly once, even in most failure situations. + * Create a RDD from Kafka using offset ranges for each topic and partition. * - * Points to note: - * - * Failure Recovery - You must checkpoint this stream, or save offsets yourself and provide them - * as the fromOffsets parameter on restart. - * Kafka must have sufficient log retention to obtain messages after failure. - * - * Getting offsets from the stream - see programming guide + * @param jsc JavaSparkContext object + * @param kafkaParams Kafka + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers) specified in + * host1:port1,host2:port2 form. + * @param offsetRanges Each OffsetRange in the batch corresponds to a + * range of offsets for a given Kafka topic/partition + */ + @Experimental + def createRDD[K, V, KD <: Decoder[K], VD <: Decoder[V]]( + jsc: JavaSparkContext, + keyClass: Class[K], + valueClass: Class[V], + keyDecoderClass: Class[KD], + valueDecoderClass: Class[VD], + kafkaParams: JMap[String, String], + offsetRanges: Array[OffsetRange] + ): JavaPairRDD[K, V] = { + implicit val keyCmt: ClassTag[K] = ClassTag(keyClass) + implicit val valueCmt: ClassTag[V] = ClassTag(valueClass) + implicit val keyDecoderCmt: ClassTag[KD] = ClassTag(keyDecoderClass) + implicit val valueDecoderCmt: ClassTag[VD] = ClassTag(valueDecoderClass) + new JavaPairRDD(createRDD[K, V, KD, VD]( + jsc.sc, Map(kafkaParams.toSeq: _*), offsetRanges)) + } + + /** + * :: Experimental :: + * Create a RDD from Kafka using offset ranges for each topic and partition. This allows you + * specify the Kafka leader to connect to (to optimize fetching) and access the message as well + * as the metadata. * -. * Zookeeper - This does not use Zookeeper to store offsets. For interop with Kafka monitors - * that depend on Zookeeper, you must store offsets in ZK yourself. + * @param jsc JavaSparkContext object + * @param kafkaParams Kafka + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers) specified in + * host1:port1,host2:port2 form. + * @param offsetRanges Each OffsetRange in the batch corresponds to a + * range of offsets for a given Kafka topic/partition + * @param leaders Kafka leaders for each offset range in batch + * @param messageHandler Function for translating each message and metadata into the desired type + */ + @Experimental + def createRDD[K, V, KD <: Decoder[K], VD <: Decoder[V], R]( + jsc: JavaSparkContext, + keyClass: Class[K], + valueClass: Class[V], + keyDecoderClass: Class[KD], + valueDecoderClass: Class[VD], + recordClass: Class[R], + kafkaParams: JMap[String, String], + offsetRanges: Array[OffsetRange], + leaders: Array[Leader], + messageHandler: JFunction[MessageAndMetadata[K, V], R] + ): JavaRDD[R] = { + implicit val keyCmt: ClassTag[K] = ClassTag(keyClass) + implicit val valueCmt: ClassTag[V] = ClassTag(valueClass) + implicit val keyDecoderCmt: ClassTag[KD] = ClassTag(keyDecoderClass) + implicit val valueDecoderCmt: ClassTag[VD] = ClassTag(valueDecoderClass) + implicit val recordCmt: ClassTag[R] = ClassTag(recordClass) + createRDD[K, V, KD, VD, R]( + jsc.sc, Map(kafkaParams.toSeq: _*), offsetRanges, leaders, messageHandler.call _) + } + + /** + * :: Experimental :: + * Create an input stream that directly pulls messages from Kafka Brokers + * without using any receiver. This stream can guarantee that each message + * from Kafka is included in transformations exactly once (see points below). * - * End-to-end semantics - This does not guarantee that any output operation will push each record - * exactly once. To ensure end-to-end exactly-once semantics (that is, receiving exactly once and - * outputting exactly once), you have to either ensure that the output operation is - * idempotent, or transactionally store offsets with the output. See the programming guide for - * more details. + * Points to note: + * - No receivers: This stream does not use any receiver. It directly queries Kafka + * - Offsets: This does not use Zookeeper to store offsets. The consumed offsets are tracked + * by the stream itself. For interoperability with Kafka monitoring tools that depend on + * Zookeeper, you have to update Kafka/Zookeeper yourself from the streaming application. + * You can access the offsets used in each batch from the generated RDDs (see + * [[org.apache.spark.streaming.kafka.HasOffsetRanges]]). + * - Failure Recovery: To recover from driver failures, you have to enable checkpointing + * in the [[StreamingContext]]. The information on consumed offset can be + * recovered from the checkpoint. See the programming guide for details (constraints, etc.). + * - End-to-end semantics: This stream ensures that every records is effectively received and + * transformed exactly once, but gives no guarantees on whether the transformed data are + * outputted exactly once. For end-to-end exactly-once semantics, you have to either ensure + * that the output operation is idempotent, or use transactions to output records atomically. + * See the programming guide for more details. * * @param ssc StreamingContext object * @param kafkaParams Kafka - * configuration parameters. - * Requires "metadata.broker.list" or "bootstrap.servers" to be set with Kafka broker(s), - * NOT zookeeper servers, specified in host1:port1,host2:port2 form. - * @param messageHandler function for translating each message into the desired type - * @param fromOffsets per-topic/partition Kafka offsets defining the (inclusive) - * starting point of the stream + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers) specified in + * host1:port1,host2:port2 form. + * @param fromOffsets Per-topic/partition Kafka offsets defining the (inclusive) + * starting point of the stream + * @param messageHandler Function for translating each message and metadata into the desired type */ @Experimental def createDirectStream[ K: ClassTag, V: ClassTag, - U <: Decoder[_]: ClassTag, - T <: Decoder[_]: ClassTag, + KD <: Decoder[K]: ClassTag, + VD <: Decoder[V]: ClassTag, R: ClassTag] ( ssc: StreamingContext, kafkaParams: Map[String, String], fromOffsets: Map[TopicAndPartition, Long], messageHandler: MessageAndMetadata[K, V] => R ): InputDStream[R] = { - new DirectKafkaInputDStream[K, V, U, T, R]( + new DirectKafkaInputDStream[K, V, KD, VD, R]( ssc, kafkaParams, fromOffsets, messageHandler) } /** - * This stream can guarantee that each message from Kafka is included in transformations - * (as opposed to output actions) exactly once, even in most failure situations. + * :: Experimental :: + * Create an input stream that directly pulls messages from Kafka Brokers + * without using any receiver. This stream can guarantee that each message + * from Kafka is included in transformations exactly once (see points below). * * Points to note: - * - * Failure Recovery - You must checkpoint this stream. - * Kafka must have sufficient log retention to obtain messages after failure. - * - * Getting offsets from the stream - see programming guide - * -. * Zookeeper - This does not use Zookeeper to store offsets. For interop with Kafka monitors - * that depend on Zookeeper, you must store offsets in ZK yourself. - * - * End-to-end semantics - This does not guarantee that any output operation will push each record - * exactly once. To ensure end-to-end exactly-once semantics (that is, receiving exactly once and - * outputting exactly once), you have to ensure that the output operation is idempotent. + * - No receivers: This stream does not use any receiver. It directly queries Kafka + * - Offsets: This does not use Zookeeper to store offsets. The consumed offsets are tracked + * by the stream itself. For interoperability with Kafka monitoring tools that depend on + * Zookeeper, you have to update Kafka/Zookeeper yourself from the streaming application. + * You can access the offsets used in each batch from the generated RDDs (see + * [[org.apache.spark.streaming.kafka.HasOffsetRanges]]). + * - Failure Recovery: To recover from driver failures, you have to enable checkpointing + * in the [[StreamingContext]]. The information on consumed offset can be + * recovered from the checkpoint. See the programming guide for details (constraints, etc.). + * - End-to-end semantics: This stream ensures that every records is effectively received and + * transformed exactly once, but gives no guarantees on whether the transformed data are + * outputted exactly once. For end-to-end exactly-once semantics, you have to either ensure + * that the output operation is idempotent, or use transactions to output records atomically. + * See the programming guide for more details. * * @param ssc StreamingContext object * @param kafkaParams Kafka - * configuration parameters. - * Requires "metadata.broker.list" or "bootstrap.servers" to be set with Kafka broker(s), - * NOT zookeeper servers, specified in host1:port1,host2:port2 form. - * If starting without a checkpoint, "auto.offset.reset" may be set to "largest" or "smallest" + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers), specified in + * host1:port1,host2:port2 form. + * If not starting from a checkpoint, "auto.offset.reset" may be set to "largest" or "smallest" * to determine where the stream starts (defaults to "largest") - * @param topics names of the topics to consume + * @param topics Names of the topics to consume */ @Experimental def createDirectStream[ K: ClassTag, V: ClassTag, - U <: Decoder[_]: ClassTag, - T <: Decoder[_]: ClassTag] ( + KD <: Decoder[K]: ClassTag, + VD <: Decoder[V]: ClassTag] ( ssc: StreamingContext, kafkaParams: Map[String, String], topics: Set[String] @@ -313,11 +391,128 @@ object KafkaUtils { val fromOffsets = leaderOffsets.map { case (tp, lo) => (tp, lo.offset) } - new DirectKafkaInputDStream[K, V, U, T, (K, V)]( + new DirectKafkaInputDStream[K, V, KD, VD, (K, V)]( ssc, kafkaParams, fromOffsets, messageHandler) }).fold( errs => throw new SparkException(errs.mkString("\n")), ok => ok ) } + + /** + * :: Experimental :: + * Create an input stream that directly pulls messages from Kafka Brokers + * without using any receiver. This stream can guarantee that each message + * from Kafka is included in transformations exactly once (see points below). + * + * Points to note: + * - No receivers: This stream does not use any receiver. It directly queries Kafka + * - Offsets: This does not use Zookeeper to store offsets. The consumed offsets are tracked + * by the stream itself. For interoperability with Kafka monitoring tools that depend on + * Zookeeper, you have to update Kafka/Zookeeper yourself from the streaming application. + * You can access the offsets used in each batch from the generated RDDs (see + * [[org.apache.spark.streaming.kafka.HasOffsetRanges]]). + * - Failure Recovery: To recover from driver failures, you have to enable checkpointing + * in the [[StreamingContext]]. The information on consumed offset can be + * recovered from the checkpoint. See the programming guide for details (constraints, etc.). + * - End-to-end semantics: This stream ensures that every records is effectively received and + * transformed exactly once, but gives no guarantees on whether the transformed data are + * outputted exactly once. For end-to-end exactly-once semantics, you have to either ensure + * that the output operation is idempotent, or use transactions to output records atomically. + * See the programming guide for more details. + * + * @param jssc JavaStreamingContext object + * @param keyClass Class of the keys in the Kafka records + * @param valueClass Class of the values in the Kafka records + * @param keyDecoderClass Class of the key decoder + * @param valueDecoderClass Class of the value decoder + * @param recordClass Class of the records in DStream + * @param kafkaParams Kafka + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers), specified in + * host1:port1,host2:port2 form. + * @param fromOffsets Per-topic/partition Kafka offsets defining the (inclusive) + * starting point of the stream + * @param messageHandler Function for translating each message and metadata into the desired type + */ + @Experimental + def createDirectStream[K, V, KD <: Decoder[K], VD <: Decoder[V], R]( + jssc: JavaStreamingContext, + keyClass: Class[K], + valueClass: Class[V], + keyDecoderClass: Class[KD], + valueDecoderClass: Class[VD], + recordClass: Class[R], + kafkaParams: JMap[String, String], + fromOffsets: JMap[TopicAndPartition, JLong], + messageHandler: JFunction[MessageAndMetadata[K, V], R] + ): JavaInputDStream[R] = { + implicit val keyCmt: ClassTag[K] = ClassTag(keyClass) + implicit val valueCmt: ClassTag[V] = ClassTag(valueClass) + implicit val keyDecoderCmt: ClassTag[KD] = ClassTag(keyDecoderClass) + implicit val valueDecoderCmt: ClassTag[VD] = ClassTag(valueDecoderClass) + implicit val recordCmt: ClassTag[R] = ClassTag(recordClass) + createDirectStream[K, V, KD, VD, R]( + jssc.ssc, + Map(kafkaParams.toSeq: _*), + Map(fromOffsets.mapValues { _.longValue() }.toSeq: _*), + messageHandler.call _ + ) + } + + /** + * :: Experimental :: + * Create an input stream that directly pulls messages from Kafka Brokers + * without using any receiver. This stream can guarantee that each message + * from Kafka is included in transformations exactly once (see points below). + * + * Points to note: + * - No receivers: This stream does not use any receiver. It directly queries Kafka + * - Offsets: This does not use Zookeeper to store offsets. The consumed offsets are tracked + * by the stream itself. For interoperability with Kafka monitoring tools that depend on + * Zookeeper, you have to update Kafka/Zookeeper yourself from the streaming application. + * You can access the offsets used in each batch from the generated RDDs (see + * [[org.apache.spark.streaming.kafka.HasOffsetRanges]]). + * - Failure Recovery: To recover from driver failures, you have to enable checkpointing + * in the [[StreamingContext]]. The information on consumed offset can be + * recovered from the checkpoint. See the programming guide for details (constraints, etc.). + * - End-to-end semantics: This stream ensures that every records is effectively received and + * transformed exactly once, but gives no guarantees on whether the transformed data are + * outputted exactly once. For end-to-end exactly-once semantics, you have to either ensure + * that the output operation is idempotent, or use transactions to output records atomically. + * See the programming guide for more details. + * + * @param jssc JavaStreamingContext object + * @param keyClass Class of the keys in the Kafka records + * @param valueClass Class of the values in the Kafka records + * @param keyDecoderClass Class of the key decoder + * @param valueDecoderClass Class type of the value decoder + * @param kafkaParams Kafka + * configuration parameters. Requires "metadata.broker.list" or "bootstrap.servers" + * to be set with Kafka broker(s) (NOT zookeeper servers), specified in + * host1:port1,host2:port2 form. + * If not starting from a checkpoint, "auto.offset.reset" may be set to "largest" or "smallest" + * to determine where the stream starts (defaults to "largest") + * @param topics Names of the topics to consume + */ + @Experimental + def createDirectStream[K, V, KD <: Decoder[K], VD <: Decoder[V], R]( + jssc: JavaStreamingContext, + keyClass: Class[K], + valueClass: Class[V], + keyDecoderClass: Class[KD], + valueDecoderClass: Class[VD], + kafkaParams: JMap[String, String], + topics: JSet[String] + ): JavaPairInputDStream[K, V] = { + implicit val keyCmt: ClassTag[K] = ClassTag(keyClass) + implicit val valueCmt: ClassTag[V] = ClassTag(valueClass) + implicit val keyDecoderCmt: ClassTag[KD] = ClassTag(keyDecoderClass) + implicit val valueDecoderCmt: ClassTag[VD] = ClassTag(valueDecoderClass) + createDirectStream[K, V, KD, VD]( + jssc.ssc, + Map(kafkaParams.toSeq: _*), + Set(topics.toSeq: _*) + ) + } } diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/Leader.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/Leader.scala index 3454d92e72b47..c129a26836c0d 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/Leader.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/Leader.scala @@ -19,17 +19,28 @@ package org.apache.spark.streaming.kafka import kafka.common.TopicAndPartition -/** Host info for the leader of a Kafka TopicAndPartition */ +import org.apache.spark.annotation.Experimental + +/** + * :: Experimental :: + * Represent the host info for the leader of a Kafka partition. + */ +@Experimental final class Leader private( - /** kafka topic name */ + /** Kafka topic name */ val topic: String, - /** kafka partition id */ + /** Kafka partition id */ val partition: Int, - /** kafka hostname */ + /** Leader's hostname */ val host: String, - /** kafka host's port */ + /** Leader's port */ val port: Int) extends Serializable +/** + * :: Experimental :: + * Companion object the provides methods to create instances of [[Leader]]. + */ +@Experimental object Leader { def create(topic: String, partition: Int, host: String, port: Int): Leader = new Leader(topic, partition, host, port) diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/OffsetRange.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/OffsetRange.scala index 334c12e4627b4..9c3dfeb8f5928 100644 --- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/OffsetRange.scala +++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/OffsetRange.scala @@ -19,16 +19,35 @@ package org.apache.spark.streaming.kafka import kafka.common.TopicAndPartition -/** Something that has a collection of OffsetRanges */ +import org.apache.spark.annotation.Experimental + +/** + * :: Experimental :: + * Represents any object that has a collection of [[OffsetRange]]s. This can be used access the + * offset ranges in RDDs generated by the direct Kafka DStream (see + * [[KafkaUtils.createDirectStream()]]). + * {{{ + * KafkaUtils.createDirectStream(...).foreachRDD { rdd => + * val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges + * ... + * } + * }}} + */ +@Experimental trait HasOffsetRanges { def offsetRanges: Array[OffsetRange] } -/** Represents a range of offsets from a single Kafka TopicAndPartition */ +/** + * :: Experimental :: + * Represents a range of offsets from a single Kafka TopicAndPartition. Instances of this class + * can be created with `OffsetRange.create()`. + */ +@Experimental final class OffsetRange private( - /** kafka topic name */ + /** Kafka topic name */ val topic: String, - /** kafka partition id */ + /** Kafka partition id */ val partition: Int, /** inclusive starting offset */ val fromOffset: Long, @@ -36,11 +55,33 @@ final class OffsetRange private( val untilOffset: Long) extends Serializable { import OffsetRange.OffsetRangeTuple + override def equals(obj: Any): Boolean = obj match { + case that: OffsetRange => + this.topic == that.topic && + this.partition == that.partition && + this.fromOffset == that.fromOffset && + this.untilOffset == that.untilOffset + case _ => false + } + + override def hashCode(): Int = { + toTuple.hashCode() + } + + override def toString(): String = { + s"OffsetRange(topic: '$topic', partition: $partition, range: [$fromOffset -> $untilOffset]" + } + /** this is to avoid ClassNotFoundException during checkpoint restore */ private[streaming] def toTuple: OffsetRangeTuple = (topic, partition, fromOffset, untilOffset) } +/** + * :: Experimental :: + * Companion object the provides methods to create instances of [[OffsetRange]]. + */ +@Experimental object OffsetRange { def create(topic: String, partition: Int, fromOffset: Long, untilOffset: Long): OffsetRange = new OffsetRange(topic, partition, fromOffset, untilOffset) @@ -61,10 +102,10 @@ object OffsetRange { new OffsetRange(topicAndPartition.topic, topicAndPartition.partition, fromOffset, untilOffset) /** this is to avoid ClassNotFoundException during checkpoint restore */ - private[spark] + private[kafka] type OffsetRangeTuple = (String, Int, Long, Long) - private[streaming] + private[kafka] def apply(t: OffsetRangeTuple) = new OffsetRange(t._1, t._2, t._3, t._4) } diff --git a/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaDirectKafkaStreamSuite.java b/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaDirectKafkaStreamSuite.java new file mode 100644 index 0000000000000..1334cc8fd1b57 --- /dev/null +++ b/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaDirectKafkaStreamSuite.java @@ -0,0 +1,159 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.streaming.kafka; + +import java.io.Serializable; +import java.util.HashMap; +import java.util.HashSet; +import java.util.Random; +import java.util.Arrays; + +import org.apache.spark.SparkConf; + +import scala.Tuple2; + +import junit.framework.Assert; + +import kafka.common.TopicAndPartition; +import kafka.message.MessageAndMetadata; +import kafka.serializer.StringDecoder; + +import org.apache.spark.api.java.JavaRDD; +import org.apache.spark.api.java.function.Function; +import org.apache.spark.streaming.api.java.JavaDStream; +import org.apache.spark.streaming.Durations; +import org.apache.spark.streaming.api.java.JavaStreamingContext; + +import org.junit.Test; +import org.junit.After; +import org.junit.Before; + +public class JavaDirectKafkaStreamSuite implements Serializable { + private transient JavaStreamingContext ssc = null; + private transient Random random = new Random(); + private transient KafkaStreamSuiteBase suiteBase = null; + + @Before + public void setUp() { + suiteBase = new KafkaStreamSuiteBase() { }; + suiteBase.setupKafka(); + System.clearProperty("spark.driver.port"); + SparkConf sparkConf = new SparkConf() + .setMaster("local[4]").setAppName(this.getClass().getSimpleName()); + ssc = new JavaStreamingContext(sparkConf, Durations.milliseconds(200)); + } + + @After + public void tearDown() { + ssc.stop(); + ssc = null; + System.clearProperty("spark.driver.port"); + suiteBase.tearDownKafka(); + } + + @Test + public void testKafkaStream() throws InterruptedException { + String topic1 = "topic1"; + String topic2 = "topic2"; + + String[] topic1data = createTopicAndSendData(topic1); + String[] topic2data = createTopicAndSendData(topic2); + + HashSet sent = new HashSet(); + sent.addAll(Arrays.asList(topic1data)); + sent.addAll(Arrays.asList(topic2data)); + + HashMap kafkaParams = new HashMap(); + kafkaParams.put("metadata.broker.list", suiteBase.brokerAddress()); + kafkaParams.put("auto.offset.reset", "smallest"); + + JavaDStream stream1 = KafkaUtils.createDirectStream( + ssc, + String.class, + String.class, + StringDecoder.class, + StringDecoder.class, + kafkaParams, + topicToSet(topic1) + ).map( + new Function, String>() { + @Override + public String call(scala.Tuple2 kv) throws Exception { + return kv._2(); + } + } + ); + + JavaDStream stream2 = KafkaUtils.createDirectStream( + ssc, + String.class, + String.class, + StringDecoder.class, + StringDecoder.class, + String.class, + kafkaParams, + topicOffsetToMap(topic2, (long) 0), + new Function, String>() { + @Override + public String call(MessageAndMetadata msgAndMd) throws Exception { + return msgAndMd.message(); + } + } + ); + JavaDStream unifiedStream = stream1.union(stream2); + + final HashSet result = new HashSet(); + unifiedStream.foreachRDD( + new Function, Void>() { + @Override + public Void call(org.apache.spark.api.java.JavaRDD rdd) throws Exception { + result.addAll(rdd.collect()); + return null; + } + } + ); + ssc.start(); + long startTime = System.currentTimeMillis(); + boolean matches = false; + while (!matches && System.currentTimeMillis() - startTime < 20000) { + matches = sent.size() == result.size(); + Thread.sleep(50); + } + Assert.assertEquals(sent, result); + ssc.stop(); + } + + private HashSet topicToSet(String topic) { + HashSet topicSet = new HashSet(); + topicSet.add(topic); + return topicSet; + } + + private HashMap topicOffsetToMap(String topic, Long offsetToStart) { + HashMap topicMap = new HashMap(); + topicMap.put(new TopicAndPartition(topic, 0), offsetToStart); + return topicMap; + } + + private String[] createTopicAndSendData(String topic) { + String[] data = { topic + "-1", topic + "-2", topic + "-3"}; + suiteBase.createTopic(topic); + suiteBase.sendMessages(topic, data); + return data; + } +} diff --git a/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaKafkaStreamSuite.java b/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaKafkaStreamSuite.java index 6e1abf3f385ee..208cc51b29876 100644 --- a/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaKafkaStreamSuite.java +++ b/external/kafka/src/test/java/org/apache/spark/streaming/kafka/JavaKafkaStreamSuite.java @@ -79,9 +79,10 @@ public void testKafkaStream() throws InterruptedException { suiteBase.createTopic(topic); HashMap tmp = new HashMap(sent); - suiteBase.produceAndSendMessage(topic, + suiteBase.sendMessages(topic, JavaConverters.mapAsScalaMapConverter(tmp).asScala().toMap( - Predef.>conforms())); + Predef.>conforms()) + ); HashMap kafkaParams = new HashMap(); kafkaParams.put("zookeeper.connect", suiteBase.zkAddress()); diff --git a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/DirectKafkaStreamSuite.scala b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/DirectKafkaStreamSuite.scala new file mode 100644 index 0000000000000..b25c2120d54f7 --- /dev/null +++ b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/DirectKafkaStreamSuite.scala @@ -0,0 +1,302 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.streaming.kafka + +import java.io.File + +import scala.collection.mutable +import scala.concurrent.duration._ +import scala.language.postfixOps + +import kafka.serializer.StringDecoder +import org.scalatest.{BeforeAndAfter, BeforeAndAfterAll} +import org.scalatest.concurrent.{Eventually, Timeouts} + +import org.apache.spark.{SparkContext, SparkConf} +import org.apache.spark.rdd.RDD +import org.apache.spark.streaming.{Milliseconds, StreamingContext, Time} +import org.apache.spark.streaming.dstream.{DStream, InputDStream} +import org.apache.spark.util.Utils +import kafka.common.TopicAndPartition +import kafka.message.MessageAndMetadata + +class DirectKafkaStreamSuite extends KafkaStreamSuiteBase + with BeforeAndAfter with BeforeAndAfterAll with Eventually { + val sparkConf = new SparkConf() + .setMaster("local[4]") + .setAppName(this.getClass.getSimpleName) + + var sc: SparkContext = _ + var ssc: StreamingContext = _ + var testDir: File = _ + + override def beforeAll { + setupKafka() + } + + override def afterAll { + tearDownKafka() + } + + after { + if (ssc != null) { + ssc.stop() + sc = null + } + if (sc != null) { + sc.stop() + } + if (testDir != null) { + Utils.deleteRecursively(testDir) + } + } + + + test("basic stream receiving with multiple topics and smallest starting offset") { + val topics = Set("basic1", "basic2", "basic3") + val data = Map("a" -> 7, "b" -> 9) + topics.foreach { t => + createTopic(t) + sendMessages(t, data) + } + val kafkaParams = Map( + "metadata.broker.list" -> s"$brokerAddress", + "auto.offset.reset" -> "smallest" + ) + + ssc = new StreamingContext(sparkConf, Milliseconds(200)) + val stream = withClue("Error creating direct stream") { + KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder]( + ssc, kafkaParams, topics) + } + var total = 0L + + stream.foreachRDD { rdd => + // Get the offset ranges in the RDD + val offsets = rdd.asInstanceOf[HasOffsetRanges].offsetRanges + val collected = rdd.mapPartitionsWithIndex { (i, iter) => + // For each partition, get size of the range in the partition, + // and the number of items in the partition + val off = offsets(i) + val all = iter.toSeq + val partSize = all.size + val rangeSize = off.untilOffset - off.fromOffset + Iterator((partSize, rangeSize)) + }.collect + + // Verify whether number of elements in each partition + // matches with the corresponding offset range + collected.foreach { case (partSize, rangeSize) => + assert(partSize === rangeSize, "offset ranges are wrong") + } + total += collected.size // Add up all the collected items + } + ssc.start() + eventually(timeout(20000.milliseconds), interval(200.milliseconds)) { + assert(total === data.values.sum * topics.size, "didn't get all messages") + } + ssc.stop() + } + + test("receiving from largest starting offset") { + val topic = "largest" + val topicPartition = TopicAndPartition(topic, 0) + val data = Map("a" -> 10) + createTopic(topic) + val kafkaParams = Map( + "metadata.broker.list" -> s"$brokerAddress", + "auto.offset.reset" -> "largest" + ) + val kc = new KafkaCluster(kafkaParams) + def getLatestOffset(): Long = { + kc.getLatestLeaderOffsets(Set(topicPartition)).right.get(topicPartition).offset + } + + // Send some initial messages before starting context + sendMessages(topic, data) + eventually(timeout(10 seconds), interval(20 milliseconds)) { + assert(getLatestOffset() > 3) + } + val offsetBeforeStart = getLatestOffset() + + // Setup context and kafka stream with largest offset + ssc = new StreamingContext(sparkConf, Milliseconds(200)) + val stream = withClue("Error creating direct stream") { + KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder]( + ssc, kafkaParams, Set(topic)) + } + assert( + stream.asInstanceOf[DirectKafkaInputDStream[_, _, _, _, _]] + .fromOffsets(topicPartition) >= offsetBeforeStart, + "Start offset not from latest" + ) + + val collectedData = new mutable.ArrayBuffer[String]() + stream.map { _._2 }.foreachRDD { rdd => collectedData ++= rdd.collect() } + ssc.start() + val newData = Map("b" -> 10) + sendMessages(topic, newData) + eventually(timeout(10 seconds), interval(50 milliseconds)) { + collectedData.contains("b") + } + assert(!collectedData.contains("a")) + } + + + test("creating stream by offset") { + val topic = "offset" + val topicPartition = TopicAndPartition(topic, 0) + val data = Map("a" -> 10) + createTopic(topic) + val kafkaParams = Map( + "metadata.broker.list" -> s"$brokerAddress", + "auto.offset.reset" -> "largest" + ) + val kc = new KafkaCluster(kafkaParams) + def getLatestOffset(): Long = { + kc.getLatestLeaderOffsets(Set(topicPartition)).right.get(topicPartition).offset + } + + // Send some initial messages before starting context + sendMessages(topic, data) + eventually(timeout(10 seconds), interval(20 milliseconds)) { + assert(getLatestOffset() >= 10) + } + val offsetBeforeStart = getLatestOffset() + + // Setup context and kafka stream with largest offset + ssc = new StreamingContext(sparkConf, Milliseconds(200)) + val stream = withClue("Error creating direct stream") { + KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder, String]( + ssc, kafkaParams, Map(topicPartition -> 11L), + (m: MessageAndMetadata[String, String]) => m.message()) + } + assert( + stream.asInstanceOf[DirectKafkaInputDStream[_, _, _, _, _]] + .fromOffsets(topicPartition) >= offsetBeforeStart, + "Start offset not from latest" + ) + + val collectedData = new mutable.ArrayBuffer[String]() + stream.foreachRDD { rdd => collectedData ++= rdd.collect() } + ssc.start() + val newData = Map("b" -> 10) + sendMessages(topic, newData) + eventually(timeout(10 seconds), interval(50 milliseconds)) { + collectedData.contains("b") + } + assert(!collectedData.contains("a")) + } + + // Test to verify the offset ranges can be recovered from the checkpoints + test("offset recovery") { + val topic = "recovery" + createTopic(topic) + testDir = Utils.createTempDir() + + val kafkaParams = Map( + "metadata.broker.list" -> s"$brokerAddress", + "auto.offset.reset" -> "smallest" + ) + + // Send data to Kafka and wait for it to be received + def sendDataAndWaitForReceive(data: Seq[Int]) { + val strings = data.map { _.toString} + sendMessages(topic, strings.map { _ -> 1}.toMap) + eventually(timeout(10 seconds), interval(50 milliseconds)) { + assert(strings.forall { DirectKafkaStreamSuite.collectedData.contains }) + } + } + + // Setup the streaming context + ssc = new StreamingContext(sparkConf, Milliseconds(100)) + val kafkaStream = withClue("Error creating direct stream") { + KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder]( + ssc, kafkaParams, Set(topic)) + } + val keyedStream = kafkaStream.map { v => "key" -> v._2.toInt } + val stateStream = keyedStream.updateStateByKey { (values: Seq[Int], state: Option[Int]) => + Some(values.sum + state.getOrElse(0)) + } + ssc.checkpoint(testDir.getAbsolutePath) + + // This is to collect the raw data received from Kafka + kafkaStream.foreachRDD { (rdd: RDD[(String, String)], time: Time) => + val data = rdd.map { _._2 }.collect() + DirectKafkaStreamSuite.collectedData.appendAll(data) + } + + // This is ensure all the data is eventually receiving only once + stateStream.foreachRDD { (rdd: RDD[(String, Int)]) => + rdd.collect().headOption.foreach { x => DirectKafkaStreamSuite.total = x._2 } + } + ssc.start() + + // Send some data and wait for them to be received + for (i <- (1 to 10).grouped(4)) { + sendDataAndWaitForReceive(i) + } + + // Verify that offset ranges were generated + val offsetRangesBeforeStop = getOffsetRanges(kafkaStream) + assert(offsetRangesBeforeStop.size >= 1, "No offset ranges generated") + assert( + offsetRangesBeforeStop.head._2.forall { _.fromOffset === 0 }, + "starting offset not zero" + ) + ssc.stop() + logInfo("====== RESTARTING ========") + + // Recover context from checkpoints + ssc = new StreamingContext(testDir.getAbsolutePath) + val recoveredStream = ssc.graph.getInputStreams().head.asInstanceOf[DStream[(String, String)]] + + // Verify offset ranges have been recovered + val recoveredOffsetRanges = getOffsetRanges(recoveredStream) + assert(recoveredOffsetRanges.size > 0, "No offset ranges recovered") + val earlierOffsetRangesAsSets = offsetRangesBeforeStop.map { x => (x._1, x._2.toSet) } + assert( + recoveredOffsetRanges.forall { or => + earlierOffsetRangesAsSets.contains((or._1, or._2.toSet)) + }, + "Recovered ranges are not the same as the ones generated" + ) + + // Restart context, give more data and verify the total at the end + // If the total is write that means each records has been received only once + ssc.start() + sendDataAndWaitForReceive(11 to 20) + eventually(timeout(10 seconds), interval(50 milliseconds)) { + assert(DirectKafkaStreamSuite.total === (1 to 20).sum) + } + ssc.stop() + } + + /** Get the generated offset ranges from the DirectKafkaStream */ + private def getOffsetRanges[K, V]( + kafkaStream: DStream[(K, V)]): Seq[(Time, Array[OffsetRange])] = { + kafkaStream.generatedRDDs.mapValues { rdd => + rdd.asInstanceOf[KafkaRDD[K, V, _, _, (K, V)]].offsetRanges + }.toSeq.sortBy { _._1 } + } +} + +object DirectKafkaStreamSuite { + val collectedData = new mutable.ArrayBuffer[String]() + var total = -1L +} diff --git a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaClusterSuite.scala b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaClusterSuite.scala index e57c8f6987fdc..fc9275b7207be 100644 --- a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaClusterSuite.scala +++ b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaClusterSuite.scala @@ -19,33 +19,29 @@ package org.apache.spark.streaming.kafka import scala.util.Random -import org.scalatest.BeforeAndAfter import kafka.common.TopicAndPartition +import org.scalatest.BeforeAndAfterAll -class KafkaClusterSuite extends KafkaStreamSuiteBase with BeforeAndAfter { - val brokerHost = "localhost" - - val kafkaParams = Map("metadata.broker.list" -> s"$brokerHost:$brokerPort") - - val kc = new KafkaCluster(kafkaParams) - +class KafkaClusterSuite extends KafkaStreamSuiteBase with BeforeAndAfterAll { val topic = "kcsuitetopic" + Random.nextInt(10000) - val topicAndPartition = TopicAndPartition(topic, 0) + var kc: KafkaCluster = null - before { + override def beforeAll() { setupKafka() createTopic(topic) - produceAndSendMessage(topic, Map("a" -> 1)) + sendMessages(topic, Map("a" -> 1)) + kc = new KafkaCluster(Map("metadata.broker.list" -> s"$brokerAddress")) } - after { + override def afterAll() { tearDownKafka() } test("metadata apis") { - val leader = kc.findLeaders(Set(topicAndPartition)).right.get - assert(leader(topicAndPartition) === (brokerHost, brokerPort), "didn't get leader") + val leader = kc.findLeaders(Set(topicAndPartition)).right.get(topicAndPartition) + val leaderAddress = s"${leader._1}:${leader._2}" + assert(leaderAddress === brokerAddress, "didn't get leader") val parts = kc.getPartitions(Set(topic)).right.get assert(parts(topicAndPartition), "didn't get partitions") diff --git a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaDirectStreamSuite.scala b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaDirectStreamSuite.scala deleted file mode 100644 index 0891ce344f16a..0000000000000 --- a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaDirectStreamSuite.scala +++ /dev/null @@ -1,92 +0,0 @@ -/* - * Licensed to the Apache Software Foundation (ASF) under one or more - * contributor license agreements. See the NOTICE file distributed with - * this work for additional information regarding copyright ownership. - * The ASF licenses this file to You under the Apache License, Version 2.0 - * (the "License"); you may not use this file except in compliance with - * the License. You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -package org.apache.spark.streaming.kafka - -import scala.util.Random -import scala.concurrent.duration._ - -import org.scalatest.BeforeAndAfter -import org.scalatest.concurrent.Eventually - -import kafka.serializer.StringDecoder - -import org.apache.spark.SparkConf -import org.apache.spark.storage.StorageLevel -import org.apache.spark.streaming.{Milliseconds, StreamingContext} - -class KafkaDirectStreamSuite extends KafkaStreamSuiteBase with BeforeAndAfter with Eventually { - val sparkConf = new SparkConf() - .setMaster("local[4]") - .setAppName(this.getClass.getSimpleName) - - val brokerHost = "localhost" - - val kafkaParams = Map( - "metadata.broker.list" -> s"$brokerHost:$brokerPort", - "auto.offset.reset" -> "smallest" - ) - - var ssc: StreamingContext = _ - - before { - setupKafka() - - ssc = new StreamingContext(sparkConf, Milliseconds(500)) - } - - after { - if (ssc != null) { - ssc.stop() - } - tearDownKafka() - } - - test("multi topic stream") { - val topics = Set("newA", "newB") - val data = Map("a" -> 7, "b" -> 9) - topics.foreach { t => - createTopic(t) - produceAndSendMessage(t, data) - } - val stream = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder]( - ssc, kafkaParams, topics) - var total = 0L; - - stream.foreachRDD { rdd => - val offsets = rdd.asInstanceOf[HasOffsetRanges].offsetRanges - val collected = rdd.mapPartitionsWithIndex { (i, iter) => - val off = offsets(i) - val all = iter.toSeq - val partSize = all.size - val rangeSize = off.untilOffset - off.fromOffset - all.map { _ => - (partSize, rangeSize) - }.toIterator - }.collect - collected.foreach { case (partSize, rangeSize) => - assert(partSize === rangeSize, "offset ranges are wrong") - } - total += collected.size - } - ssc.start() - eventually(timeout(20000.milliseconds), interval(200.milliseconds)) { - assert(total === data.values.sum * topics.size, "didn't get all messages") - } - ssc.stop() - } -} diff --git a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaRDDSuite.scala b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaRDDSuite.scala index 9b9e3f5fce8bd..6774db854a0d0 100644 --- a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaRDDSuite.scala +++ b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaRDDSuite.scala @@ -46,9 +46,9 @@ class KafkaRDDSuite extends KafkaStreamSuiteBase with BeforeAndAfter { val topic = "topic1" val sent = Map("a" -> 5, "b" -> 3, "c" -> 10) createTopic(topic) - produceAndSendMessage(topic, sent) + sendMessages(topic, sent) - val kafkaParams = Map("metadata.broker.list" -> s"localhost:$brokerPort", + val kafkaParams = Map("metadata.broker.list" -> brokerAddress, "group.id" -> s"test-consumer-${Random.nextInt(10000)}") val kc = new KafkaCluster(kafkaParams) @@ -65,14 +65,14 @@ class KafkaRDDSuite extends KafkaStreamSuiteBase with BeforeAndAfter { val rdd2 = getRdd(kc, Set(topic)) val sent2 = Map("d" -> 1) - produceAndSendMessage(topic, sent2) + sendMessages(topic, sent2) // this is the "0 messages" case // make sure we dont get anything, since messages were sent after rdd was defined assert(rdd2.isDefined) assert(rdd2.get.count === 0) val rdd3 = getRdd(kc, Set(topic)) - produceAndSendMessage(topic, Map("extra" -> 22)) + sendMessages(topic, Map("extra" -> 22)) // this is the "exactly 1 message" case // make sure we get exactly one message, despite there being lots more available assert(rdd3.isDefined) diff --git a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaStreamSuite.scala b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaStreamSuite.scala index f207dc6d4fa04..e4966eebb9b34 100644 --- a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaStreamSuite.scala +++ b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/KafkaStreamSuite.scala @@ -48,30 +48,41 @@ import org.apache.spark.util.Utils */ abstract class KafkaStreamSuiteBase extends FunSuite with Eventually with Logging { - var zkAddress: String = _ - var zkClient: ZkClient = _ - private val zkHost = "localhost" + private var zkPort: Int = 0 private val zkConnectionTimeout = 6000 private val zkSessionTimeout = 6000 private var zookeeper: EmbeddedZookeeper = _ - private var zkPort: Int = 0 - protected var brokerPort = 9092 + private val brokerHost = "localhost" + private var brokerPort = 9092 private var brokerConf: KafkaConfig = _ private var server: KafkaServer = _ private var producer: Producer[String, String] = _ + private var zkReady = false + private var brokerReady = false + + protected var zkClient: ZkClient = _ + + def zkAddress: String = { + assert(zkReady, "Zookeeper not setup yet or already torn down, cannot get zookeeper address") + s"$zkHost:$zkPort" + } + + def brokerAddress: String = { + assert(brokerReady, "Kafka not setup yet or already torn down, cannot get broker address") + s"$brokerHost:$brokerPort" + } def setupKafka() { // Zookeeper server startup zookeeper = new EmbeddedZookeeper(s"$zkHost:$zkPort") // Get the actual zookeeper binding port zkPort = zookeeper.actualPort - zkAddress = s"$zkHost:$zkPort" - logInfo("==================== 0 ====================") + zkReady = true + logInfo("==================== Zookeeper Started ====================") - zkClient = new ZkClient(zkAddress, zkSessionTimeout, zkConnectionTimeout, - ZKStringSerializer) - logInfo("==================== 1 ====================") + zkClient = new ZkClient(zkAddress, zkSessionTimeout, zkConnectionTimeout, ZKStringSerializer) + logInfo("==================== Zookeeper Client Created ====================") // Kafka broker startup var bindSuccess: Boolean = false @@ -80,9 +91,8 @@ abstract class KafkaStreamSuiteBase extends FunSuite with Eventually with Loggin val brokerProps = getBrokerConfig() brokerConf = new KafkaConfig(brokerProps) server = new KafkaServer(brokerConf) - logInfo("==================== 2 ====================") server.startup() - logInfo("==================== 3 ====================") + logInfo("==================== Kafka Broker Started ====================") bindSuccess = true } catch { case e: KafkaException => @@ -94,10 +104,13 @@ abstract class KafkaStreamSuiteBase extends FunSuite with Eventually with Loggin } Thread.sleep(2000) - logInfo("==================== 4 ====================") + logInfo("==================== Kafka + Zookeeper Ready ====================") + brokerReady = true } def tearDownKafka() { + brokerReady = false + zkReady = false if (producer != null) { producer.close() producer = null @@ -121,26 +134,23 @@ abstract class KafkaStreamSuiteBase extends FunSuite with Eventually with Loggin } } - private def createTestMessage(topic: String, sent: Map[String, Int]) - : Seq[KeyedMessage[String, String]] = { - val messages = for ((s, freq) <- sent; i <- 0 until freq) yield { - new KeyedMessage[String, String](topic, s) - } - messages.toSeq - } - def createTopic(topic: String) { AdminUtils.createTopic(zkClient, topic, 1, 1) - logInfo("==================== 5 ====================") // wait until metadata is propagated waitUntilMetadataIsPropagated(topic, 0) + logInfo(s"==================== Topic $topic Created ====================") } - def produceAndSendMessage(topic: String, sent: Map[String, Int]) { + def sendMessages(topic: String, messageToFreq: Map[String, Int]) { + val messages = messageToFreq.flatMap { case (s, freq) => Seq.fill(freq)(s) }.toArray + sendMessages(topic, messages) + } + + def sendMessages(topic: String, messages: Array[String]) { producer = new Producer[String, String](new ProducerConfig(getProducerConfig())) - producer.send(createTestMessage(topic, sent): _*) + producer.send(messages.map { new KeyedMessage[String, String](topic, _ ) }: _*) producer.close() - logInfo("==================== 6 ====================") + logInfo(s"==================== Sent Messages: ${messages.mkString(", ")} ====================") } private def getBrokerConfig(): Properties = { @@ -218,7 +228,7 @@ class KafkaStreamSuite extends KafkaStreamSuiteBase with BeforeAndAfter { val topic = "topic1" val sent = Map("a" -> 5, "b" -> 3, "c" -> 10) createTopic(topic) - produceAndSendMessage(topic, sent) + sendMessages(topic, sent) val kafkaParams = Map("zookeeper.connect" -> zkAddress, "group.id" -> s"test-consumer-${Random.nextInt(10000)}", diff --git a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/ReliableKafkaStreamSuite.scala b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/ReliableKafkaStreamSuite.scala index 64ccc92c81fa9..fc53c23abda85 100644 --- a/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/ReliableKafkaStreamSuite.scala +++ b/external/kafka/src/test/scala/org/apache/spark/streaming/kafka/ReliableKafkaStreamSuite.scala @@ -79,7 +79,7 @@ class ReliableKafkaStreamSuite extends KafkaStreamSuiteBase with BeforeAndAfter test("Reliable Kafka input stream with single topic") { var topic = "test-topic" createTopic(topic) - produceAndSendMessage(topic, data) + sendMessages(topic, data) // Verify whether the offset of this group/topic/partition is 0 before starting. assert(getCommitOffset(groupId, topic, 0) === None) @@ -111,7 +111,7 @@ class ReliableKafkaStreamSuite extends KafkaStreamSuiteBase with BeforeAndAfter val topics = Map("topic1" -> 1, "topic2" -> 1, "topic3" -> 1) topics.foreach { case (t, _) => createTopic(t) - produceAndSendMessage(t, data) + sendMessages(t, data) } // Before started, verify all the group/topic/partition offsets are 0. diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/EigenValueDecomposition.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/EigenValueDecomposition.scala index 3515461b52493..866936aa4f118 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/EigenValueDecomposition.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/EigenValueDecomposition.scala @@ -79,6 +79,9 @@ private[mllib] object EigenValueDecomposition { // Mode 1: A*x = lambda*x, A symmetric iparam(6) = 1 + require(n * ncv.toLong <= Integer.MAX_VALUE && ncv * (ncv.toLong + 8) <= Integer.MAX_VALUE, + s"k = $k and/or n = $n are too large to compute an eigendecomposition") + var ido = new intW(0) var info = new intW(0) var resid = new Array[Double](n) @@ -114,7 +117,7 @@ private[mllib] object EigenValueDecomposition { info.`val` match { case 1 => throw new IllegalStateException("ARPACK returns non-zero info = " + info.`val` + " Maximum number of iterations taken. (Refer ARPACK user guide for details)") - case 2 => throw new IllegalStateException("ARPACK returns non-zero info = " + info.`val` + + case 3 => throw new IllegalStateException("ARPACK returns non-zero info = " + info.`val` + " No shifts could be applied. Try to increase NCV. " + "(Refer ARPACK user guide for details)") case _ => throw new IllegalStateException("ARPACK returns non-zero info = " + info.`val` + diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala index 84f8ac2e0d9cd..89b38679b7494 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala @@ -50,7 +50,7 @@ sealed trait Matrix extends Serializable { private[mllib] def toBreeze: BM[Double] /** Gets the (i, j)-th element. */ - private[mllib] def apply(i: Int, j: Int): Double + def apply(i: Int, j: Int): Double /** Return the index for the (i, j)-th element in the backing array. */ private[mllib] def index(i: Int, j: Int): Int @@ -163,7 +163,7 @@ class DenseMatrix( private[mllib] def apply(i: Int): Double = values(i) - private[mllib] def apply(i: Int, j: Int): Double = values(index(i, j)) + override def apply(i: Int, j: Int): Double = values(index(i, j)) private[mllib] def index(i: Int, j: Int): Int = { if (!isTransposed) i + numRows * j else j + numCols * i @@ -256,8 +256,11 @@ object DenseMatrix { * @param numCols number of columns of the matrix * @return `DenseMatrix` with size `numRows` x `numCols` and values of zeros */ - def zeros(numRows: Int, numCols: Int): DenseMatrix = + def zeros(numRows: Int, numCols: Int): DenseMatrix = { + require(numRows.toLong * numCols <= Int.MaxValue, + s"$numRows x $numCols dense matrix is too large to allocate") new DenseMatrix(numRows, numCols, new Array[Double](numRows * numCols)) + } /** * Generate a `DenseMatrix` consisting of ones. @@ -265,8 +268,11 @@ object DenseMatrix { * @param numCols number of columns of the matrix * @return `DenseMatrix` with size `numRows` x `numCols` and values of ones */ - def ones(numRows: Int, numCols: Int): DenseMatrix = + def ones(numRows: Int, numCols: Int): DenseMatrix = { + require(numRows.toLong * numCols <= Int.MaxValue, + s"$numRows x $numCols dense matrix is too large to allocate") new DenseMatrix(numRows, numCols, Array.fill(numRows * numCols)(1.0)) + } /** * Generate an Identity Matrix in `DenseMatrix` format. @@ -291,6 +297,8 @@ object DenseMatrix { * @return `DenseMatrix` with size `numRows` x `numCols` and values in U(0, 1) */ def rand(numRows: Int, numCols: Int, rng: Random): DenseMatrix = { + require(numRows.toLong * numCols <= Int.MaxValue, + s"$numRows x $numCols dense matrix is too large to allocate") new DenseMatrix(numRows, numCols, Array.fill(numRows * numCols)(rng.nextDouble())) } @@ -302,6 +310,8 @@ object DenseMatrix { * @return `DenseMatrix` with size `numRows` x `numCols` and values in N(0, 1) */ def randn(numRows: Int, numCols: Int, rng: Random): DenseMatrix = { + require(numRows.toLong * numCols <= Int.MaxValue, + s"$numRows x $numCols dense matrix is too large to allocate") new DenseMatrix(numRows, numCols, Array.fill(numRows * numCols)(rng.nextGaussian())) } @@ -398,7 +408,7 @@ class SparseMatrix( } } - private[mllib] def apply(i: Int, j: Int): Double = { + override def apply(i: Int, j: Int): Double = { val ind = index(i, j) if (ind < 0) 0.0 else values(ind) } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/recommendation/ALS.scala b/mllib/src/main/scala/org/apache/spark/mllib/recommendation/ALS.scala index 4bb28d1b1e071..caacab943030b 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/recommendation/ALS.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/recommendation/ALS.scala @@ -18,7 +18,7 @@ package org.apache.spark.mllib.recommendation import org.apache.spark.Logging -import org.apache.spark.annotation.{DeveloperApi, Experimental} +import org.apache.spark.annotation.DeveloperApi import org.apache.spark.api.java.JavaRDD import org.apache.spark.ml.recommendation.{ALS => NewALS} import org.apache.spark.rdd.RDD diff --git a/mllib/src/main/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModel.scala index ed2f8b41bcae5..9ff06ac362a31 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModel.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModel.scala @@ -17,13 +17,17 @@ package org.apache.spark.mllib.recommendation +import java.io.IOException import java.lang.{Integer => JavaInteger} +import org.apache.hadoop.fs.Path import org.jblas.DoubleMatrix -import org.apache.spark.Logging +import org.apache.spark.{Logging, SparkContext} import org.apache.spark.api.java.{JavaPairRDD, JavaRDD} +import org.apache.spark.mllib.util.{Loader, Saveable} import org.apache.spark.rdd.RDD +import org.apache.spark.sql.{Row, SQLContext} import org.apache.spark.storage.StorageLevel /** @@ -41,7 +45,8 @@ import org.apache.spark.storage.StorageLevel class MatrixFactorizationModel( val rank: Int, val userFeatures: RDD[(Int, Array[Double])], - val productFeatures: RDD[(Int, Array[Double])]) extends Serializable with Logging { + val productFeatures: RDD[(Int, Array[Double])]) + extends Saveable with Serializable with Logging { require(rank > 0) validateFeatures("User", userFeatures) @@ -125,6 +130,12 @@ class MatrixFactorizationModel( recommend(productFeatures.lookup(product).head, userFeatures, num) .map(t => Rating(t._1, product, t._2)) + protected override val formatVersion: String = "1.0" + + override def save(sc: SparkContext, path: String): Unit = { + MatrixFactorizationModel.SaveLoadV1_0.save(this, path) + } + private def recommend( recommendToFeatures: Array[Double], recommendableFeatures: RDD[(Int, Array[Double])], @@ -136,3 +147,70 @@ class MatrixFactorizationModel( scored.top(num)(Ordering.by(_._2)) } } + +object MatrixFactorizationModel extends Loader[MatrixFactorizationModel] { + + import org.apache.spark.mllib.util.Loader._ + + override def load(sc: SparkContext, path: String): MatrixFactorizationModel = { + val (loadedClassName, formatVersion, metadata) = loadMetadata(sc, path) + val classNameV1_0 = SaveLoadV1_0.thisClassName + (loadedClassName, formatVersion) match { + case (className, "1.0") if className == classNameV1_0 => + SaveLoadV1_0.load(sc, path) + case _ => + throw new IOException("MatrixFactorizationModel.load did not recognize model with" + + s"(class: $loadedClassName, version: $formatVersion). Supported:\n" + + s" ($classNameV1_0, 1.0)") + } + } + + private[recommendation] + object SaveLoadV1_0 { + + private val thisFormatVersion = "1.0" + + private[recommendation] + val thisClassName = "org.apache.spark.mllib.recommendation.MatrixFactorizationModel" + + /** + * Saves a [[MatrixFactorizationModel]], where user features are saved under `data/users` and + * product features are saved under `data/products`. + */ + def save(model: MatrixFactorizationModel, path: String): Unit = { + val sc = model.userFeatures.sparkContext + val sqlContext = new SQLContext(sc) + import sqlContext.implicits.createDataFrame + val metadata = (thisClassName, thisFormatVersion, model.rank) + val metadataRDD = sc.parallelize(Seq(metadata), 1).toDataFrame("class", "version", "rank") + metadataRDD.toJSON.saveAsTextFile(metadataPath(path)) + model.userFeatures.toDataFrame("id", "features").saveAsParquetFile(userPath(path)) + model.productFeatures.toDataFrame("id", "features").saveAsParquetFile(productPath(path)) + } + + def load(sc: SparkContext, path: String): MatrixFactorizationModel = { + val sqlContext = new SQLContext(sc) + val (className, formatVersion, metadata) = loadMetadata(sc, path) + assert(className == thisClassName) + assert(formatVersion == thisFormatVersion) + val rank = metadata.select("rank").first().getInt(0) + val userFeatures = sqlContext.parquetFile(userPath(path)) + .map { case Row(id: Int, features: Seq[Double]) => + (id, features.toArray) + } + val productFeatures = sqlContext.parquetFile(productPath(path)) + .map { case Row(id: Int, features: Seq[Double]) => + (id, features.toArray) + } + new MatrixFactorizationModel(rank, userFeatures, productFeatures) + } + + private def userPath(path: String): String = { + new Path(dataPath(path), "user").toUri.toString + } + + private def productPath(path: String): String = { + new Path(dataPath(path), "product").toUri.toString + } + } +} diff --git a/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala b/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala index c854f124451cf..ce95c063db970 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearAlgorithm.scala @@ -21,7 +21,9 @@ import scala.reflect.ClassTag import org.apache.spark.Logging import org.apache.spark.annotation.DeveloperApi +import org.apache.spark.api.java.JavaSparkContext.fakeClassTag import org.apache.spark.mllib.linalg.{Vector, Vectors} +import org.apache.spark.streaming.api.java.{JavaDStream, JavaPairDStream} import org.apache.spark.streaming.dstream.DStream /** @@ -76,7 +78,7 @@ abstract class StreamingLinearAlgorithm[ * * @param data DStream containing labeled data */ - def trainOn(data: DStream[LabeledPoint]) { + def trainOn(data: DStream[LabeledPoint]): Unit = { if (model.isEmpty) { throw new IllegalArgumentException("Model must be initialized before starting training.") } @@ -99,6 +101,9 @@ abstract class StreamingLinearAlgorithm[ } } + /** Java-friendly version of `trainOn`. */ + def trainOn(data: JavaDStream[LabeledPoint]): Unit = trainOn(data.dstream) + /** * Use the model to make predictions on batches of data from a DStream * @@ -112,6 +117,11 @@ abstract class StreamingLinearAlgorithm[ data.map(model.get.predict) } + /** Java-friendly version of `predictOn`. */ + def predictOn(data: JavaDStream[Vector]): JavaDStream[java.lang.Double] = { + JavaDStream.fromDStream(predictOn(data.dstream).asInstanceOf[DStream[java.lang.Double]]) + } + /** * Use the model to make predictions on the values of a DStream and carry over its keys. * @param data DStream containing feature vectors @@ -124,4 +134,12 @@ abstract class StreamingLinearAlgorithm[ } data.mapValues(model.get.predict) } + + + /** Java-friendly version of `predictOnValues`. */ + def predictOnValues[K](data: JavaPairDStream[K, Vector]): JavaPairDStream[K, java.lang.Double] = { + implicit val tag = fakeClassTag[K] + JavaPairDStream.fromPairDStream( + predictOnValues(data.dstream).asInstanceOf[DStream[(K, java.lang.Double)]]) + } } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/stat/KernelDensity.scala b/mllib/src/main/scala/org/apache/spark/mllib/stat/KernelDensity.scala new file mode 100644 index 0000000000000..0deef11b4511a --- /dev/null +++ b/mllib/src/main/scala/org/apache/spark/mllib/stat/KernelDensity.scala @@ -0,0 +1,71 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.mllib.stat + +import org.apache.spark.rdd.RDD + +private[stat] object KernelDensity { + /** + * Given a set of samples from a distribution, estimates its density at the set of given points. + * Uses a Gaussian kernel with the given standard deviation. + */ + def estimate(samples: RDD[Double], standardDeviation: Double, + evaluationPoints: Array[Double]): Array[Double] = { + if (standardDeviation <= 0.0) { + throw new IllegalArgumentException("Standard deviation must be positive") + } + + // This gets used in each Gaussian PDF computation, so compute it up front + val logStandardDeviationPlusHalfLog2Pi = + Math.log(standardDeviation) + 0.5 * Math.log(2 * Math.PI) + + val (points, count) = samples.aggregate((new Array[Double](evaluationPoints.length), 0))( + (x, y) => { + var i = 0 + while (i < evaluationPoints.length) { + x._1(i) += normPdf(y, standardDeviation, logStandardDeviationPlusHalfLog2Pi, + evaluationPoints(i)) + i += 1 + } + (x._1, i) + }, + (x, y) => { + var i = 0 + while (i < evaluationPoints.length) { + x._1(i) += y._1(i) + i += 1 + } + (x._1, x._2 + y._2) + }) + + var i = 0 + while (i < points.length) { + points(i) /= count + i += 1 + } + points + } + + private def normPdf(mean: Double, standardDeviation: Double, + logStandardDeviationPlusHalfLog2Pi: Double, x: Double): Double = { + val x0 = x - mean + val x1 = x0 / standardDeviation + val logDensity = -0.5 * x1 * x1 - logStandardDeviationPlusHalfLog2Pi + Math.exp(logDensity) + } +} diff --git a/mllib/src/main/scala/org/apache/spark/mllib/stat/Statistics.scala b/mllib/src/main/scala/org/apache/spark/mllib/stat/Statistics.scala index b3fad0c52d655..32561620ac914 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/stat/Statistics.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/stat/Statistics.scala @@ -149,4 +149,18 @@ object Statistics { def chiSqTest(data: RDD[LabeledPoint]): Array[ChiSqTestResult] = { ChiSqTest.chiSquaredFeatures(data) } + + /** + * Given an empirical distribution defined by the input RDD of samples, estimate its density at + * each of the given evaluation points using a Gaussian kernel. + * + * @param samples The samples RDD used to define the empirical distribution. + * @param standardDeviation The standard deviation of the kernel Gaussians. + * @param evaluationPoints The points at which to estimate densities. + * @return An array the same size as evaluationPoints with the density at each point. + */ + def kernelDensity(samples: RDD[Double], standardDeviation: Double, + evaluationPoints: Iterable[Double]): Array[Double] = { + KernelDensity.estimate(samples, standardDeviation, evaluationPoints.toArray) + } } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala index a25e625a4017a..89ecf3773dd77 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala @@ -17,11 +17,17 @@ package org.apache.spark.mllib.tree.model +import scala.collection.mutable + +import org.apache.spark.SparkContext import org.apache.spark.annotation.Experimental import org.apache.spark.api.java.JavaRDD import org.apache.spark.mllib.linalg.Vector +import org.apache.spark.mllib.tree.configuration.{Algo, FeatureType} import org.apache.spark.mllib.tree.configuration.Algo._ +import org.apache.spark.mllib.util.{Loader, Saveable} import org.apache.spark.rdd.RDD +import org.apache.spark.sql.{DataFrame, Row, SQLContext} /** * :: Experimental :: @@ -31,7 +37,7 @@ import org.apache.spark.rdd.RDD * @param algo algorithm type -- classification or regression */ @Experimental -class DecisionTreeModel(val topNode: Node, val algo: Algo) extends Serializable { +class DecisionTreeModel(val topNode: Node, val algo: Algo) extends Serializable with Saveable { /** * Predict values for a single data point using the model trained. @@ -98,4 +104,193 @@ class DecisionTreeModel(val topNode: Node, val algo: Algo) extends Serializable header + topNode.subtreeToString(2) } + override def save(sc: SparkContext, path: String): Unit = { + DecisionTreeModel.SaveLoadV1_0.save(sc, path, this) + } + + override protected def formatVersion: String = "1.0" +} + +object DecisionTreeModel extends Loader[DecisionTreeModel] { + + private[tree] object SaveLoadV1_0 { + + def thisFormatVersion = "1.0" + + // Hard-code class name string in case it changes in the future + def thisClassName = "org.apache.spark.mllib.tree.DecisionTreeModel" + + case class PredictData(predict: Double, prob: Double) { + def toPredict: Predict = new Predict(predict, prob) + } + + object PredictData { + def apply(p: Predict): PredictData = PredictData(p.predict, p.prob) + + def apply(r: Row): PredictData = PredictData(r.getDouble(0), r.getDouble(1)) + } + + case class SplitData( + feature: Int, + threshold: Double, + featureType: Int, + categories: Seq[Double]) { // TODO: Change to List once SPARK-3365 is fixed + def toSplit: Split = { + new Split(feature, threshold, FeatureType(featureType), categories.toList) + } + } + + object SplitData { + def apply(s: Split): SplitData = { + SplitData(s.feature, s.threshold, s.featureType.id, s.categories) + } + + def apply(r: Row): SplitData = { + SplitData(r.getInt(0), r.getDouble(1), r.getInt(2), r.getAs[Seq[Double]](3)) + } + } + + /** Model data for model import/export */ + case class NodeData( + treeId: Int, + nodeId: Int, + predict: PredictData, + impurity: Double, + isLeaf: Boolean, + split: Option[SplitData], + leftNodeId: Option[Int], + rightNodeId: Option[Int], + infoGain: Option[Double]) + + object NodeData { + def apply(treeId: Int, n: Node): NodeData = { + NodeData(treeId, n.id, PredictData(n.predict), n.impurity, n.isLeaf, + n.split.map(SplitData.apply), n.leftNode.map(_.id), n.rightNode.map(_.id), + n.stats.map(_.gain)) + } + + def apply(r: Row): NodeData = { + val split = if (r.isNullAt(5)) None else Some(SplitData(r.getStruct(5))) + val leftNodeId = if (r.isNullAt(6)) None else Some(r.getInt(6)) + val rightNodeId = if (r.isNullAt(7)) None else Some(r.getInt(7)) + val infoGain = if (r.isNullAt(8)) None else Some(r.getDouble(8)) + NodeData(r.getInt(0), r.getInt(1), PredictData(r.getStruct(2)), r.getDouble(3), + r.getBoolean(4), split, leftNodeId, rightNodeId, infoGain) + } + } + + def save(sc: SparkContext, path: String, model: DecisionTreeModel): Unit = { + val sqlContext = new SQLContext(sc) + import sqlContext.implicits._ + + // Create JSON metadata. + val metadataRDD = sc.parallelize( + Seq((thisClassName, thisFormatVersion, model.algo.toString, model.numNodes)), 1) + .toDataFrame("class", "version", "algo", "numNodes") + metadataRDD.toJSON.saveAsTextFile(Loader.metadataPath(path)) + + // Create Parquet data. + val nodes = model.topNode.subtreeIterator.toSeq + val dataRDD: DataFrame = sc.parallelize(nodes) + .map(NodeData.apply(0, _)) + .toDataFrame + dataRDD.saveAsParquetFile(Loader.dataPath(path)) + } + + def load(sc: SparkContext, path: String, algo: String, numNodes: Int): DecisionTreeModel = { + val datapath = Loader.dataPath(path) + val sqlContext = new SQLContext(sc) + // Load Parquet data. + val dataRDD = sqlContext.parquetFile(datapath) + // Check schema explicitly since erasure makes it hard to use match-case for checking. + Loader.checkSchema[NodeData](dataRDD.schema) + val nodes = dataRDD.map(NodeData.apply) + // Build node data into a tree. + val trees = constructTrees(nodes) + assert(trees.size == 1, + "Decision tree should contain exactly one tree but got ${trees.size} trees.") + val model = new DecisionTreeModel(trees(0), Algo.fromString(algo)) + assert(model.numNodes == numNodes, s"Unable to load DecisionTreeModel data from: $datapath." + + s" Expected $numNodes nodes but found ${model.numNodes}") + model + } + + def constructTrees(nodes: RDD[NodeData]): Array[Node] = { + val trees = nodes + .groupBy(_.treeId) + .mapValues(_.toArray) + .collect() + .map { case (treeId, data) => + (treeId, constructTree(data)) + }.sortBy(_._1) + val numTrees = trees.size + val treeIndices = trees.map(_._1).toSeq + assert(treeIndices == (0 until numTrees), + s"Tree indices must start from 0 and increment by 1, but we found $treeIndices.") + trees.map(_._2) + } + + /** + * Given a list of nodes from a tree, construct the tree. + * @param data array of all node data in a tree. + */ + def constructTree(data: Array[NodeData]): Node = { + val dataMap: Map[Int, NodeData] = data.map(n => n.nodeId -> n).toMap + assert(dataMap.contains(1), + s"DecisionTree missing root node (id = 1).") + constructNode(1, dataMap, mutable.Map.empty) + } + + /** + * Builds a node from the node data map and adds new nodes to the input nodes map. + */ + private def constructNode( + id: Int, + dataMap: Map[Int, NodeData], + nodes: mutable.Map[Int, Node]): Node = { + if (nodes.contains(id)) { + return nodes(id) + } + val data = dataMap(id) + val node = + if (data.isLeaf) { + Node(data.nodeId, data.predict.toPredict, data.impurity, data.isLeaf) + } else { + val leftNode = constructNode(data.leftNodeId.get, dataMap, nodes) + val rightNode = constructNode(data.rightNodeId.get, dataMap, nodes) + val stats = new InformationGainStats(data.infoGain.get, data.impurity, leftNode.impurity, + rightNode.impurity, leftNode.predict, rightNode.predict) + new Node(data.nodeId, data.predict.toPredict, data.impurity, data.isLeaf, + data.split.map(_.toSplit), Some(leftNode), Some(rightNode), Some(stats)) + } + nodes += node.id -> node + node + } + } + + override def load(sc: SparkContext, path: String): DecisionTreeModel = { + val (loadedClassName, version, metadata) = Loader.loadMetadata(sc, path) + val (algo: String, numNodes: Int) = try { + val algo_numNodes = metadata.select("algo", "numNodes").collect() + assert(algo_numNodes.length == 1) + algo_numNodes(0) match { + case Row(a: String, n: Int) => (a, n) + } + } catch { + // Catch both Error and Exception since the checks above can throw either. + case e: Throwable => + throw new Exception( + s"Unable to load DecisionTreeModel metadata from: ${Loader.metadataPath(path)}." + + s" Error message: ${e.getMessage}") + } + val classNameV1_0 = SaveLoadV1_0.thisClassName + (loadedClassName, version) match { + case (className, "1.0") if className == classNameV1_0 => + SaveLoadV1_0.load(sc, path, algo, numNodes) + case _ => throw new Exception( + s"DecisionTreeModel.load did not recognize model with (className, format version):" + + s"($loadedClassName, $version). Supported:\n" + + s" ($classNameV1_0, 1.0)") + } + } } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/InformationGainStats.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/InformationGainStats.scala index 9a50ecb550c38..80990aa9a603f 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/InformationGainStats.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/InformationGainStats.scala @@ -49,7 +49,9 @@ class InformationGainStats( gain == other.gain && impurity == other.impurity && leftImpurity == other.leftImpurity && - rightImpurity == other.rightImpurity + rightImpurity == other.rightImpurity && + leftPredict == other.leftPredict && + rightPredict == other.rightPredict } case _ => false } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Node.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Node.scala index 2179da8dbe03e..d961081d185e9 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Node.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Node.scala @@ -166,6 +166,11 @@ class Node ( } } + /** Returns an iterator that traverses (DFS, left to right) the subtree of this node. */ + private[tree] def subtreeIterator: Iterator[Node] = { + Iterator.single(this) ++ leftNode.map(_.subtreeIterator).getOrElse(Iterator.empty) ++ + rightNode.map(_.subtreeIterator).getOrElse(Iterator.empty) + } } private[tree] object Node { diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Predict.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Predict.scala index 004838ee5ba0e..ad4c0dbbfb3e5 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Predict.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Predict.scala @@ -32,4 +32,11 @@ class Predict( override def toString = { "predict = %f, prob = %f".format(predict, prob) } + + override def equals(other: Any): Boolean = { + other match { + case p: Predict => predict == p.predict && prob == p.prob + case _ => false + } + } } diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/treeEnsembleModels.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/treeEnsembleModels.scala index 22997110de8dd..23bd46baabf65 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/treeEnsembleModels.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/treeEnsembleModels.scala @@ -21,12 +21,17 @@ import scala.collection.mutable import com.github.fommil.netlib.BLAS.{getInstance => blas} +import org.apache.spark.SparkContext import org.apache.spark.annotation.Experimental import org.apache.spark.api.java.JavaRDD import org.apache.spark.mllib.linalg.Vector import org.apache.spark.mllib.tree.configuration.Algo._ +import org.apache.spark.mllib.tree.configuration.Algo import org.apache.spark.mllib.tree.configuration.EnsembleCombiningStrategy._ +import org.apache.spark.mllib.util.{Saveable, Loader} import org.apache.spark.rdd.RDD +import org.apache.spark.sql.{DataFrame, SQLContext} + /** * :: Experimental :: @@ -38,9 +43,42 @@ import org.apache.spark.rdd.RDD @Experimental class RandomForestModel(override val algo: Algo, override val trees: Array[DecisionTreeModel]) extends TreeEnsembleModel(algo, trees, Array.fill(trees.size)(1.0), - combiningStrategy = if (algo == Classification) Vote else Average) { + combiningStrategy = if (algo == Classification) Vote else Average) + with Saveable { require(trees.forall(_.algo == algo)) + + override def save(sc: SparkContext, path: String): Unit = { + TreeEnsembleModel.SaveLoadV1_0.save(sc, path, this, + RandomForestModel.SaveLoadV1_0.thisClassName) + } + + override protected def formatVersion: String = TreeEnsembleModel.SaveLoadV1_0.thisFormatVersion +} + +object RandomForestModel extends Loader[RandomForestModel] { + + override def load(sc: SparkContext, path: String): RandomForestModel = { + val (loadedClassName, version, metadataRDD) = Loader.loadMetadata(sc, path) + val classNameV1_0 = SaveLoadV1_0.thisClassName + (loadedClassName, version) match { + case (className, "1.0") if className == classNameV1_0 => + val metadata = TreeEnsembleModel.SaveLoadV1_0.readMetadata(metadataRDD, path) + assert(metadata.treeWeights.forall(_ == 1.0)) + val trees = + TreeEnsembleModel.SaveLoadV1_0.loadTrees(sc, path, metadata.treeAlgo) + new RandomForestModel(Algo.fromString(metadata.algo), trees) + case _ => throw new Exception(s"RandomForestModel.load did not recognize model" + + s" with (className, format version): ($loadedClassName, $version). Supported:\n" + + s" ($classNameV1_0, 1.0)") + } + } + + private object SaveLoadV1_0 { + // Hard-code class name string in case it changes in the future + def thisClassName = "org.apache.spark.mllib.tree.model.RandomForestModel" + } + } /** @@ -56,9 +94,42 @@ class GradientBoostedTreesModel( override val algo: Algo, override val trees: Array[DecisionTreeModel], override val treeWeights: Array[Double]) - extends TreeEnsembleModel(algo, trees, treeWeights, combiningStrategy = Sum) { + extends TreeEnsembleModel(algo, trees, treeWeights, combiningStrategy = Sum) + with Saveable { require(trees.size == treeWeights.size) + + override def save(sc: SparkContext, path: String): Unit = { + TreeEnsembleModel.SaveLoadV1_0.save(sc, path, this, + GradientBoostedTreesModel.SaveLoadV1_0.thisClassName) + } + + override protected def formatVersion: String = TreeEnsembleModel.SaveLoadV1_0.thisFormatVersion +} + +object GradientBoostedTreesModel extends Loader[GradientBoostedTreesModel] { + + override def load(sc: SparkContext, path: String): GradientBoostedTreesModel = { + val (loadedClassName, version, metadataRDD) = Loader.loadMetadata(sc, path) + val classNameV1_0 = SaveLoadV1_0.thisClassName + (loadedClassName, version) match { + case (className, "1.0") if className == classNameV1_0 => + val metadata = TreeEnsembleModel.SaveLoadV1_0.readMetadata(metadataRDD, path) + assert(metadata.combiningStrategy == Sum.toString) + val trees = + TreeEnsembleModel.SaveLoadV1_0.loadTrees(sc, path, metadata.treeAlgo) + new GradientBoostedTreesModel(Algo.fromString(metadata.algo), trees, metadata.treeWeights) + case _ => throw new Exception(s"GradientBoostedTreesModel.load did not recognize model" + + s" with (className, format version): ($loadedClassName, $version). Supported:\n" + + s" ($classNameV1_0, 1.0)") + } + } + + private object SaveLoadV1_0 { + // Hard-code class name string in case it changes in the future + def thisClassName = "org.apache.spark.mllib.tree.model.GradientBoostedTreesModel" + } + } /** @@ -176,3 +247,85 @@ private[tree] sealed class TreeEnsembleModel( */ def totalNumNodes: Int = trees.map(_.numNodes).sum } + +private[tree] object TreeEnsembleModel { + + object SaveLoadV1_0 { + + import DecisionTreeModel.SaveLoadV1_0.{NodeData, constructTrees} + + def thisFormatVersion = "1.0" + + case class Metadata( + algo: String, + treeAlgo: String, + combiningStrategy: String, + treeWeights: Array[Double]) + + /** + * Model data for model import/export. + * We have to duplicate NodeData here since Spark SQL does not yet support extracting subfields + * of nested fields; once that is possible, we can use something like: + * case class EnsembleNodeData(treeId: Int, node: NodeData), + * where NodeData is from DecisionTreeModel. + */ + case class EnsembleNodeData(treeId: Int, node: NodeData) + + def save(sc: SparkContext, path: String, model: TreeEnsembleModel, className: String): Unit = { + val sqlContext = new SQLContext(sc) + import sqlContext.implicits._ + + // Create JSON metadata. + val metadata = Metadata(model.algo.toString, model.trees(0).algo.toString, + model.combiningStrategy.toString, model.treeWeights) + val metadataRDD = sc.parallelize(Seq((className, thisFormatVersion, metadata)), 1) + .toDataFrame("class", "version", "metadata") + metadataRDD.toJSON.saveAsTextFile(Loader.metadataPath(path)) + + // Create Parquet data. + val dataRDD = sc.parallelize(model.trees.zipWithIndex).flatMap { case (tree, treeId) => + tree.topNode.subtreeIterator.toSeq.map(node => NodeData(treeId, node)) + }.toDataFrame + dataRDD.saveAsParquetFile(Loader.dataPath(path)) + } + + /** + * Read metadata from the loaded metadata DataFrame. + * @param path Path for loading data, used for debug messages. + */ + def readMetadata(metadata: DataFrame, path: String): Metadata = { + try { + // We rely on the try-catch for schema checking rather than creating a schema just for this. + val metadataArray = metadata.select("metadata.algo", "metadata.treeAlgo", + "metadata.combiningStrategy", "metadata.treeWeights").collect() + assert(metadataArray.size == 1) + Metadata(metadataArray(0).getString(0), metadataArray(0).getString(1), + metadataArray(0).getString(2), metadataArray(0).getAs[Seq[Double]](3).toArray) + } catch { + // Catch both Error and Exception since the checks above can throw either. + case e: Throwable => + throw new Exception( + s"Unable to load TreeEnsembleModel metadata from: ${Loader.metadataPath(path)}." + + s" Error message: ${e.getMessage}") + } + } + + /** + * Load trees for an ensemble, and return them in order. + * @param path path to load the model from + * @param treeAlgo Algorithm for individual trees (which may differ from the ensemble's + * algorithm). + */ + def loadTrees( + sc: SparkContext, + path: String, + treeAlgo: String): Array[DecisionTreeModel] = { + val datapath = Loader.dataPath(path) + val sqlContext = new SQLContext(sc) + val nodes = sqlContext.parquetFile(datapath).map(NodeData.apply) + val trees = constructTrees(nodes) + trees.map(new DecisionTreeModel(_, Algo.fromString(treeAlgo))) + } + } + +} diff --git a/mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java b/mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java new file mode 100644 index 0000000000000..ac945ba6f23c1 --- /dev/null +++ b/mllib/src/test/java/org/apache/spark/ml/classification/JavaStreamingLogisticRegressionSuite.java @@ -0,0 +1,82 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.ml.classification; + +import java.io.Serializable; +import java.util.List; + +import scala.Tuple2; + +import com.google.common.collect.Lists; +import org.junit.After; +import org.junit.Before; +import org.junit.Test; + +import org.apache.spark.SparkConf; +import org.apache.spark.mllib.classification.StreamingLogisticRegressionWithSGD; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.mllib.regression.LabeledPoint; +import org.apache.spark.streaming.Duration; +import org.apache.spark.streaming.api.java.JavaDStream; +import org.apache.spark.streaming.api.java.JavaPairDStream; +import org.apache.spark.streaming.api.java.JavaStreamingContext; +import static org.apache.spark.streaming.JavaTestUtils.*; + +public class JavaStreamingLogisticRegressionSuite implements Serializable { + + protected transient JavaStreamingContext ssc; + + @Before + public void setUp() { + SparkConf conf = new SparkConf() + .setMaster("local[2]") + .setAppName("test") + .set("spark.streaming.clock", "org.apache.spark.streaming.util.ManualClock"); + ssc = new JavaStreamingContext(conf, new Duration(1000)); + ssc.checkpoint("checkpoint"); + } + + @After + public void tearDown() { + ssc.stop(); + ssc = null; + } + + @Test + @SuppressWarnings("unchecked") + public void javaAPI() { + List trainingBatch = Lists.newArrayList( + new LabeledPoint(1.0, Vectors.dense(1.0)), + new LabeledPoint(0.0, Vectors.dense(0.0))); + JavaDStream training = + attachTestInputStream(ssc, Lists.newArrayList(trainingBatch, trainingBatch), 2); + List> testBatch = Lists.newArrayList( + new Tuple2(10, Vectors.dense(1.0)), + new Tuple2(11, Vectors.dense(0.0))); + JavaPairDStream test = JavaPairDStream.fromJavaDStream( + attachTestInputStream(ssc, Lists.newArrayList(testBatch, testBatch), 2)); + StreamingLogisticRegressionWithSGD slr = new StreamingLogisticRegressionWithSGD() + .setNumIterations(2) + .setInitialWeights(Vectors.dense(0.0)); + slr.trainOn(training); + JavaPairDStream prediction = slr.predictOnValues(test); + attachTestOutputStream(prediction.count()); + runStreams(ssc, 2, 2); + } +} diff --git a/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java b/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java new file mode 100644 index 0000000000000..a4dd1ac39a3c8 --- /dev/null +++ b/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java @@ -0,0 +1,80 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.mllib.regression; + +import java.io.Serializable; +import java.util.List; + +import scala.Tuple2; + +import com.google.common.collect.Lists; +import org.junit.After; +import org.junit.Before; +import org.junit.Test; + +import org.apache.spark.SparkConf; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.streaming.Duration; +import org.apache.spark.streaming.api.java.JavaDStream; +import org.apache.spark.streaming.api.java.JavaPairDStream; +import org.apache.spark.streaming.api.java.JavaStreamingContext; +import static org.apache.spark.streaming.JavaTestUtils.*; + +public class JavaStreamingLinearRegressionSuite implements Serializable { + + protected transient JavaStreamingContext ssc; + + @Before + public void setUp() { + SparkConf conf = new SparkConf() + .setMaster("local[2]") + .setAppName("test") + .set("spark.streaming.clock", "org.apache.spark.streaming.util.ManualClock"); + ssc = new JavaStreamingContext(conf, new Duration(1000)); + ssc.checkpoint("checkpoint"); + } + + @After + public void tearDown() { + ssc.stop(); + ssc = null; + } + + @Test + @SuppressWarnings("unchecked") + public void javaAPI() { + List trainingBatch = Lists.newArrayList( + new LabeledPoint(1.0, Vectors.dense(1.0)), + new LabeledPoint(0.0, Vectors.dense(0.0))); + JavaDStream training = + attachTestInputStream(ssc, Lists.newArrayList(trainingBatch, trainingBatch), 2); + List> testBatch = Lists.newArrayList( + new Tuple2(10, Vectors.dense(1.0)), + new Tuple2(11, Vectors.dense(0.0))); + JavaPairDStream test = JavaPairDStream.fromJavaDStream( + attachTestInputStream(ssc, Lists.newArrayList(testBatch, testBatch), 2)); + StreamingLinearRegressionWithSGD slr = new StreamingLinearRegressionWithSGD() + .setNumIterations(2) + .setInitialWeights(Vectors.dense(0.0)); + slr.trainOn(training); + JavaPairDStream prediction = slr.predictOnValues(test); + attachTestOutputStream(prediction.count()); + runStreams(ssc, 2, 2); + } +} diff --git a/mllib/src/test/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModelSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModelSuite.scala index b9caecc904a23..9801e87576744 100644 --- a/mllib/src/test/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModelSuite.scala +++ b/mllib/src/test/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModelSuite.scala @@ -22,6 +22,7 @@ import org.scalatest.FunSuite import org.apache.spark.mllib.util.MLlibTestSparkContext import org.apache.spark.mllib.util.TestingUtils._ import org.apache.spark.rdd.RDD +import org.apache.spark.util.Utils class MatrixFactorizationModelSuite extends FunSuite with MLlibTestSparkContext { @@ -53,4 +54,22 @@ class MatrixFactorizationModelSuite extends FunSuite with MLlibTestSparkContext new MatrixFactorizationModel(rank, userFeatures, prodFeatures1) } } + + test("save/load") { + val model = new MatrixFactorizationModel(rank, userFeatures, prodFeatures) + val tempDir = Utils.createTempDir() + val path = tempDir.toURI.toString + def collect(features: RDD[(Int, Array[Double])]): Set[(Int, Seq[Double])] = { + features.mapValues(_.toSeq).collect().toSet + } + try { + model.save(sc, path) + val newModel = MatrixFactorizationModel.load(sc, path) + assert(newModel.rank === rank) + assert(collect(newModel.userFeatures) === collect(userFeatures)) + assert(collect(newModel.productFeatures) === collect(prodFeatures)) + } finally { + Utils.deleteRecursively(tempDir) + } + } } diff --git a/mllib/src/test/scala/org/apache/spark/mllib/stat/KernelDensitySuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/stat/KernelDensitySuite.scala new file mode 100644 index 0000000000000..f6a1e19f50296 --- /dev/null +++ b/mllib/src/test/scala/org/apache/spark/mllib/stat/KernelDensitySuite.scala @@ -0,0 +1,47 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.mllib.stat + +import org.scalatest.FunSuite + +import org.apache.commons.math3.distribution.NormalDistribution + +import org.apache.spark.mllib.util.LocalClusterSparkContext + +class KernelDensitySuite extends FunSuite with LocalClusterSparkContext { + test("kernel density single sample") { + val rdd = sc.parallelize(Array(5.0)) + val evaluationPoints = Array(5.0, 6.0) + val densities = KernelDensity.estimate(rdd, 3.0, evaluationPoints) + val normal = new NormalDistribution(5.0, 3.0) + val acceptableErr = 1e-6 + assert(densities(0) - normal.density(5.0) < acceptableErr) + assert(densities(0) - normal.density(6.0) < acceptableErr) + } + + test("kernel density multiple samples") { + val rdd = sc.parallelize(Array(5.0, 10.0)) + val evaluationPoints = Array(5.0, 6.0) + val densities = KernelDensity.estimate(rdd, 3.0, evaluationPoints) + val normal1 = new NormalDistribution(5.0, 3.0) + val normal2 = new NormalDistribution(10.0, 3.0) + val acceptableErr = 1e-6 + assert(densities(0) - (normal1.density(5.0) + normal2.density(5.0)) / 2 < acceptableErr) + assert(densities(0) - (normal1.density(6.0) + normal2.density(6.0)) / 2 < acceptableErr) + } +} diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala index 9347eaf9221a8..7b1aed5ffeb3e 100644 --- a/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala +++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala @@ -29,8 +29,10 @@ import org.apache.spark.mllib.tree.configuration.FeatureType._ import org.apache.spark.mllib.tree.configuration.{QuantileStrategy, Strategy} import org.apache.spark.mllib.tree.impl.{BaggedPoint, DecisionTreeMetadata, TreePoint} import org.apache.spark.mllib.tree.impurity.{Entropy, Gini, Variance} -import org.apache.spark.mllib.tree.model.{InformationGainStats, DecisionTreeModel, Node} +import org.apache.spark.mllib.tree.model._ import org.apache.spark.mllib.util.MLlibTestSparkContext +import org.apache.spark.util.Utils + class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext { @@ -857,9 +859,32 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext { assert(topNode.leftNode.get.impurity === 0.0) assert(topNode.rightNode.get.impurity === 0.0) } + + test("Node.subtreeIterator") { + val model = DecisionTreeSuite.createModel(Classification) + val nodeIds = model.topNode.subtreeIterator.map(_.id).toArray.sorted + assert(nodeIds === DecisionTreeSuite.createdModelNodeIds) + } + + test("model save/load") { + val tempDir = Utils.createTempDir() + val path = tempDir.toURI.toString + + Array(Classification, Regression).foreach { algo => + val model = DecisionTreeSuite.createModel(algo) + // Save model, load it back, and compare. + try { + model.save(sc, path) + val sameModel = DecisionTreeModel.load(sc, path) + DecisionTreeSuite.checkEqual(model, sameModel) + } finally { + Utils.deleteRecursively(tempDir) + } + } + } } -object DecisionTreeSuite { +object DecisionTreeSuite extends FunSuite { def validateClassifier( model: DecisionTreeModel, @@ -979,4 +1004,95 @@ object DecisionTreeSuite { arr } + /** Create a leaf node with the given node ID */ + private def createLeafNode(id: Int): Node = { + Node(nodeIndex = id, new Predict(0.0, 1.0), impurity = 0.5, isLeaf = true) + } + + /** + * Create an internal node with the given node ID and feature type. + * Note: This does NOT set the child nodes. + */ + private def createInternalNode(id: Int, featureType: FeatureType): Node = { + val node = Node(nodeIndex = id, new Predict(0.0, 1.0), impurity = 0.5, isLeaf = false) + featureType match { + case Continuous => + node.split = Some(new Split(feature = 0, threshold = 0.5, Continuous, + categories = List.empty[Double])) + case Categorical => + node.split = Some(new Split(feature = 1, threshold = 0.0, Categorical, + categories = List(0.0, 1.0))) + } + // TODO: The information gain stats should be consistent with the same info stored in children. + node.stats = Some(new InformationGainStats(gain = 0.1, impurity = 0.2, + leftImpurity = 0.3, rightImpurity = 0.4, new Predict(1.0, 0.4), new Predict(0.0, 0.6))) + node + } + + /** + * Create a tree model. This is deterministic and contains a variety of node and feature types. + */ + private[tree] def createModel(algo: Algo): DecisionTreeModel = { + val topNode = createInternalNode(id = 1, Continuous) + val (node2, node3) = (createLeafNode(id = 2), createInternalNode(id = 3, Categorical)) + val (node6, node7) = (createLeafNode(id = 6), createLeafNode(id = 7)) + topNode.leftNode = Some(node2) + topNode.rightNode = Some(node3) + node3.leftNode = Some(node6) + node3.rightNode = Some(node7) + new DecisionTreeModel(topNode, algo) + } + + /** Sorted Node IDs matching the model returned by [[createModel()]] */ + private val createdModelNodeIds = Array(1, 2, 3, 6, 7) + + /** + * Check if the two trees are exactly the same. + * Note: I hesitate to override Node.equals since it could cause problems if users + * make mistakes such as creating loops of Nodes. + * If the trees are not equal, this prints the two trees and throws an exception. + */ + private[tree] def checkEqual(a: DecisionTreeModel, b: DecisionTreeModel): Unit = { + try { + assert(a.algo === b.algo) + checkEqual(a.topNode, b.topNode) + } catch { + case ex: Exception => + throw new AssertionError("checkEqual failed since the two trees were not identical.\n" + + "TREE A:\n" + a.toDebugString + "\n" + + "TREE B:\n" + b.toDebugString + "\n", ex) + } + } + + /** + * Return true iff the two nodes and their descendents are exactly the same. + * Note: I hesitate to override Node.equals since it could cause problems if users + * make mistakes such as creating loops of Nodes. + */ + private def checkEqual(a: Node, b: Node): Unit = { + assert(a.id === b.id) + assert(a.predict === b.predict) + assert(a.impurity === b.impurity) + assert(a.isLeaf === b.isLeaf) + assert(a.split === b.split) + (a.stats, b.stats) match { + // TODO: Check other fields besides the infomation gain. + case (Some(aStats), Some(bStats)) => assert(aStats.gain === bStats.gain) + case (None, None) => + case _ => throw new AssertionError( + s"Only one instance has stats defined. (a.stats: ${a.stats}, b.stats: ${b.stats})") + } + (a.leftNode, b.leftNode) match { + case (Some(aNode), Some(bNode)) => checkEqual(aNode, bNode) + case (None, None) => + case _ => throw new AssertionError("Only one instance has leftNode defined. " + + s"(a.leftNode: ${a.leftNode}, b.leftNode: ${b.leftNode})") + } + (a.rightNode, b.rightNode) match { + case (Some(aNode: Node), Some(bNode: Node)) => checkEqual(aNode, bNode) + case (None, None) => + case _ => throw new AssertionError("Only one instance has rightNode defined. " + + s"(a.rightNode: ${a.rightNode}, b.rightNode: ${b.rightNode})") + } + } } diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala index e8341a5d0d104..bde47606eb001 100644 --- a/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala +++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala @@ -24,8 +24,10 @@ import org.apache.spark.mllib.tree.configuration.Algo._ import org.apache.spark.mllib.tree.configuration.{BoostingStrategy, Strategy} import org.apache.spark.mllib.tree.impurity.Variance import org.apache.spark.mllib.tree.loss.{AbsoluteError, SquaredError, LogLoss} - +import org.apache.spark.mllib.tree.model.GradientBoostedTreesModel import org.apache.spark.mllib.util.MLlibTestSparkContext +import org.apache.spark.util.Utils + /** * Test suite for [[GradientBoostedTrees]]. @@ -35,32 +37,30 @@ class GradientBoostedTreesSuite extends FunSuite with MLlibTestSparkContext { test("Regression with continuous features: SquaredError") { GradientBoostedTreesSuite.testCombinations.foreach { case (numIterations, learningRate, subsamplingRate) => - GradientBoostedTreesSuite.randomSeeds.foreach { randomSeed => - val rdd = sc.parallelize(GradientBoostedTreesSuite.data, 2) - - val treeStrategy = new Strategy(algo = Regression, impurity = Variance, maxDepth = 2, - categoricalFeaturesInfo = Map.empty, subsamplingRate = subsamplingRate) - val boostingStrategy = - new BoostingStrategy(treeStrategy, SquaredError, numIterations, learningRate) - - val gbt = GradientBoostedTrees.train(rdd, boostingStrategy) - - assert(gbt.trees.size === numIterations) - try { - EnsembleTestHelper.validateRegressor(gbt, GradientBoostedTreesSuite.data, 0.06) - } catch { - case e: java.lang.AssertionError => - println(s"FAILED for numIterations=$numIterations, learningRate=$learningRate," + - s" subsamplingRate=$subsamplingRate") - throw e - } - - val remappedInput = rdd.map(x => new LabeledPoint((x.label * 2) - 1, x.features)) - val dt = DecisionTree.train(remappedInput, treeStrategy) - - // Make sure trees are the same. - assert(gbt.trees.head.toString == dt.toString) + val rdd = sc.parallelize(GradientBoostedTreesSuite.data, 2) + + val treeStrategy = new Strategy(algo = Regression, impurity = Variance, maxDepth = 2, + categoricalFeaturesInfo = Map.empty, subsamplingRate = subsamplingRate) + val boostingStrategy = + new BoostingStrategy(treeStrategy, SquaredError, numIterations, learningRate) + + val gbt = GradientBoostedTrees.train(rdd, boostingStrategy) + + assert(gbt.trees.size === numIterations) + try { + EnsembleTestHelper.validateRegressor(gbt, GradientBoostedTreesSuite.data, 0.06) + } catch { + case e: java.lang.AssertionError => + println(s"FAILED for numIterations=$numIterations, learningRate=$learningRate," + + s" subsamplingRate=$subsamplingRate") + throw e } + + val remappedInput = rdd.map(x => new LabeledPoint((x.label * 2) - 1, x.features)) + val dt = DecisionTree.train(remappedInput, treeStrategy) + + // Make sure trees are the same. + assert(gbt.trees.head.toString == dt.toString) } } @@ -133,14 +133,37 @@ class GradientBoostedTreesSuite extends FunSuite with MLlibTestSparkContext { BoostingStrategy.defaultParams(algo) } } + + test("model save/load") { + val tempDir = Utils.createTempDir() + val path = tempDir.toURI.toString + + val trees = Range(0, 3).map(_ => DecisionTreeSuite.createModel(Regression)).toArray + val treeWeights = Array(0.1, 0.3, 1.1) + + Array(Classification, Regression).foreach { algo => + val model = new GradientBoostedTreesModel(algo, trees, treeWeights) + + // Save model, load it back, and compare. + try { + model.save(sc, path) + val sameModel = GradientBoostedTreesModel.load(sc, path) + assert(model.algo == sameModel.algo) + model.trees.zip(sameModel.trees).foreach { case (treeA, treeB) => + DecisionTreeSuite.checkEqual(treeA, treeB) + } + assert(model.treeWeights === sameModel.treeWeights) + } finally { + Utils.deleteRecursively(tempDir) + } + } + } } -object GradientBoostedTreesSuite { +private object GradientBoostedTreesSuite { // Combinations for estimators, learning rates and subsamplingRate val testCombinations = Array((10, 1.0, 1.0), (10, 0.1, 1.0), (10, 0.5, 0.75), (10, 0.1, 0.75)) - val randomSeeds = Array(681283, 4398) - val data = EnsembleTestHelper.generateOrderedLabeledPoints(numFeatures = 10, 100) } diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala index 55e963977b54f..ee3bc98486862 100644 --- a/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala +++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala @@ -27,8 +27,10 @@ import org.apache.spark.mllib.tree.configuration.Algo._ import org.apache.spark.mllib.tree.configuration.Strategy import org.apache.spark.mllib.tree.impl.DecisionTreeMetadata import org.apache.spark.mllib.tree.impurity.{Gini, Variance} -import org.apache.spark.mllib.tree.model.Node +import org.apache.spark.mllib.tree.model.{Node, RandomForestModel} import org.apache.spark.mllib.util.MLlibTestSparkContext +import org.apache.spark.util.Utils + /** * Test suite for [[RandomForest]]. @@ -212,6 +214,26 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext { assert(rf1.toDebugString != rf2.toDebugString) } -} - + test("model save/load") { + val tempDir = Utils.createTempDir() + val path = tempDir.toURI.toString + + Array(Classification, Regression).foreach { algo => + val trees = Range(0, 3).map(_ => DecisionTreeSuite.createModel(algo)).toArray + val model = new RandomForestModel(algo, trees) + + // Save model, load it back, and compare. + try { + model.save(sc, path) + val sameModel = RandomForestModel.load(sc, path) + assert(model.algo == sameModel.algo) + model.trees.zip(sameModel.trees).foreach { case (treeA, treeB) => + DecisionTreeSuite.checkEqual(treeA, treeB) + } + } finally { + Utils.deleteRecursively(tempDir) + } + } + } +} diff --git a/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java b/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java index 625c3257d764e..ef209991804b4 100644 --- a/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java +++ b/network/common/src/main/java/org/apache/spark/network/server/TransportServer.java @@ -100,8 +100,7 @@ protected void initChannel(SocketChannel ch) throws Exception { } }); - channelFuture = bootstrap.bind(new InetSocketAddress(portToBind)); - channelFuture.syncUninterruptibly(); + bindRightPort(portToBind); port = ((InetSocketAddress) channelFuture.channel().localAddress()).getPort(); logger.debug("Shuffle server started on port :" + port); @@ -123,4 +122,37 @@ public void close() { bootstrap = null; } + /** + * Attempt to bind to the specified port up to a fixed number of retries. + * If all attempts fail after the max number of retries, exit. + */ + private void bindRightPort(int portToBind) { + int maxPortRetries = conf.portMaxRetries(); + + for (int i = 0; i <= maxPortRetries; i++) { + int tryPort = -1; + if (0 == portToBind) { + // Do not increment port if tryPort is 0, which is treated as a special port + tryPort = 0; + } else { + // If the new port wraps around, do not try a privilege port + tryPort = ((portToBind + i - 1024) % (65536 - 1024)) + 1024; + } + try { + channelFuture = bootstrap.bind(new InetSocketAddress(tryPort)); + channelFuture.syncUninterruptibly(); + return; + } catch (Exception e) { + logger.warn("Netty service could not bind on port " + tryPort + + ". Attempting the next port."); + if (i >= maxPortRetries) { + logger.error(e.getMessage() + ": Netty server failed after " + + maxPortRetries + " retries."); + + // If it can't find a right port, it should exit directly. + System.exit(-1); + } + } + } + } } diff --git a/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java b/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java index 6c9178688693f..2eaf3b71d9a49 100644 --- a/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java +++ b/network/common/src/main/java/org/apache/spark/network/util/TransportConf.java @@ -98,4 +98,11 @@ public int memoryMapBytes() { public boolean lazyFileDescriptor() { return conf.getBoolean("spark.shuffle.io.lazyFD", true); } + + /** + * Maximum number of retries when binding to a port before giving up. + */ + public int portMaxRetries() { + return conf.getInt("spark.port.maxRetries", 16); + } } diff --git a/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java b/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java index a34aabe9e78a6..63b21222e7b77 100644 --- a/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java +++ b/network/yarn/src/main/java/org/apache/spark/network/yarn/YarnShuffleService.java @@ -76,6 +76,9 @@ public class YarnShuffleService extends AuxiliaryService { // The actual server that serves shuffle files private TransportServer shuffleServer = null; + // Handles registering executors and opening shuffle blocks + private ExternalShuffleBlockHandler blockHandler; + public YarnShuffleService() { super("spark_shuffle"); logger.info("Initializing YARN shuffle service for Spark"); @@ -99,7 +102,8 @@ protected void serviceInit(Configuration conf) { // If authentication is enabled, set up the shuffle server to use a // special RPC handler that filters out unauthenticated fetch requests boolean authEnabled = conf.getBoolean(SPARK_AUTHENTICATE_KEY, DEFAULT_SPARK_AUTHENTICATE); - RpcHandler rpcHandler = new ExternalShuffleBlockHandler(transportConf); + blockHandler = new ExternalShuffleBlockHandler(transportConf); + RpcHandler rpcHandler = blockHandler; if (authEnabled) { secretManager = new ShuffleSecretManager(); rpcHandler = new SaslRpcHandler(rpcHandler, secretManager); @@ -136,6 +140,7 @@ public void stopApplication(ApplicationTerminationContext context) { if (isAuthenticationEnabled()) { secretManager.unregisterApp(appId); } + blockHandler.applicationRemoved(appId, false /* clean up local dirs */); } catch (Exception e) { logger.error("Exception when stopping application {}", appId, e); } diff --git a/pom.xml b/pom.xml index aef450ae63121..56e37d42265c0 100644 --- a/pom.xml +++ b/pom.xml @@ -154,6 +154,7 @@ org.scala-lang 3.6.3 1.8.8 + 2.4.4 1.1.1.6 @@ -394,7 +395,7 @@ provided - + org.apache.commons commons-lang3 @@ -577,6 +578,16 @@ metrics-graphite ${codahale.metrics.version} + + com.fasterxml.jackson.core + jackson-databind + ${fasterxml.jackson.version} + + + com.fasterxml.jackson.module + jackson-module-scala_2.10 + ${fasterxml.jackson.version} + org.scala-lang scala-compiler @@ -928,6 +939,16 @@ ${codehaus.jackson.version} ${hadoop.deps.scope} + + org.codehaus.jackson + jackson-xc + ${codehaus.jackson.version} + + + org.codehaus.jackson + jackson-jaxrs + ${codehaus.jackson.version} + ${hive.group} hive-beeline @@ -954,6 +975,10 @@ com.esotericsoftware.kryo kryo + + org.apache.avro + avro-mapred + @@ -1153,13 +1178,19 @@ ${project.build.directory}/surefire-reports -Xmx3g -XX:MaxPermSize=${MaxPermGen} -XX:ReservedCodeCacheSize=512m + + + ${test_classpath} + true ${session.executionRootDirectory} 1 false false - ${test_classpath} true false @@ -1538,6 +1569,7 @@ 2.5.0 0.98.7-hadoop2 hadoop2 + 1.9.13 @@ -1546,10 +1578,11 @@ 2.3.0 2.5.0 - 0.9.0 + 0.9.3 0.98.7-hadoop2 3.1.1 hadoop2 + 1.9.13 @@ -1558,10 +1591,11 @@ 2.4.0 2.5.0 - 0.9.0 + 0.9.3 0.98.7-hadoop2 3.1.1 hadoop2 + 1.9.13 diff --git a/project/SparkBuild.scala b/project/SparkBuild.scala index 5e3051c091030..8fb1239b4a96b 100644 --- a/project/SparkBuild.scala +++ b/project/SparkBuild.scala @@ -177,6 +177,29 @@ object SparkBuild extends PomBuild { enable(Flume.settings)(streamingFlumeSink) + + /** + * Adds the ability to run the spark shell directly from SBT without building an assembly + * jar. + * + * Usage: `build/sbt sparkShell` + */ + val sparkShell = taskKey[Unit]("start a spark-shell.") + + enable(Seq( + connectInput in run := true, + fork := true, + outputStrategy in run := Some (StdoutOutput), + + javaOptions ++= Seq("-Xmx2G", "-XX:MaxPermSize=1g"), + + sparkShell := { + (runMain in Compile).toTask(" org.apache.spark.repl.Main -usejavacp").value + } + ))(assembly) + + enable(Seq(sparkShell := sparkShell in "assembly"))(spark) + // TODO: move this to its upstream project. override def projectDefinitions(baseDirectory: File): Seq[Project] = { super.projectDefinitions(baseDirectory).map { x => @@ -388,6 +411,10 @@ object TestSettings { lazy val settings = Seq ( // Fork new JVMs for tests and set Java options for those fork := true, + // Setting SPARK_DIST_CLASSPATH is a simple way to make sure any child processes + // launched by the tests have access to the correct test-time classpath. + envVars in Test += ("SPARK_DIST_CLASSPATH" -> + (fullClasspath in Test).value.files.map(_.getAbsolutePath).mkString(":").stripSuffix(":")), javaOptions in Test += "-Dspark.test.home=" + sparkHome, javaOptions in Test += "-Dspark.testing=1", javaOptions in Test += "-Dspark.port.maxRetries=100", @@ -400,10 +427,6 @@ object TestSettings { javaOptions in Test += "-ea", javaOptions in Test ++= "-Xmx3g -XX:PermSize=128M -XX:MaxNewSize=256m -XX:MaxPermSize=1g" .split(" ").toSeq, - // This places test scope jars on the classpath of executors during tests. - javaOptions in Test += - "-Dspark.executor.extraClassPath=" + (fullClasspath in Test).value.files. - map(_.getAbsolutePath).mkString(":").stripSuffix(":"), javaOptions += "-Xmx3g", // Show full stack trace and duration in test cases. testOptions in Test += Tests.Argument("-oDF"), diff --git a/python/docs/pyspark.ml.rst b/python/docs/pyspark.ml.rst index f10d1339a9a8f..4da6d4a74a299 100644 --- a/python/docs/pyspark.ml.rst +++ b/python/docs/pyspark.ml.rst @@ -1,11 +1,8 @@ pyspark.ml package ===================== -Submodules ----------- - -pyspark.ml module ------------------ +Module Context +-------------- .. automodule:: pyspark.ml :members: diff --git a/python/docs/pyspark.mllib.rst b/python/docs/pyspark.mllib.rst index 4548b8739ed91..21f66ca344a3c 100644 --- a/python/docs/pyspark.mllib.rst +++ b/python/docs/pyspark.mllib.rst @@ -1,9 +1,6 @@ pyspark.mllib package ===================== -Submodules ----------- - pyspark.mllib.classification module ----------------------------------- diff --git a/python/docs/pyspark.sql.rst b/python/docs/pyspark.sql.rst index 65b3650ae10ab..80c6f02a9df41 100644 --- a/python/docs/pyspark.sql.rst +++ b/python/docs/pyspark.sql.rst @@ -1,10 +1,18 @@ pyspark.sql module ================== -Module contents ---------------- +Module Context +-------------- .. automodule:: pyspark.sql :members: :undoc-members: :show-inheritance: + + +pyspark.sql.types module +------------------------ +.. automodule:: pyspark.sql.types + :members: + :undoc-members: + :show-inheritance: diff --git a/python/pyspark/mllib/linalg.py b/python/pyspark/mllib/linalg.py index 7f21190ed8c25..597012b1c967c 100644 --- a/python/pyspark/mllib/linalg.py +++ b/python/pyspark/mllib/linalg.py @@ -29,7 +29,7 @@ import numpy as np -from pyspark.sql import UserDefinedType, StructField, StructType, ArrayType, DoubleType, \ +from pyspark.sql.types import UserDefinedType, StructField, StructType, ArrayType, DoubleType, \ IntegerType, ByteType diff --git a/python/pyspark/rdd.py b/python/pyspark/rdd.py index 6e029bf7f13fc..bd4f16e058045 100644 --- a/python/pyspark/rdd.py +++ b/python/pyspark/rdd.py @@ -1366,10 +1366,14 @@ def saveAsPickleFile(self, path, batchSize=10): ser = BatchedSerializer(PickleSerializer(), batchSize) self._reserialize(ser)._jrdd.saveAsObjectFile(path) - def saveAsTextFile(self, path): + def saveAsTextFile(self, path, compressionCodecClass=None): """ Save this RDD as a text file, using string representations of elements. + @param path: path to text file + @param compressionCodecClass: (None by default) string i.e. + "org.apache.hadoop.io.compress.GzipCodec" + >>> tempFile = NamedTemporaryFile(delete=True) >>> tempFile.close() >>> sc.parallelize(range(10)).saveAsTextFile(tempFile.name) @@ -1385,6 +1389,16 @@ def saveAsTextFile(self, path): >>> sc.parallelize(['', 'foo', '', 'bar', '']).saveAsTextFile(tempFile2.name) >>> ''.join(sorted(input(glob(tempFile2.name + "/part-0000*")))) '\\n\\n\\nbar\\nfoo\\n' + + Using compressionCodecClass + + >>> tempFile3 = NamedTemporaryFile(delete=True) + >>> tempFile3.close() + >>> codec = "org.apache.hadoop.io.compress.GzipCodec" + >>> sc.parallelize(['foo', 'bar']).saveAsTextFile(tempFile3.name, codec) + >>> from fileinput import input, hook_compressed + >>> ''.join(sorted(input(glob(tempFile3.name + "/part*.gz"), openhook=hook_compressed))) + 'bar\\nfoo\\n' """ def func(split, iterator): for x in iterator: @@ -1395,7 +1409,11 @@ def func(split, iterator): yield x keyed = self.mapPartitionsWithIndex(func) keyed._bypass_serializer = True - keyed._jrdd.map(self.ctx._jvm.BytesToString()).saveAsTextFile(path) + if compressionCodecClass: + compressionCodec = self.ctx._jvm.java.lang.Class.forName(compressionCodecClass) + keyed._jrdd.map(self.ctx._jvm.BytesToString()).saveAsTextFile(path, compressionCodec) + else: + keyed._jrdd.map(self.ctx._jvm.BytesToString()).saveAsTextFile(path) # Pair functions diff --git a/python/pyspark/sql.py b/python/pyspark/sql.py deleted file mode 100644 index e55f285a778c4..0000000000000 --- a/python/pyspark/sql.py +++ /dev/null @@ -1,2711 +0,0 @@ -# -# Licensed to the Apache Software Foundation (ASF) under one or more -# contributor license agreements. See the NOTICE file distributed with -# this work for additional information regarding copyright ownership. -# The ASF licenses this file to You under the Apache License, Version 2.0 -# (the "License"); you may not use this file except in compliance with -# the License. You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -# - -""" -public classes of Spark SQL: - - - L{SQLContext} - Main entry point for SQL functionality. - - L{DataFrame} - A Resilient Distributed Dataset (RDD) with Schema information for the data contained. In - addition to normal RDD operations, DataFrames also support SQL. - - L{GroupedData} - - L{Column} - Column is a DataFrame with a single column. - - L{Row} - A Row of data returned by a Spark SQL query. - - L{HiveContext} - Main entry point for accessing data stored in Apache Hive.. -""" - -import sys -import itertools -import decimal -import datetime -import keyword -import warnings -import json -import re -import random -import os -from tempfile import NamedTemporaryFile -from array import array -from operator import itemgetter -from itertools import imap - -from py4j.protocol import Py4JError -from py4j.java_collections import ListConverter, MapConverter - -from pyspark.context import SparkContext -from pyspark.rdd import RDD, _prepare_for_python_RDD -from pyspark.serializers import BatchedSerializer, AutoBatchedSerializer, PickleSerializer, \ - CloudPickleSerializer, UTF8Deserializer -from pyspark.storagelevel import StorageLevel -from pyspark.traceback_utils import SCCallSiteSync - - -__all__ = [ - "StringType", "BinaryType", "BooleanType", "DateType", "TimestampType", "DecimalType", - "DoubleType", "FloatType", "ByteType", "IntegerType", "LongType", - "ShortType", "ArrayType", "MapType", "StructField", "StructType", - "SQLContext", "HiveContext", "DataFrame", "GroupedData", "Column", "Row", "Dsl", - "SchemaRDD"] - - -class DataType(object): - - """Spark SQL DataType""" - - def __repr__(self): - return self.__class__.__name__ - - def __hash__(self): - return hash(str(self)) - - def __eq__(self, other): - return (isinstance(other, self.__class__) and - self.__dict__ == other.__dict__) - - def __ne__(self, other): - return not self.__eq__(other) - - @classmethod - def typeName(cls): - return cls.__name__[:-4].lower() - - def jsonValue(self): - return self.typeName() - - def json(self): - return json.dumps(self.jsonValue(), - separators=(',', ':'), - sort_keys=True) - - -class PrimitiveTypeSingleton(type): - - """Metaclass for PrimitiveType""" - - _instances = {} - - def __call__(cls): - if cls not in cls._instances: - cls._instances[cls] = super(PrimitiveTypeSingleton, cls).__call__() - return cls._instances[cls] - - -class PrimitiveType(DataType): - - """Spark SQL PrimitiveType""" - - __metaclass__ = PrimitiveTypeSingleton - - def __eq__(self, other): - # because they should be the same object - return self is other - - -class NullType(PrimitiveType): - - """Spark SQL NullType - - The data type representing None, used for the types which has not - been inferred. - """ - - -class StringType(PrimitiveType): - - """Spark SQL StringType - - The data type representing string values. - """ - - -class BinaryType(PrimitiveType): - - """Spark SQL BinaryType - - The data type representing bytearray values. - """ - - -class BooleanType(PrimitiveType): - - """Spark SQL BooleanType - - The data type representing bool values. - """ - - -class DateType(PrimitiveType): - - """Spark SQL DateType - - The data type representing datetime.date values. - """ - - -class TimestampType(PrimitiveType): - - """Spark SQL TimestampType - - The data type representing datetime.datetime values. - """ - - -class DecimalType(DataType): - - """Spark SQL DecimalType - - The data type representing decimal.Decimal values. - """ - - def __init__(self, precision=None, scale=None): - self.precision = precision - self.scale = scale - self.hasPrecisionInfo = precision is not None - - def jsonValue(self): - if self.hasPrecisionInfo: - return "decimal(%d,%d)" % (self.precision, self.scale) - else: - return "decimal" - - def __repr__(self): - if self.hasPrecisionInfo: - return "DecimalType(%d,%d)" % (self.precision, self.scale) - else: - return "DecimalType()" - - -class DoubleType(PrimitiveType): - - """Spark SQL DoubleType - - The data type representing float values. - """ - - -class FloatType(PrimitiveType): - - """Spark SQL FloatType - - The data type representing single precision floating-point values. - """ - - -class ByteType(PrimitiveType): - - """Spark SQL ByteType - - The data type representing int values with 1 singed byte. - """ - - -class IntegerType(PrimitiveType): - - """Spark SQL IntegerType - - The data type representing int values. - """ - - -class LongType(PrimitiveType): - - """Spark SQL LongType - - The data type representing long values. If the any value is - beyond the range of [-9223372036854775808, 9223372036854775807], - please use DecimalType. - """ - - -class ShortType(PrimitiveType): - - """Spark SQL ShortType - - The data type representing int values with 2 signed bytes. - """ - - -class ArrayType(DataType): - - """Spark SQL ArrayType - - The data type representing list values. An ArrayType object - comprises two fields, elementType (a DataType) and containsNull (a bool). - The field of elementType is used to specify the type of array elements. - The field of containsNull is used to specify if the array has None values. - - """ - - def __init__(self, elementType, containsNull=True): - """Creates an ArrayType - - :param elementType: the data type of elements. - :param containsNull: indicates whether the list contains None values. - - >>> ArrayType(StringType) == ArrayType(StringType, True) - True - >>> ArrayType(StringType, False) == ArrayType(StringType) - False - """ - self.elementType = elementType - self.containsNull = containsNull - - def __repr__(self): - return "ArrayType(%s,%s)" % (self.elementType, - str(self.containsNull).lower()) - - def jsonValue(self): - return {"type": self.typeName(), - "elementType": self.elementType.jsonValue(), - "containsNull": self.containsNull} - - @classmethod - def fromJson(cls, json): - return ArrayType(_parse_datatype_json_value(json["elementType"]), - json["containsNull"]) - - -class MapType(DataType): - - """Spark SQL MapType - - The data type representing dict values. A MapType object comprises - three fields, keyType (a DataType), valueType (a DataType) and - valueContainsNull (a bool). - - The field of keyType is used to specify the type of keys in the map. - The field of valueType is used to specify the type of values in the map. - The field of valueContainsNull is used to specify if values of this - map has None values. - - For values of a MapType column, keys are not allowed to have None values. - - """ - - def __init__(self, keyType, valueType, valueContainsNull=True): - """Creates a MapType - :param keyType: the data type of keys. - :param valueType: the data type of values. - :param valueContainsNull: indicates whether values contains - null values. - - >>> (MapType(StringType, IntegerType) - ... == MapType(StringType, IntegerType, True)) - True - >>> (MapType(StringType, IntegerType, False) - ... == MapType(StringType, FloatType)) - False - """ - self.keyType = keyType - self.valueType = valueType - self.valueContainsNull = valueContainsNull - - def __repr__(self): - return "MapType(%s,%s,%s)" % (self.keyType, self.valueType, - str(self.valueContainsNull).lower()) - - def jsonValue(self): - return {"type": self.typeName(), - "keyType": self.keyType.jsonValue(), - "valueType": self.valueType.jsonValue(), - "valueContainsNull": self.valueContainsNull} - - @classmethod - def fromJson(cls, json): - return MapType(_parse_datatype_json_value(json["keyType"]), - _parse_datatype_json_value(json["valueType"]), - json["valueContainsNull"]) - - -class StructField(DataType): - - """Spark SQL StructField - - Represents a field in a StructType. - A StructField object comprises three fields, name (a string), - dataType (a DataType) and nullable (a bool). The field of name - is the name of a StructField. The field of dataType specifies - the data type of a StructField. - - The field of nullable specifies if values of a StructField can - contain None values. - - """ - - def __init__(self, name, dataType, nullable=True, metadata=None): - """Creates a StructField - :param name: the name of this field. - :param dataType: the data type of this field. - :param nullable: indicates whether values of this field - can be null. - :param metadata: metadata of this field, which is a map from string - to simple type that can be serialized to JSON - automatically - - >>> (StructField("f1", StringType, True) - ... == StructField("f1", StringType, True)) - True - >>> (StructField("f1", StringType, True) - ... == StructField("f2", StringType, True)) - False - """ - self.name = name - self.dataType = dataType - self.nullable = nullable - self.metadata = metadata or {} - - def __repr__(self): - return "StructField(%s,%s,%s)" % (self.name, self.dataType, - str(self.nullable).lower()) - - def jsonValue(self): - return {"name": self.name, - "type": self.dataType.jsonValue(), - "nullable": self.nullable, - "metadata": self.metadata} - - @classmethod - def fromJson(cls, json): - return StructField(json["name"], - _parse_datatype_json_value(json["type"]), - json["nullable"], - json["metadata"]) - - -class StructType(DataType): - - """Spark SQL StructType - - The data type representing rows. - A StructType object comprises a list of L{StructField}. - - """ - - def __init__(self, fields): - """Creates a StructType - - >>> struct1 = StructType([StructField("f1", StringType, True)]) - >>> struct2 = StructType([StructField("f1", StringType, True)]) - >>> struct1 == struct2 - True - >>> struct1 = StructType([StructField("f1", StringType, True)]) - >>> struct2 = StructType([StructField("f1", StringType, True), - ... [StructField("f2", IntegerType, False)]]) - >>> struct1 == struct2 - False - """ - self.fields = fields - - def __repr__(self): - return ("StructType(List(%s))" % - ",".join(str(field) for field in self.fields)) - - def jsonValue(self): - return {"type": self.typeName(), - "fields": [f.jsonValue() for f in self.fields]} - - @classmethod - def fromJson(cls, json): - return StructType([StructField.fromJson(f) for f in json["fields"]]) - - -class UserDefinedType(DataType): - """ - .. note:: WARN: Spark Internal Use Only - SQL User-Defined Type (UDT). - """ - - @classmethod - def typeName(cls): - return cls.__name__.lower() - - @classmethod - def sqlType(cls): - """ - Underlying SQL storage type for this UDT. - """ - raise NotImplementedError("UDT must implement sqlType().") - - @classmethod - def module(cls): - """ - The Python module of the UDT. - """ - raise NotImplementedError("UDT must implement module().") - - @classmethod - def scalaUDT(cls): - """ - The class name of the paired Scala UDT. - """ - raise NotImplementedError("UDT must have a paired Scala UDT.") - - def serialize(self, obj): - """ - Converts the a user-type object into a SQL datum. - """ - raise NotImplementedError("UDT must implement serialize().") - - def deserialize(self, datum): - """ - Converts a SQL datum into a user-type object. - """ - raise NotImplementedError("UDT must implement deserialize().") - - def json(self): - return json.dumps(self.jsonValue(), separators=(',', ':'), sort_keys=True) - - def jsonValue(self): - schema = { - "type": "udt", - "class": self.scalaUDT(), - "pyClass": "%s.%s" % (self.module(), type(self).__name__), - "sqlType": self.sqlType().jsonValue() - } - return schema - - @classmethod - def fromJson(cls, json): - pyUDT = json["pyClass"] - split = pyUDT.rfind(".") - pyModule = pyUDT[:split] - pyClass = pyUDT[split+1:] - m = __import__(pyModule, globals(), locals(), [pyClass], -1) - UDT = getattr(m, pyClass) - return UDT() - - def __eq__(self, other): - return type(self) == type(other) - - -_all_primitive_types = dict((v.typeName(), v) - for v in globals().itervalues() - if type(v) is PrimitiveTypeSingleton and - v.__base__ == PrimitiveType) - - -_all_complex_types = dict((v.typeName(), v) - for v in [ArrayType, MapType, StructType]) - - -def _parse_datatype_json_string(json_string): - """Parses the given data type JSON string. - >>> def check_datatype(datatype): - ... scala_datatype = sqlCtx._ssql_ctx.parseDataType(datatype.json()) - ... python_datatype = _parse_datatype_json_string(scala_datatype.json()) - ... return datatype == python_datatype - >>> all(check_datatype(cls()) for cls in _all_primitive_types.values()) - True - >>> # Simple ArrayType. - >>> simple_arraytype = ArrayType(StringType(), True) - >>> check_datatype(simple_arraytype) - True - >>> # Simple MapType. - >>> simple_maptype = MapType(StringType(), LongType()) - >>> check_datatype(simple_maptype) - True - >>> # Simple StructType. - >>> simple_structtype = StructType([ - ... StructField("a", DecimalType(), False), - ... StructField("b", BooleanType(), True), - ... StructField("c", LongType(), True), - ... StructField("d", BinaryType(), False)]) - >>> check_datatype(simple_structtype) - True - >>> # Complex StructType. - >>> complex_structtype = StructType([ - ... StructField("simpleArray", simple_arraytype, True), - ... StructField("simpleMap", simple_maptype, True), - ... StructField("simpleStruct", simple_structtype, True), - ... StructField("boolean", BooleanType(), False), - ... StructField("withMeta", DoubleType(), False, {"name": "age"})]) - >>> check_datatype(complex_structtype) - True - >>> # Complex ArrayType. - >>> complex_arraytype = ArrayType(complex_structtype, True) - >>> check_datatype(complex_arraytype) - True - >>> # Complex MapType. - >>> complex_maptype = MapType(complex_structtype, - ... complex_arraytype, False) - >>> check_datatype(complex_maptype) - True - >>> check_datatype(ExamplePointUDT()) - True - >>> structtype_with_udt = StructType([StructField("label", DoubleType(), False), - ... StructField("point", ExamplePointUDT(), False)]) - >>> check_datatype(structtype_with_udt) - True - """ - return _parse_datatype_json_value(json.loads(json_string)) - - -_FIXED_DECIMAL = re.compile("decimal\\((\\d+),(\\d+)\\)") - - -def _parse_datatype_json_value(json_value): - if type(json_value) is unicode: - if json_value in _all_primitive_types.keys(): - return _all_primitive_types[json_value]() - elif json_value == u'decimal': - return DecimalType() - elif _FIXED_DECIMAL.match(json_value): - m = _FIXED_DECIMAL.match(json_value) - return DecimalType(int(m.group(1)), int(m.group(2))) - else: - raise ValueError("Could not parse datatype: %s" % json_value) - else: - tpe = json_value["type"] - if tpe in _all_complex_types: - return _all_complex_types[tpe].fromJson(json_value) - elif tpe == 'udt': - return UserDefinedType.fromJson(json_value) - else: - raise ValueError("not supported type: %s" % tpe) - - -# Mapping Python types to Spark SQL DataType -_type_mappings = { - type(None): NullType, - bool: BooleanType, - int: IntegerType, - long: LongType, - float: DoubleType, - str: StringType, - unicode: StringType, - bytearray: BinaryType, - decimal.Decimal: DecimalType, - datetime.date: DateType, - datetime.datetime: TimestampType, - datetime.time: TimestampType, -} - - -def _infer_type(obj): - """Infer the DataType from obj - - >>> p = ExamplePoint(1.0, 2.0) - >>> _infer_type(p) - ExamplePointUDT - """ - if obj is None: - raise ValueError("Can not infer type for None") - - if hasattr(obj, '__UDT__'): - return obj.__UDT__ - - dataType = _type_mappings.get(type(obj)) - if dataType is not None: - return dataType() - - if isinstance(obj, dict): - for key, value in obj.iteritems(): - if key is not None and value is not None: - return MapType(_infer_type(key), _infer_type(value), True) - else: - return MapType(NullType(), NullType(), True) - elif isinstance(obj, (list, array)): - for v in obj: - if v is not None: - return ArrayType(_infer_type(obj[0]), True) - else: - return ArrayType(NullType(), True) - else: - try: - return _infer_schema(obj) - except ValueError: - raise ValueError("not supported type: %s" % type(obj)) - - -def _infer_schema(row): - """Infer the schema from dict/namedtuple/object""" - if isinstance(row, dict): - items = sorted(row.items()) - - elif isinstance(row, tuple): - if hasattr(row, "_fields"): # namedtuple - items = zip(row._fields, tuple(row)) - elif hasattr(row, "__FIELDS__"): # Row - items = zip(row.__FIELDS__, tuple(row)) - elif all(isinstance(x, tuple) and len(x) == 2 for x in row): - items = row - else: - raise ValueError("Can't infer schema from tuple") - - elif hasattr(row, "__dict__"): # object - items = sorted(row.__dict__.items()) - - else: - raise ValueError("Can not infer schema for type: %s" % type(row)) - - fields = [StructField(k, _infer_type(v), True) for k, v in items] - return StructType(fields) - - -def _need_python_to_sql_conversion(dataType): - """ - Checks whether we need python to sql conversion for the given type. - For now, only UDTs need this conversion. - - >>> _need_python_to_sql_conversion(DoubleType()) - False - >>> schema0 = StructType([StructField("indices", ArrayType(IntegerType(), False), False), - ... StructField("values", ArrayType(DoubleType(), False), False)]) - >>> _need_python_to_sql_conversion(schema0) - False - >>> _need_python_to_sql_conversion(ExamplePointUDT()) - True - >>> schema1 = ArrayType(ExamplePointUDT(), False) - >>> _need_python_to_sql_conversion(schema1) - True - >>> schema2 = StructType([StructField("label", DoubleType(), False), - ... StructField("point", ExamplePointUDT(), False)]) - >>> _need_python_to_sql_conversion(schema2) - True - """ - if isinstance(dataType, StructType): - return any([_need_python_to_sql_conversion(f.dataType) for f in dataType.fields]) - elif isinstance(dataType, ArrayType): - return _need_python_to_sql_conversion(dataType.elementType) - elif isinstance(dataType, MapType): - return _need_python_to_sql_conversion(dataType.keyType) or \ - _need_python_to_sql_conversion(dataType.valueType) - elif isinstance(dataType, UserDefinedType): - return True - else: - return False - - -def _python_to_sql_converter(dataType): - """ - Returns a converter that converts a Python object into a SQL datum for the given type. - - >>> conv = _python_to_sql_converter(DoubleType()) - >>> conv(1.0) - 1.0 - >>> conv = _python_to_sql_converter(ArrayType(DoubleType(), False)) - >>> conv([1.0, 2.0]) - [1.0, 2.0] - >>> conv = _python_to_sql_converter(ExamplePointUDT()) - >>> conv(ExamplePoint(1.0, 2.0)) - [1.0, 2.0] - >>> schema = StructType([StructField("label", DoubleType(), False), - ... StructField("point", ExamplePointUDT(), False)]) - >>> conv = _python_to_sql_converter(schema) - >>> conv((1.0, ExamplePoint(1.0, 2.0))) - (1.0, [1.0, 2.0]) - """ - if not _need_python_to_sql_conversion(dataType): - return lambda x: x - - if isinstance(dataType, StructType): - names, types = zip(*[(f.name, f.dataType) for f in dataType.fields]) - converters = map(_python_to_sql_converter, types) - - def converter(obj): - if isinstance(obj, dict): - return tuple(c(obj.get(n)) for n, c in zip(names, converters)) - elif isinstance(obj, tuple): - if hasattr(obj, "_fields") or hasattr(obj, "__FIELDS__"): - return tuple(c(v) for c, v in zip(converters, obj)) - elif all(isinstance(x, tuple) and len(x) == 2 for x in obj): # k-v pairs - d = dict(obj) - return tuple(c(d.get(n)) for n, c in zip(names, converters)) - else: - return tuple(c(v) for c, v in zip(converters, obj)) - else: - raise ValueError("Unexpected tuple %r with type %r" % (obj, dataType)) - return converter - elif isinstance(dataType, ArrayType): - element_converter = _python_to_sql_converter(dataType.elementType) - return lambda a: [element_converter(v) for v in a] - elif isinstance(dataType, MapType): - key_converter = _python_to_sql_converter(dataType.keyType) - value_converter = _python_to_sql_converter(dataType.valueType) - return lambda m: dict([(key_converter(k), value_converter(v)) for k, v in m.items()]) - elif isinstance(dataType, UserDefinedType): - return lambda obj: dataType.serialize(obj) - else: - raise ValueError("Unexpected type %r" % dataType) - - -def _has_nulltype(dt): - """ Return whether there is NullType in `dt` or not """ - if isinstance(dt, StructType): - return any(_has_nulltype(f.dataType) for f in dt.fields) - elif isinstance(dt, ArrayType): - return _has_nulltype((dt.elementType)) - elif isinstance(dt, MapType): - return _has_nulltype(dt.keyType) or _has_nulltype(dt.valueType) - else: - return isinstance(dt, NullType) - - -def _merge_type(a, b): - if isinstance(a, NullType): - return b - elif isinstance(b, NullType): - return a - elif type(a) is not type(b): - # TODO: type cast (such as int -> long) - raise TypeError("Can not merge type %s and %s" % (a, b)) - - # same type - if isinstance(a, StructType): - nfs = dict((f.name, f.dataType) for f in b.fields) - fields = [StructField(f.name, _merge_type(f.dataType, nfs.get(f.name, NullType()))) - for f in a.fields] - names = set([f.name for f in fields]) - for n in nfs: - if n not in names: - fields.append(StructField(n, nfs[n])) - return StructType(fields) - - elif isinstance(a, ArrayType): - return ArrayType(_merge_type(a.elementType, b.elementType), True) - - elif isinstance(a, MapType): - return MapType(_merge_type(a.keyType, b.keyType), - _merge_type(a.valueType, b.valueType), - True) - else: - return a - - -def _create_converter(dataType): - """Create an converter to drop the names of fields in obj """ - if isinstance(dataType, ArrayType): - conv = _create_converter(dataType.elementType) - return lambda row: map(conv, row) - - elif isinstance(dataType, MapType): - kconv = _create_converter(dataType.keyType) - vconv = _create_converter(dataType.valueType) - return lambda row: dict((kconv(k), vconv(v)) for k, v in row.iteritems()) - - elif isinstance(dataType, NullType): - return lambda x: None - - elif not isinstance(dataType, StructType): - return lambda x: x - - # dataType must be StructType - names = [f.name for f in dataType.fields] - converters = [_create_converter(f.dataType) for f in dataType.fields] - - def convert_struct(obj): - if obj is None: - return - - if isinstance(obj, tuple): - if hasattr(obj, "_fields"): - d = dict(zip(obj._fields, obj)) - elif hasattr(obj, "__FIELDS__"): - d = dict(zip(obj.__FIELDS__, obj)) - elif all(isinstance(x, tuple) and len(x) == 2 for x in obj): - d = dict(obj) - else: - raise ValueError("unexpected tuple: %s" % str(obj)) - - elif isinstance(obj, dict): - d = obj - elif hasattr(obj, "__dict__"): # object - d = obj.__dict__ - else: - raise ValueError("Unexpected obj: %s" % obj) - - return tuple([conv(d.get(name)) for name, conv in zip(names, converters)]) - - return convert_struct - - -_BRACKETS = {'(': ')', '[': ']', '{': '}'} - - -def _split_schema_abstract(s): - """ - split the schema abstract into fields - - >>> _split_schema_abstract("a b c") - ['a', 'b', 'c'] - >>> _split_schema_abstract("a(a b)") - ['a(a b)'] - >>> _split_schema_abstract("a b[] c{a b}") - ['a', 'b[]', 'c{a b}'] - >>> _split_schema_abstract(" ") - [] - """ - - r = [] - w = '' - brackets = [] - for c in s: - if c == ' ' and not brackets: - if w: - r.append(w) - w = '' - else: - w += c - if c in _BRACKETS: - brackets.append(c) - elif c in _BRACKETS.values(): - if not brackets or c != _BRACKETS[brackets.pop()]: - raise ValueError("unexpected " + c) - - if brackets: - raise ValueError("brackets not closed: %s" % brackets) - if w: - r.append(w) - return r - - -def _parse_field_abstract(s): - """ - Parse a field in schema abstract - - >>> _parse_field_abstract("a") - StructField(a,None,true) - >>> _parse_field_abstract("b(c d)") - StructField(b,StructType(...c,None,true),StructField(d... - >>> _parse_field_abstract("a[]") - StructField(a,ArrayType(None,true),true) - >>> _parse_field_abstract("a{[]}") - StructField(a,MapType(None,ArrayType(None,true),true),true) - """ - if set(_BRACKETS.keys()) & set(s): - idx = min((s.index(c) for c in _BRACKETS if c in s)) - name = s[:idx] - return StructField(name, _parse_schema_abstract(s[idx:]), True) - else: - return StructField(s, None, True) - - -def _parse_schema_abstract(s): - """ - parse abstract into schema - - >>> _parse_schema_abstract("a b c") - StructType...a...b...c... - >>> _parse_schema_abstract("a[b c] b{}") - StructType...a,ArrayType...b...c...b,MapType... - >>> _parse_schema_abstract("c{} d{a b}") - StructType...c,MapType...d,MapType...a...b... - >>> _parse_schema_abstract("a b(t)").fields[1] - StructField(b,StructType(List(StructField(t,None,true))),true) - """ - s = s.strip() - if not s: - return - - elif s.startswith('('): - return _parse_schema_abstract(s[1:-1]) - - elif s.startswith('['): - return ArrayType(_parse_schema_abstract(s[1:-1]), True) - - elif s.startswith('{'): - return MapType(None, _parse_schema_abstract(s[1:-1])) - - parts = _split_schema_abstract(s) - fields = [_parse_field_abstract(p) for p in parts] - return StructType(fields) - - -def _infer_schema_type(obj, dataType): - """ - Fill the dataType with types inferred from obj - - >>> schema = _parse_schema_abstract("a b c d") - >>> row = (1, 1.0, "str", datetime.date(2014, 10, 10)) - >>> _infer_schema_type(row, schema) - StructType...IntegerType...DoubleType...StringType...DateType... - >>> row = [[1], {"key": (1, 2.0)}] - >>> schema = _parse_schema_abstract("a[] b{c d}") - >>> _infer_schema_type(row, schema) - StructType...a,ArrayType...b,MapType(StringType,...c,IntegerType... - """ - if dataType is None: - return _infer_type(obj) - - if not obj: - return NullType() - - if isinstance(dataType, ArrayType): - eType = _infer_schema_type(obj[0], dataType.elementType) - return ArrayType(eType, True) - - elif isinstance(dataType, MapType): - k, v = obj.iteritems().next() - return MapType(_infer_schema_type(k, dataType.keyType), - _infer_schema_type(v, dataType.valueType)) - - elif isinstance(dataType, StructType): - fs = dataType.fields - assert len(fs) == len(obj), \ - "Obj(%s) have different length with fields(%s)" % (obj, fs) - fields = [StructField(f.name, _infer_schema_type(o, f.dataType), True) - for o, f in zip(obj, fs)] - return StructType(fields) - - else: - raise ValueError("Unexpected dataType: %s" % dataType) - - -_acceptable_types = { - BooleanType: (bool,), - ByteType: (int, long), - ShortType: (int, long), - IntegerType: (int, long), - LongType: (int, long), - FloatType: (float,), - DoubleType: (float,), - DecimalType: (decimal.Decimal,), - StringType: (str, unicode), - BinaryType: (bytearray,), - DateType: (datetime.date,), - TimestampType: (datetime.datetime,), - ArrayType: (list, tuple, array), - MapType: (dict,), - StructType: (tuple, list), -} - - -def _verify_type(obj, dataType): - """ - Verify the type of obj against dataType, raise an exception if - they do not match. - - >>> _verify_type(None, StructType([])) - >>> _verify_type("", StringType()) - >>> _verify_type(0, IntegerType()) - >>> _verify_type(range(3), ArrayType(ShortType())) - >>> _verify_type(set(), ArrayType(StringType())) # doctest: +IGNORE_EXCEPTION_DETAIL - Traceback (most recent call last): - ... - TypeError:... - >>> _verify_type({}, MapType(StringType(), IntegerType())) - >>> _verify_type((), StructType([])) - >>> _verify_type([], StructType([])) - >>> _verify_type([1], StructType([])) # doctest: +IGNORE_EXCEPTION_DETAIL - Traceback (most recent call last): - ... - ValueError:... - >>> _verify_type(ExamplePoint(1.0, 2.0), ExamplePointUDT()) - >>> _verify_type([1.0, 2.0], ExamplePointUDT()) # doctest: +IGNORE_EXCEPTION_DETAIL - Traceback (most recent call last): - ... - ValueError:... - """ - # all objects are nullable - if obj is None: - return - - if isinstance(dataType, UserDefinedType): - if not (hasattr(obj, '__UDT__') and obj.__UDT__ == dataType): - raise ValueError("%r is not an instance of type %r" % (obj, dataType)) - _verify_type(dataType.serialize(obj), dataType.sqlType()) - return - - _type = type(dataType) - assert _type in _acceptable_types, "unkown datatype: %s" % dataType - - # subclass of them can not be deserialized in JVM - if type(obj) not in _acceptable_types[_type]: - raise TypeError("%s can not accept object in type %s" - % (dataType, type(obj))) - - if isinstance(dataType, ArrayType): - for i in obj: - _verify_type(i, dataType.elementType) - - elif isinstance(dataType, MapType): - for k, v in obj.iteritems(): - _verify_type(k, dataType.keyType) - _verify_type(v, dataType.valueType) - - elif isinstance(dataType, StructType): - if len(obj) != len(dataType.fields): - raise ValueError("Length of object (%d) does not match with" - "length of fields (%d)" % (len(obj), len(dataType.fields))) - for v, f in zip(obj, dataType.fields): - _verify_type(v, f.dataType) - - -_cached_cls = {} - - -def _restore_object(dataType, obj): - """ Restore object during unpickling. """ - # use id(dataType) as key to speed up lookup in dict - # Because of batched pickling, dataType will be the - # same object in most cases. - k = id(dataType) - cls = _cached_cls.get(k) - if cls is None: - # use dataType as key to avoid create multiple class - cls = _cached_cls.get(dataType) - if cls is None: - cls = _create_cls(dataType) - _cached_cls[dataType] = cls - _cached_cls[k] = cls - return cls(obj) - - -def _create_object(cls, v): - """ Create an customized object with class `cls`. """ - # datetime.date would be deserialized as datetime.datetime - # from java type, so we need to set it back. - if cls is datetime.date and isinstance(v, datetime.datetime): - return v.date() - return cls(v) if v is not None else v - - -def _create_getter(dt, i): - """ Create a getter for item `i` with schema """ - cls = _create_cls(dt) - - def getter(self): - return _create_object(cls, self[i]) - - return getter - - -def _has_struct_or_date(dt): - """Return whether `dt` is or has StructType/DateType in it""" - if isinstance(dt, StructType): - return True - elif isinstance(dt, ArrayType): - return _has_struct_or_date(dt.elementType) - elif isinstance(dt, MapType): - return _has_struct_or_date(dt.keyType) or _has_struct_or_date(dt.valueType) - elif isinstance(dt, DateType): - return True - elif isinstance(dt, UserDefinedType): - return True - return False - - -def _create_properties(fields): - """Create properties according to fields""" - ps = {} - for i, f in enumerate(fields): - name = f.name - if (name.startswith("__") and name.endswith("__") - or keyword.iskeyword(name)): - warnings.warn("field name %s can not be accessed in Python," - "use position to access it instead" % name) - if _has_struct_or_date(f.dataType): - # delay creating object until accessing it - getter = _create_getter(f.dataType, i) - else: - getter = itemgetter(i) - ps[name] = property(getter) - return ps - - -def _create_cls(dataType): - """ - Create an class by dataType - - The created class is similar to namedtuple, but can have nested schema. - - >>> schema = _parse_schema_abstract("a b c") - >>> row = (1, 1.0, "str") - >>> schema = _infer_schema_type(row, schema) - >>> obj = _create_cls(schema)(row) - >>> import pickle - >>> pickle.loads(pickle.dumps(obj)) - Row(a=1, b=1.0, c='str') - - >>> row = [[1], {"key": (1, 2.0)}] - >>> schema = _parse_schema_abstract("a[] b{c d}") - >>> schema = _infer_schema_type(row, schema) - >>> obj = _create_cls(schema)(row) - >>> pickle.loads(pickle.dumps(obj)) - Row(a=[1], b={'key': Row(c=1, d=2.0)}) - >>> pickle.loads(pickle.dumps(obj.a)) - [1] - >>> pickle.loads(pickle.dumps(obj.b)) - {'key': Row(c=1, d=2.0)} - """ - - if isinstance(dataType, ArrayType): - cls = _create_cls(dataType.elementType) - - def List(l): - if l is None: - return - return [_create_object(cls, v) for v in l] - - return List - - elif isinstance(dataType, MapType): - kcls = _create_cls(dataType.keyType) - vcls = _create_cls(dataType.valueType) - - def Dict(d): - if d is None: - return - return dict((_create_object(kcls, k), _create_object(vcls, v)) for k, v in d.items()) - - return Dict - - elif isinstance(dataType, DateType): - return datetime.date - - elif isinstance(dataType, UserDefinedType): - return lambda datum: dataType.deserialize(datum) - - elif not isinstance(dataType, StructType): - # no wrapper for primitive types - return lambda x: x - - class Row(tuple): - - """ Row in DataFrame """ - __DATATYPE__ = dataType - __FIELDS__ = tuple(f.name for f in dataType.fields) - __slots__ = () - - # create property for fast access - locals().update(_create_properties(dataType.fields)) - - def asDict(self): - """ Return as a dict """ - return dict((n, getattr(self, n)) for n in self.__FIELDS__) - - def __repr__(self): - # call collect __repr__ for nested objects - return ("Row(%s)" % ", ".join("%s=%r" % (n, getattr(self, n)) - for n in self.__FIELDS__)) - - def __reduce__(self): - return (_restore_object, (self.__DATATYPE__, tuple(self))) - - return Row - - -class SQLContext(object): - - """Main entry point for Spark SQL functionality. - - A SQLContext can be used create L{DataFrame}, register L{DataFrame} as - tables, execute SQL over tables, cache tables, and read parquet files. - """ - - def __init__(self, sparkContext, sqlContext=None): - """Create a new SQLContext. - - :param sparkContext: The SparkContext to wrap. - :param sqlContext: An optional JVM Scala SQLContext. If set, we do not instatiate a new - SQLContext in the JVM, instead we make all calls to this object. - - >>> df = sqlCtx.inferSchema(rdd) - >>> sqlCtx.inferSchema(df) # doctest: +IGNORE_EXCEPTION_DETAIL - Traceback (most recent call last): - ... - TypeError:... - - >>> bad_rdd = sc.parallelize([1,2,3]) - >>> sqlCtx.inferSchema(bad_rdd) # doctest: +IGNORE_EXCEPTION_DETAIL - Traceback (most recent call last): - ... - ValueError:... - - >>> from datetime import datetime - >>> allTypes = sc.parallelize([Row(i=1, s="string", d=1.0, l=1L, - ... b=True, list=[1, 2, 3], dict={"s": 0}, row=Row(a=1), - ... time=datetime(2014, 8, 1, 14, 1, 5))]) - >>> df = sqlCtx.inferSchema(allTypes) - >>> df.registerTempTable("allTypes") - >>> sqlCtx.sql('select i+1, d+1, not b, list[1], dict["s"], time, row.a ' - ... 'from allTypes where b and i > 0').collect() - [Row(c0=2, c1=2.0, c2=False, c3=2, c4=0...8, 1, 14, 1, 5), a=1)] - >>> df.map(lambda x: (x.i, x.s, x.d, x.l, x.b, x.time, - ... x.row.a, x.list)).collect() - [(1, u'string', 1.0, 1, True, ...(2014, 8, 1, 14, 1, 5), 1, [1, 2, 3])] - """ - self._sc = sparkContext - self._jsc = self._sc._jsc - self._jvm = self._sc._jvm - self._scala_SQLContext = sqlContext - - @property - def _ssql_ctx(self): - """Accessor for the JVM Spark SQL context. - - Subclasses can override this property to provide their own - JVM Contexts. - """ - if self._scala_SQLContext is None: - self._scala_SQLContext = self._jvm.SQLContext(self._jsc.sc()) - return self._scala_SQLContext - - def registerFunction(self, name, f, returnType=StringType()): - """Registers a lambda function as a UDF so it can be used in SQL statements. - - In addition to a name and the function itself, the return type can be optionally specified. - When the return type is not given it default to a string and conversion will automatically - be done. For any other return type, the produced object must match the specified type. - - >>> sqlCtx.registerFunction("stringLengthString", lambda x: len(x)) - >>> sqlCtx.sql("SELECT stringLengthString('test')").collect() - [Row(c0=u'4')] - >>> sqlCtx.registerFunction("stringLengthInt", lambda x: len(x), IntegerType()) - >>> sqlCtx.sql("SELECT stringLengthInt('test')").collect() - [Row(c0=4)] - """ - func = lambda _, it: imap(lambda x: f(*x), it) - ser = AutoBatchedSerializer(PickleSerializer()) - command = (func, None, ser, ser) - pickled_cmd, bvars, env, includes = _prepare_for_python_RDD(self._sc, command, self) - self._ssql_ctx.udf().registerPython(name, - bytearray(pickled_cmd), - env, - includes, - self._sc.pythonExec, - bvars, - self._sc._javaAccumulator, - returnType.json()) - - def inferSchema(self, rdd, samplingRatio=None): - """Infer and apply a schema to an RDD of L{Row}. - - When samplingRatio is specified, the schema is inferred by looking - at the types of each row in the sampled dataset. Otherwise, the - first 100 rows of the RDD are inspected. Nested collections are - supported, which can include array, dict, list, Row, tuple, - namedtuple, or object. - - Each row could be L{pyspark.sql.Row} object or namedtuple or objects. - Using top level dicts is deprecated, as dict is used to represent Maps. - - If a single column has multiple distinct inferred types, it may cause - runtime exceptions. - - >>> rdd = sc.parallelize( - ... [Row(field1=1, field2="row1"), - ... Row(field1=2, field2="row2"), - ... Row(field1=3, field2="row3")]) - >>> df = sqlCtx.inferSchema(rdd) - >>> df.collect()[0] - Row(field1=1, field2=u'row1') - - >>> NestedRow = Row("f1", "f2") - >>> nestedRdd1 = sc.parallelize([ - ... NestedRow(array('i', [1, 2]), {"row1": 1.0}), - ... NestedRow(array('i', [2, 3]), {"row2": 2.0})]) - >>> df = sqlCtx.inferSchema(nestedRdd1) - >>> df.collect() - [Row(f1=[1, 2], f2={u'row1': 1.0}), ..., f2={u'row2': 2.0})] - - >>> nestedRdd2 = sc.parallelize([ - ... NestedRow([[1, 2], [2, 3]], [1, 2]), - ... NestedRow([[2, 3], [3, 4]], [2, 3])]) - >>> df = sqlCtx.inferSchema(nestedRdd2) - >>> df.collect() - [Row(f1=[[1, 2], [2, 3]], f2=[1, 2]), ..., f2=[2, 3])] - - >>> from collections import namedtuple - >>> CustomRow = namedtuple('CustomRow', 'field1 field2') - >>> rdd = sc.parallelize( - ... [CustomRow(field1=1, field2="row1"), - ... CustomRow(field1=2, field2="row2"), - ... CustomRow(field1=3, field2="row3")]) - >>> df = sqlCtx.inferSchema(rdd) - >>> df.collect()[0] - Row(field1=1, field2=u'row1') - """ - - if isinstance(rdd, DataFrame): - raise TypeError("Cannot apply schema to DataFrame") - - first = rdd.first() - if not first: - raise ValueError("The first row in RDD is empty, " - "can not infer schema") - if type(first) is dict: - warnings.warn("Using RDD of dict to inferSchema is deprecated," - "please use pyspark.sql.Row instead") - - if samplingRatio is None: - schema = _infer_schema(first) - if _has_nulltype(schema): - for row in rdd.take(100)[1:]: - schema = _merge_type(schema, _infer_schema(row)) - if not _has_nulltype(schema): - break - else: - warnings.warn("Some of types cannot be determined by the " - "first 100 rows, please try again with sampling") - else: - if samplingRatio > 0.99: - rdd = rdd.sample(False, float(samplingRatio)) - schema = rdd.map(_infer_schema).reduce(_merge_type) - - converter = _create_converter(schema) - rdd = rdd.map(converter) - return self.applySchema(rdd, schema) - - def applySchema(self, rdd, schema): - """ - Applies the given schema to the given RDD of L{tuple} or L{list}. - - These tuples or lists can contain complex nested structures like - lists, maps or nested rows. - - The schema should be a StructType. - - It is important that the schema matches the types of the objects - in each row or exceptions could be thrown at runtime. - - >>> rdd2 = sc.parallelize([(1, "row1"), (2, "row2"), (3, "row3")]) - >>> schema = StructType([StructField("field1", IntegerType(), False), - ... StructField("field2", StringType(), False)]) - >>> df = sqlCtx.applySchema(rdd2, schema) - >>> sqlCtx.registerRDDAsTable(df, "table1") - >>> df2 = sqlCtx.sql("SELECT * from table1") - >>> df2.collect() - [Row(field1=1, field2=u'row1'),..., Row(field1=3, field2=u'row3')] - - >>> from datetime import date, datetime - >>> rdd = sc.parallelize([(127, -128L, -32768, 32767, 2147483647L, 1.0, - ... date(2010, 1, 1), - ... datetime(2010, 1, 1, 1, 1, 1), - ... {"a": 1}, (2,), [1, 2, 3], None)]) - >>> schema = StructType([ - ... StructField("byte1", ByteType(), False), - ... StructField("byte2", ByteType(), False), - ... StructField("short1", ShortType(), False), - ... StructField("short2", ShortType(), False), - ... StructField("int", IntegerType(), False), - ... StructField("float", FloatType(), False), - ... StructField("date", DateType(), False), - ... StructField("time", TimestampType(), False), - ... StructField("map", - ... MapType(StringType(), IntegerType(), False), False), - ... StructField("struct", - ... StructType([StructField("b", ShortType(), False)]), False), - ... StructField("list", ArrayType(ByteType(), False), False), - ... StructField("null", DoubleType(), True)]) - >>> df = sqlCtx.applySchema(rdd, schema) - >>> results = df.map( - ... lambda x: (x.byte1, x.byte2, x.short1, x.short2, x.int, x.float, x.date, - ... x.time, x.map["a"], x.struct.b, x.list, x.null)) - >>> results.collect()[0] # doctest: +NORMALIZE_WHITESPACE - (127, -128, -32768, 32767, 2147483647, 1.0, datetime.date(2010, 1, 1), - datetime.datetime(2010, 1, 1, 1, 1, 1), 1, 2, [1, 2, 3], None) - - >>> df.registerTempTable("table2") - >>> sqlCtx.sql( - ... "SELECT byte1 - 1 AS byte1, byte2 + 1 AS byte2, " + - ... "short1 + 1 AS short1, short2 - 1 AS short2, int - 1 AS int, " + - ... "float + 1.5 as float FROM table2").collect() - [Row(byte1=126, byte2=-127, short1=-32767, short2=32766, int=2147483646, float=2.5)] - - >>> rdd = sc.parallelize([(127, -32768, 1.0, - ... datetime(2010, 1, 1, 1, 1, 1), - ... {"a": 1}, (2,), [1, 2, 3])]) - >>> abstract = "byte short float time map{} struct(b) list[]" - >>> schema = _parse_schema_abstract(abstract) - >>> typedSchema = _infer_schema_type(rdd.first(), schema) - >>> df = sqlCtx.applySchema(rdd, typedSchema) - >>> df.collect() - [Row(byte=127, short=-32768, float=1.0, time=..., list=[1, 2, 3])] - """ - - if isinstance(rdd, DataFrame): - raise TypeError("Cannot apply schema to DataFrame") - - if not isinstance(schema, StructType): - raise TypeError("schema should be StructType") - - # take the first few rows to verify schema - rows = rdd.take(10) - # Row() cannot been deserialized by Pyrolite - if rows and isinstance(rows[0], tuple) and rows[0].__class__.__name__ == 'Row': - rdd = rdd.map(tuple) - rows = rdd.take(10) - - for row in rows: - _verify_type(row, schema) - - # convert python objects to sql data - converter = _python_to_sql_converter(schema) - rdd = rdd.map(converter) - - jrdd = self._jvm.SerDeUtil.toJavaArray(rdd._to_java_object_rdd()) - df = self._ssql_ctx.applySchemaToPythonRDD(jrdd.rdd(), schema.json()) - return DataFrame(df, self) - - def registerRDDAsTable(self, rdd, tableName): - """Registers the given RDD as a temporary table in the catalog. - - Temporary tables exist only during the lifetime of this instance of - SQLContext. - - >>> df = sqlCtx.inferSchema(rdd) - >>> sqlCtx.registerRDDAsTable(df, "table1") - """ - if (rdd.__class__ is DataFrame): - df = rdd._jdf - self._ssql_ctx.registerRDDAsTable(df, tableName) - else: - raise ValueError("Can only register DataFrame as table") - - def parquetFile(self, *paths): - """Loads a Parquet file, returning the result as a L{DataFrame}. - - >>> import tempfile, shutil - >>> parquetFile = tempfile.mkdtemp() - >>> shutil.rmtree(parquetFile) - >>> df = sqlCtx.inferSchema(rdd) - >>> df.saveAsParquetFile(parquetFile) - >>> df2 = sqlCtx.parquetFile(parquetFile) - >>> sorted(df.collect()) == sorted(df2.collect()) - True - """ - gateway = self._sc._gateway - jpath = paths[0] - jpaths = gateway.new_array(gateway.jvm.java.lang.String, len(paths) - 1) - for i in range(1, len(paths)): - jpaths[i] = paths[i] - jdf = self._ssql_ctx.parquetFile(jpath, jpaths) - return DataFrame(jdf, self) - - def jsonFile(self, path, schema=None, samplingRatio=1.0): - """ - Loads a text file storing one JSON object per line as a - L{DataFrame}. - - If the schema is provided, applies the given schema to this - JSON dataset. - - Otherwise, it samples the dataset with ratio `samplingRatio` to - determine the schema. - - >>> import tempfile, shutil - >>> jsonFile = tempfile.mkdtemp() - >>> shutil.rmtree(jsonFile) - >>> ofn = open(jsonFile, 'w') - >>> for json in jsonStrings: - ... print>>ofn, json - >>> ofn.close() - >>> df1 = sqlCtx.jsonFile(jsonFile) - >>> sqlCtx.registerRDDAsTable(df1, "table1") - >>> df2 = sqlCtx.sql( - ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " - ... "field6 as f4 from table1") - >>> for r in df2.collect(): - ... print r - Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) - Row(f1=2, f2=None, f3=Row(field4=22,..., f4=[Row(field7=u'row2')]) - Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) - - >>> df3 = sqlCtx.jsonFile(jsonFile, df1.schema()) - >>> sqlCtx.registerRDDAsTable(df3, "table2") - >>> df4 = sqlCtx.sql( - ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " - ... "field6 as f4 from table2") - >>> for r in df4.collect(): - ... print r - Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) - Row(f1=2, f2=None, f3=Row(field4=22,..., f4=[Row(field7=u'row2')]) - Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) - - >>> schema = StructType([ - ... StructField("field2", StringType(), True), - ... StructField("field3", - ... StructType([ - ... StructField("field5", - ... ArrayType(IntegerType(), False), True)]), False)]) - >>> df5 = sqlCtx.jsonFile(jsonFile, schema) - >>> sqlCtx.registerRDDAsTable(df5, "table3") - >>> df6 = sqlCtx.sql( - ... "SELECT field2 AS f1, field3.field5 as f2, " - ... "field3.field5[0] as f3 from table3") - >>> df6.collect() - [Row(f1=u'row1', f2=None, f3=None)...Row(f1=u'row3', f2=[], f3=None)] - """ - if schema is None: - df = self._ssql_ctx.jsonFile(path, samplingRatio) - else: - scala_datatype = self._ssql_ctx.parseDataType(schema.json()) - df = self._ssql_ctx.jsonFile(path, scala_datatype) - return DataFrame(df, self) - - def jsonRDD(self, rdd, schema=None, samplingRatio=1.0): - """Loads an RDD storing one JSON object per string as a L{DataFrame}. - - If the schema is provided, applies the given schema to this - JSON dataset. - - Otherwise, it samples the dataset with ratio `samplingRatio` to - determine the schema. - - >>> df1 = sqlCtx.jsonRDD(json) - >>> sqlCtx.registerRDDAsTable(df1, "table1") - >>> df2 = sqlCtx.sql( - ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " - ... "field6 as f4 from table1") - >>> for r in df2.collect(): - ... print r - Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) - Row(f1=2, f2=None, f3=Row(field4=22..., f4=[Row(field7=u'row2')]) - Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) - - >>> df3 = sqlCtx.jsonRDD(json, df1.schema()) - >>> sqlCtx.registerRDDAsTable(df3, "table2") - >>> df4 = sqlCtx.sql( - ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " - ... "field6 as f4 from table2") - >>> for r in df4.collect(): - ... print r - Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) - Row(f1=2, f2=None, f3=Row(field4=22..., f4=[Row(field7=u'row2')]) - Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) - - >>> schema = StructType([ - ... StructField("field2", StringType(), True), - ... StructField("field3", - ... StructType([ - ... StructField("field5", - ... ArrayType(IntegerType(), False), True)]), False)]) - >>> df5 = sqlCtx.jsonRDD(json, schema) - >>> sqlCtx.registerRDDAsTable(df5, "table3") - >>> df6 = sqlCtx.sql( - ... "SELECT field2 AS f1, field3.field5 as f2, " - ... "field3.field5[0] as f3 from table3") - >>> df6.collect() - [Row(f1=u'row1', f2=None,...Row(f1=u'row3', f2=[], f3=None)] - - >>> sqlCtx.jsonRDD(sc.parallelize(['{}', - ... '{"key0": {"key1": "value1"}}'])).collect() - [Row(key0=None), Row(key0=Row(key1=u'value1'))] - >>> sqlCtx.jsonRDD(sc.parallelize(['{"key0": null}', - ... '{"key0": {"key1": "value1"}}'])).collect() - [Row(key0=None), Row(key0=Row(key1=u'value1'))] - """ - - def func(iterator): - for x in iterator: - if not isinstance(x, basestring): - x = unicode(x) - if isinstance(x, unicode): - x = x.encode("utf-8") - yield x - keyed = rdd.mapPartitions(func) - keyed._bypass_serializer = True - jrdd = keyed._jrdd.map(self._jvm.BytesToString()) - if schema is None: - df = self._ssql_ctx.jsonRDD(jrdd.rdd(), samplingRatio) - else: - scala_datatype = self._ssql_ctx.parseDataType(schema.json()) - df = self._ssql_ctx.jsonRDD(jrdd.rdd(), scala_datatype) - return DataFrame(df, self) - - def sql(self, sqlQuery): - """Return a L{DataFrame} representing the result of the given query. - - >>> df = sqlCtx.inferSchema(rdd) - >>> sqlCtx.registerRDDAsTable(df, "table1") - >>> df2 = sqlCtx.sql("SELECT field1 AS f1, field2 as f2 from table1") - >>> df2.collect() - [Row(f1=1, f2=u'row1'), Row(f1=2, f2=u'row2'), Row(f1=3, f2=u'row3')] - """ - return DataFrame(self._ssql_ctx.sql(sqlQuery), self) - - def table(self, tableName): - """Returns the specified table as a L{DataFrame}. - - >>> df = sqlCtx.inferSchema(rdd) - >>> sqlCtx.registerRDDAsTable(df, "table1") - >>> df2 = sqlCtx.table("table1") - >>> sorted(df.collect()) == sorted(df2.collect()) - True - """ - return DataFrame(self._ssql_ctx.table(tableName), self) - - def cacheTable(self, tableName): - """Caches the specified table in-memory.""" - self._ssql_ctx.cacheTable(tableName) - - def uncacheTable(self, tableName): - """Removes the specified table from the in-memory cache.""" - self._ssql_ctx.uncacheTable(tableName) - - -class HiveContext(SQLContext): - - """A variant of Spark SQL that integrates with data stored in Hive. - - Configuration for Hive is read from hive-site.xml on the classpath. - It supports running both SQL and HiveQL commands. - """ - - def __init__(self, sparkContext, hiveContext=None): - """Create a new HiveContext. - - :param sparkContext: The SparkContext to wrap. - :param hiveContext: An optional JVM Scala HiveContext. If set, we do not instatiate a new - HiveContext in the JVM, instead we make all calls to this object. - """ - SQLContext.__init__(self, sparkContext) - - if hiveContext: - self._scala_HiveContext = hiveContext - - @property - def _ssql_ctx(self): - try: - if not hasattr(self, '_scala_HiveContext'): - self._scala_HiveContext = self._get_hive_ctx() - return self._scala_HiveContext - except Py4JError as e: - raise Exception("You must build Spark with Hive. " - "Export 'SPARK_HIVE=true' and run " - "build/sbt assembly", e) - - def _get_hive_ctx(self): - return self._jvm.HiveContext(self._jsc.sc()) - - -def _create_row(fields, values): - row = Row(*values) - row.__FIELDS__ = fields - return row - - -class Row(tuple): - - """ - A row in L{DataFrame}. The fields in it can be accessed like attributes. - - Row can be used to create a row object by using named arguments, - the fields will be sorted by names. - - >>> row = Row(name="Alice", age=11) - >>> row - Row(age=11, name='Alice') - >>> row.name, row.age - ('Alice', 11) - - Row also can be used to create another Row like class, then it - could be used to create Row objects, such as - - >>> Person = Row("name", "age") - >>> Person - - >>> Person("Alice", 11) - Row(name='Alice', age=11) - """ - - def __new__(self, *args, **kwargs): - if args and kwargs: - raise ValueError("Can not use both args " - "and kwargs to create Row") - if args: - # create row class or objects - return tuple.__new__(self, args) - - elif kwargs: - # create row objects - names = sorted(kwargs.keys()) - values = tuple(kwargs[n] for n in names) - row = tuple.__new__(self, values) - row.__FIELDS__ = names - return row - - else: - raise ValueError("No args or kwargs") - - def asDict(self): - """ - Return as an dict - """ - if not hasattr(self, "__FIELDS__"): - raise TypeError("Cannot convert a Row class into dict") - return dict(zip(self.__FIELDS__, self)) - - # let obect acs like class - def __call__(self, *args): - """create new Row object""" - return _create_row(self, args) - - def __getattr__(self, item): - if item.startswith("__"): - raise AttributeError(item) - try: - # it will be slow when it has many fields, - # but this will not be used in normal cases - idx = self.__FIELDS__.index(item) - return self[idx] - except IndexError: - raise AttributeError(item) - - def __reduce__(self): - if hasattr(self, "__FIELDS__"): - return (_create_row, (self.__FIELDS__, tuple(self))) - else: - return tuple.__reduce__(self) - - def __repr__(self): - if hasattr(self, "__FIELDS__"): - return "Row(%s)" % ", ".join("%s=%r" % (k, v) - for k, v in zip(self.__FIELDS__, self)) - else: - return "" % ", ".join(self) - - -class DataFrame(object): - - """A collection of rows that have the same columns. - - A :class:`DataFrame` is equivalent to a relational table in Spark SQL, - and can be created using various functions in :class:`SQLContext`:: - - people = sqlContext.parquetFile("...") - - Once created, it can be manipulated using the various domain-specific-language - (DSL) functions defined in: :class:`DataFrame`, :class:`Column`. - - To select a column from the data frame, use the apply method:: - - ageCol = people.age - - Note that the :class:`Column` type can also be manipulated - through its various functions:: - - # The following creates a new column that increases everybody's age by 10. - people.age + 10 - - - A more concrete example:: - - # To create DataFrame using SQLContext - people = sqlContext.parquetFile("...") - department = sqlContext.parquetFile("...") - - people.filter(people.age > 30).join(department, people.deptId == department.id)) \ - .groupBy(department.name, "gender").agg({"salary": "avg", "age": "max"}) - """ - - def __init__(self, jdf, sql_ctx): - self._jdf = jdf - self.sql_ctx = sql_ctx - self._sc = sql_ctx and sql_ctx._sc - self.is_cached = False - - @property - def rdd(self): - """ - Return the content of the :class:`DataFrame` as an :class:`RDD` - of :class:`Row` s. - """ - if not hasattr(self, '_lazy_rdd'): - jrdd = self._jdf.javaToPython() - rdd = RDD(jrdd, self.sql_ctx._sc, BatchedSerializer(PickleSerializer())) - schema = self.schema() - - def applySchema(it): - cls = _create_cls(schema) - return itertools.imap(cls, it) - - self._lazy_rdd = rdd.mapPartitions(applySchema) - - return self._lazy_rdd - - def toJSON(self, use_unicode=False): - """Convert a DataFrame into a MappedRDD of JSON documents; one document per row. - - >>> df1 = sqlCtx.jsonRDD(json) - >>> sqlCtx.registerRDDAsTable(df1, "table1") - >>> df2 = sqlCtx.sql( "SELECT * from table1") - >>> df2.toJSON().take(1)[0] == '{"field1":1,"field2":"row1","field3":{"field4":11}}' - True - >>> df3 = sqlCtx.sql( "SELECT field3.field4 from table1") - >>> df3.toJSON().collect() == ['{"field4":11}', '{"field4":22}', '{"field4":33}'] - True - """ - rdd = self._jdf.toJSON() - return RDD(rdd.toJavaRDD(), self._sc, UTF8Deserializer(use_unicode)) - - def saveAsParquetFile(self, path): - """Save the contents as a Parquet file, preserving the schema. - - Files that are written out using this method can be read back in as - a DataFrame using the L{SQLContext.parquetFile} method. - - >>> import tempfile, shutil - >>> parquetFile = tempfile.mkdtemp() - >>> shutil.rmtree(parquetFile) - >>> df.saveAsParquetFile(parquetFile) - >>> df2 = sqlCtx.parquetFile(parquetFile) - >>> sorted(df2.collect()) == sorted(df.collect()) - True - """ - self._jdf.saveAsParquetFile(path) - - def registerTempTable(self, name): - """Registers this RDD as a temporary table using the given name. - - The lifetime of this temporary table is tied to the L{SQLContext} - that was used to create this DataFrame. - - >>> df.registerTempTable("people") - >>> df2 = sqlCtx.sql("select * from people") - >>> sorted(df.collect()) == sorted(df2.collect()) - True - """ - self._jdf.registerTempTable(name) - - def registerAsTable(self, name): - """DEPRECATED: use registerTempTable() instead""" - warnings.warn("Use registerTempTable instead of registerAsTable.", DeprecationWarning) - self.registerTempTable(name) - - def insertInto(self, tableName, overwrite=False): - """Inserts the contents of this DataFrame into the specified table. - - Optionally overwriting any existing data. - """ - self._jdf.insertInto(tableName, overwrite) - - def saveAsTable(self, tableName): - """Creates a new table with the contents of this DataFrame.""" - self._jdf.saveAsTable(tableName) - - def schema(self): - """Returns the schema of this DataFrame (represented by - a L{StructType}). - - >>> df.schema() - StructType(List(StructField(age,IntegerType,true),StructField(name,StringType,true))) - """ - return _parse_datatype_json_string(self._jdf.schema().json()) - - def printSchema(self): - """Prints out the schema in the tree format. - - >>> df.printSchema() - root - |-- age: integer (nullable = true) - |-- name: string (nullable = true) - - """ - print (self._jdf.schema().treeString()) - - def count(self): - """Return the number of elements in this RDD. - - Unlike the base RDD implementation of count, this implementation - leverages the query optimizer to compute the count on the DataFrame, - which supports features such as filter pushdown. - - >>> df.count() - 2L - """ - return self._jdf.count() - - def collect(self): - """Return a list that contains all of the rows. - - Each object in the list is a Row, the fields can be accessed as - attributes. - - >>> df.collect() - [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] - """ - with SCCallSiteSync(self._sc) as css: - bytesInJava = self._jdf.javaToPython().collect().iterator() - tempFile = NamedTemporaryFile(delete=False, dir=self._sc._temp_dir) - tempFile.close() - self._sc._writeToFile(bytesInJava, tempFile.name) - # Read the data into Python and deserialize it: - with open(tempFile.name, 'rb') as tempFile: - rs = list(BatchedSerializer(PickleSerializer()).load_stream(tempFile)) - os.unlink(tempFile.name) - cls = _create_cls(self.schema()) - return [cls(r) for r in rs] - - def limit(self, num): - """Limit the result count to the number specified. - - >>> df.limit(1).collect() - [Row(age=2, name=u'Alice')] - >>> df.limit(0).collect() - [] - """ - jdf = self._jdf.limit(num) - return DataFrame(jdf, self.sql_ctx) - - def take(self, num): - """Take the first num rows of the RDD. - - Each object in the list is a Row, the fields can be accessed as - attributes. - - >>> df.take(2) - [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] - """ - return self.limit(num).collect() - - def map(self, f): - """ Return a new RDD by applying a function to each Row, it's a - shorthand for df.rdd.map() - - >>> df.map(lambda p: p.name).collect() - [u'Alice', u'Bob'] - """ - return self.rdd.map(f) - - def mapPartitions(self, f, preservesPartitioning=False): - """ - Return a new RDD by applying a function to each partition. - - >>> rdd = sc.parallelize([1, 2, 3, 4], 4) - >>> def f(iterator): yield 1 - >>> rdd.mapPartitions(f).sum() - 4 - """ - return self.rdd.mapPartitions(f, preservesPartitioning) - - def cache(self): - """ Persist with the default storage level (C{MEMORY_ONLY_SER}). - """ - self.is_cached = True - self._jdf.cache() - return self - - def persist(self, storageLevel=StorageLevel.MEMORY_ONLY_SER): - """ Set the storage level to persist its values across operations - after the first time it is computed. This can only be used to assign - a new storage level if the RDD does not have a storage level set yet. - If no storage level is specified defaults to (C{MEMORY_ONLY_SER}). - """ - self.is_cached = True - javaStorageLevel = self._sc._getJavaStorageLevel(storageLevel) - self._jdf.persist(javaStorageLevel) - return self - - def unpersist(self, blocking=True): - """ Mark it as non-persistent, and remove all blocks for it from - memory and disk. - """ - self.is_cached = False - self._jdf.unpersist(blocking) - return self - - # def coalesce(self, numPartitions, shuffle=False): - # rdd = self._jdf.coalesce(numPartitions, shuffle, None) - # return DataFrame(rdd, self.sql_ctx) - - def repartition(self, numPartitions): - """ Return a new :class:`DataFrame` that has exactly `numPartitions` - partitions. - """ - rdd = self._jdf.repartition(numPartitions, None) - return DataFrame(rdd, self.sql_ctx) - - def sample(self, withReplacement, fraction, seed=None): - """ - Return a sampled subset of this DataFrame. - - >>> df = sqlCtx.inferSchema(rdd) - >>> df.sample(False, 0.5, 97).count() - 2L - """ - assert fraction >= 0.0, "Negative fraction value: %s" % fraction - seed = seed if seed is not None else random.randint(0, sys.maxint) - rdd = self._jdf.sample(withReplacement, fraction, long(seed)) - return DataFrame(rdd, self.sql_ctx) - - # def takeSample(self, withReplacement, num, seed=None): - # """Return a fixed-size sampled subset of this DataFrame. - # - # >>> df = sqlCtx.inferSchema(rdd) - # >>> df.takeSample(False, 2, 97) - # [Row(field1=3, field2=u'row3'), Row(field1=1, field2=u'row1')] - # """ - # seed = seed if seed is not None else random.randint(0, sys.maxint) - # with SCCallSiteSync(self.context) as css: - # bytesInJava = self._jdf \ - # .takeSampleToPython(withReplacement, num, long(seed)) \ - # .iterator() - # cls = _create_cls(self.schema()) - # return map(cls, self._collect_iterator_through_file(bytesInJava)) - - @property - def dtypes(self): - """Return all column names and their data types as a list. - - >>> df.dtypes - [('age', 'integer'), ('name', 'string')] - """ - return [(str(f.name), f.dataType.jsonValue()) for f in self.schema().fields] - - @property - def columns(self): - """ Return all column names as a list. - - >>> df.columns - [u'age', u'name'] - """ - return [f.name for f in self.schema().fields] - - def join(self, other, joinExprs=None, joinType=None): - """ - Join with another DataFrame, using the given join expression. - The following performs a full outer join between `df1` and `df2`:: - - :param other: Right side of the join - :param joinExprs: Join expression - :param joinType: One of `inner`, `outer`, `left_outer`, `right_outer`, `semijoin`. - - >>> df.join(df2, df.name == df2.name, 'outer').select(df.name, df2.height).collect() - [Row(name=None, height=80), Row(name=u'Bob', height=85), Row(name=u'Alice', height=None)] - """ - - if joinExprs is None: - jdf = self._jdf.join(other._jdf) - else: - assert isinstance(joinExprs, Column), "joinExprs should be Column" - if joinType is None: - jdf = self._jdf.join(other._jdf, joinExprs._jc) - else: - assert isinstance(joinType, basestring), "joinType should be basestring" - jdf = self._jdf.join(other._jdf, joinExprs._jc, joinType) - return DataFrame(jdf, self.sql_ctx) - - def sort(self, *cols): - """ Return a new :class:`DataFrame` sorted by the specified column. - - :param cols: The columns or expressions used for sorting - - >>> df.sort(df.age.desc()).collect() - [Row(age=5, name=u'Bob'), Row(age=2, name=u'Alice')] - >>> df.sortBy(df.age.desc()).collect() - [Row(age=5, name=u'Bob'), Row(age=2, name=u'Alice')] - """ - if not cols: - raise ValueError("should sort by at least one column") - jcols = ListConverter().convert([_to_java_column(c) for c in cols], - self._sc._gateway._gateway_client) - jdf = self._jdf.sort(self._sc._jvm.PythonUtils.toSeq(jcols)) - return DataFrame(jdf, self.sql_ctx) - - sortBy = sort - - def head(self, n=None): - """ Return the first `n` rows or the first row if n is None. - - >>> df.head() - Row(age=2, name=u'Alice') - >>> df.head(1) - [Row(age=2, name=u'Alice')] - """ - if n is None: - rs = self.head(1) - return rs[0] if rs else None - return self.take(n) - - def first(self): - """ Return the first row. - - >>> df.first() - Row(age=2, name=u'Alice') - """ - return self.head() - - def __getitem__(self, item): - """ Return the column by given name - - >>> df['age'].collect() - [Row(age=2), Row(age=5)] - >>> df[ ["name", "age"]].collect() - [Row(name=u'Alice', age=2), Row(name=u'Bob', age=5)] - >>> df[ df.age > 3 ].collect() - [Row(age=5, name=u'Bob')] - """ - if isinstance(item, basestring): - jc = self._jdf.apply(item) - return Column(jc, self.sql_ctx) - elif isinstance(item, Column): - return self.filter(item) - elif isinstance(item, list): - return self.select(*item) - else: - raise IndexError("unexpected index: %s" % item) - - def __getattr__(self, name): - """ Return the column by given name - - >>> df.age.collect() - [Row(age=2), Row(age=5)] - """ - if name.startswith("__"): - raise AttributeError(name) - jc = self._jdf.apply(name) - return Column(jc, self.sql_ctx) - - def select(self, *cols): - """ Selecting a set of expressions. - - >>> df.select().collect() - [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] - >>> df.select('*').collect() - [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] - >>> df.select('name', 'age').collect() - [Row(name=u'Alice', age=2), Row(name=u'Bob', age=5)] - >>> df.select(df.name, (df.age + 10).alias('age')).collect() - [Row(name=u'Alice', age=12), Row(name=u'Bob', age=15)] - """ - if not cols: - cols = ["*"] - jcols = ListConverter().convert([_to_java_column(c) for c in cols], - self._sc._gateway._gateway_client) - jdf = self._jdf.select(self.sql_ctx._sc._jvm.PythonUtils.toSeq(jcols)) - return DataFrame(jdf, self.sql_ctx) - - def selectExpr(self, *expr): - """ - Selects a set of SQL expressions. This is a variant of - `select` that accepts SQL expressions. - - >>> df.selectExpr("age * 2", "abs(age)").collect() - [Row(('age * 2)=4, Abs('age)=2), Row(('age * 2)=10, Abs('age)=5)] - """ - jexpr = ListConverter().convert(expr, self._sc._gateway._gateway_client) - jdf = self._jdf.selectExpr(self._sc._jvm.PythonUtils.toSeq(jexpr)) - return DataFrame(jdf, self.sql_ctx) - - def filter(self, condition): - """ Filtering rows using the given condition, which could be - Column expression or string of SQL expression. - - where() is an alias for filter(). - - >>> df.filter(df.age > 3).collect() - [Row(age=5, name=u'Bob')] - >>> df.where(df.age == 2).collect() - [Row(age=2, name=u'Alice')] - - >>> df.filter("age > 3").collect() - [Row(age=5, name=u'Bob')] - >>> df.where("age = 2").collect() - [Row(age=2, name=u'Alice')] - """ - if isinstance(condition, basestring): - jdf = self._jdf.filter(condition) - elif isinstance(condition, Column): - jdf = self._jdf.filter(condition._jc) - else: - raise TypeError("condition should be string or Column") - return DataFrame(jdf, self.sql_ctx) - - where = filter - - def groupBy(self, *cols): - """ Group the :class:`DataFrame` using the specified columns, - so we can run aggregation on them. See :class:`GroupedData` - for all the available aggregate functions. - - >>> df.groupBy().avg().collect() - [Row(AVG(age#0)=3.5)] - >>> df.groupBy('name').agg({'age': 'mean'}).collect() - [Row(name=u'Bob', AVG(age#0)=5.0), Row(name=u'Alice', AVG(age#0)=2.0)] - >>> df.groupBy(df.name).avg().collect() - [Row(name=u'Bob', AVG(age#0)=5.0), Row(name=u'Alice', AVG(age#0)=2.0)] - """ - jcols = ListConverter().convert([_to_java_column(c) for c in cols], - self._sc._gateway._gateway_client) - jdf = self._jdf.groupBy(self.sql_ctx._sc._jvm.PythonUtils.toSeq(jcols)) - return GroupedData(jdf, self.sql_ctx) - - def agg(self, *exprs): - """ Aggregate on the entire :class:`DataFrame` without groups - (shorthand for df.groupBy.agg()). - - >>> df.agg({"age": "max"}).collect() - [Row(MAX(age#0)=5)] - >>> from pyspark.sql import Dsl - >>> df.agg(Dsl.min(df.age)).collect() - [Row(MIN(age#0)=2)] - """ - return self.groupBy().agg(*exprs) - - def unionAll(self, other): - """ Return a new DataFrame containing union of rows in this - frame and another frame. - - This is equivalent to `UNION ALL` in SQL. - """ - return DataFrame(self._jdf.unionAll(other._jdf), self.sql_ctx) - - def intersect(self, other): - """ Return a new :class:`DataFrame` containing rows only in - both this frame and another frame. - - This is equivalent to `INTERSECT` in SQL. - """ - return DataFrame(self._jdf.intersect(other._jdf), self.sql_ctx) - - def subtract(self, other): - """ Return a new :class:`DataFrame` containing rows in this frame - but not in another frame. - - This is equivalent to `EXCEPT` in SQL. - """ - return DataFrame(getattr(self._jdf, "except")(other._jdf), self.sql_ctx) - - def addColumn(self, colName, col): - """ Return a new :class:`DataFrame` by adding a column. - - >>> df.addColumn('age2', df.age + 2).collect() - [Row(age=2, name=u'Alice', age2=4), Row(age=5, name=u'Bob', age2=7)] - """ - return self.select('*', col.alias(colName)) - - -# Having SchemaRDD for backward compatibility (for docs) -class SchemaRDD(DataFrame): - """ - SchemaRDD is deprecated, please use DataFrame - """ - - -def dfapi(f): - def _api(self): - name = f.__name__ - jdf = getattr(self._jdf, name)() - return DataFrame(jdf, self.sql_ctx) - _api.__name__ = f.__name__ - _api.__doc__ = f.__doc__ - return _api - - -class GroupedData(object): - - """ - A set of methods for aggregations on a :class:`DataFrame`, - created by DataFrame.groupBy(). - """ - - def __init__(self, jdf, sql_ctx): - self._jdf = jdf - self.sql_ctx = sql_ctx - - def agg(self, *exprs): - """ Compute aggregates by specifying a map from column name - to aggregate methods. - - The available aggregate methods are `avg`, `max`, `min`, - `sum`, `count`. - - :param exprs: list or aggregate columns or a map from column - name to aggregate methods. - - >>> gdf = df.groupBy(df.name) - >>> gdf.agg({"age": "max"}).collect() - [Row(name=u'Bob', MAX(age#0)=5), Row(name=u'Alice', MAX(age#0)=2)] - >>> from pyspark.sql import Dsl - >>> gdf.agg(Dsl.min(df.age)).collect() - [Row(MIN(age#0)=5), Row(MIN(age#0)=2)] - """ - assert exprs, "exprs should not be empty" - if len(exprs) == 1 and isinstance(exprs[0], dict): - jmap = MapConverter().convert(exprs[0], - self.sql_ctx._sc._gateway._gateway_client) - jdf = self._jdf.agg(jmap) - else: - # Columns - assert all(isinstance(c, Column) for c in exprs), "all exprs should be Column" - jcols = ListConverter().convert([c._jc for c in exprs[1:]], - self.sql_ctx._sc._gateway._gateway_client) - jdf = self._jdf.agg(exprs[0]._jc, self.sql_ctx._sc._jvm.PythonUtils.toSeq(jcols)) - return DataFrame(jdf, self.sql_ctx) - - @dfapi - def count(self): - """ Count the number of rows for each group. - - >>> df.groupBy(df.age).count().collect() - [Row(age=2, count=1), Row(age=5, count=1)] - """ - - @dfapi - def mean(self): - """Compute the average value for each numeric columns - for each group. This is an alias for `avg`.""" - - @dfapi - def avg(self): - """Compute the average value for each numeric columns - for each group.""" - - @dfapi - def max(self): - """Compute the max value for each numeric columns for - each group. """ - - @dfapi - def min(self): - """Compute the min value for each numeric column for - each group.""" - - @dfapi - def sum(self): - """Compute the sum for each numeric columns for each - group.""" - - -def _create_column_from_literal(literal): - sc = SparkContext._active_spark_context - return sc._jvm.Dsl.lit(literal) - - -def _create_column_from_name(name): - sc = SparkContext._active_spark_context - return sc._jvm.Dsl.col(name) - - -def _to_java_column(col): - if isinstance(col, Column): - jcol = col._jc - else: - jcol = _create_column_from_name(col) - return jcol - - -def _unary_op(name, doc="unary operator"): - """ Create a method for given unary operator """ - def _(self): - jc = getattr(self._jc, name)() - return Column(jc, self.sql_ctx) - _.__doc__ = doc - return _ - - -def _dsl_op(name, doc=''): - def _(self): - jc = getattr(self._sc._jvm.Dsl, name)(self._jc) - return Column(jc, self.sql_ctx) - _.__doc__ = doc - return _ - - -def _bin_op(name, doc="binary operator"): - """ Create a method for given binary operator - """ - def _(self, other): - jc = other._jc if isinstance(other, Column) else other - njc = getattr(self._jc, name)(jc) - return Column(njc, self.sql_ctx) - _.__doc__ = doc - return _ - - -def _reverse_op(name, doc="binary operator"): - """ Create a method for binary operator (this object is on right side) - """ - def _(self, other): - jother = _create_column_from_literal(other) - jc = getattr(jother, name)(self._jc) - return Column(jc, self.sql_ctx) - _.__doc__ = doc - return _ - - -class Column(DataFrame): - - """ - A column in a DataFrame. - - `Column` instances can be created by:: - - # 1. Select a column out of a DataFrame - df.colName - df["colName"] - - # 2. Create from an expression - df.colName + 1 - 1 / df.colName - """ - - def __init__(self, jc, sql_ctx=None): - self._jc = jc - super(Column, self).__init__(jc, sql_ctx) - - # arithmetic operators - __neg__ = _dsl_op("negate") - __add__ = _bin_op("plus") - __sub__ = _bin_op("minus") - __mul__ = _bin_op("multiply") - __div__ = _bin_op("divide") - __mod__ = _bin_op("mod") - __radd__ = _bin_op("plus") - __rsub__ = _reverse_op("minus") - __rmul__ = _bin_op("multiply") - __rdiv__ = _reverse_op("divide") - __rmod__ = _reverse_op("mod") - - # logistic operators - __eq__ = _bin_op("equalTo") - __ne__ = _bin_op("notEqual") - __lt__ = _bin_op("lt") - __le__ = _bin_op("leq") - __ge__ = _bin_op("geq") - __gt__ = _bin_op("gt") - - # `and`, `or`, `not` cannot be overloaded in Python, - # so use bitwise operators as boolean operators - __and__ = _bin_op('and') - __or__ = _bin_op('or') - __invert__ = _dsl_op('not') - __rand__ = _bin_op("and") - __ror__ = _bin_op("or") - - # container operators - __contains__ = _bin_op("contains") - __getitem__ = _bin_op("getItem") - getField = _bin_op("getField", "An expression that gets a field by name in a StructField.") - - # string methods - rlike = _bin_op("rlike") - like = _bin_op("like") - startswith = _bin_op("startsWith") - endswith = _bin_op("endsWith") - - def substr(self, startPos, length): - """ - Return a Column which is a substring of the column - - :param startPos: start position (int or Column) - :param length: length of the substring (int or Column) - - >>> df.name.substr(1, 3).collect() - [Row(col=u'Ali'), Row(col=u'Bob')] - """ - if type(startPos) != type(length): - raise TypeError("Can not mix the type") - if isinstance(startPos, (int, long)): - jc = self._jc.substr(startPos, length) - elif isinstance(startPos, Column): - jc = self._jc.substr(startPos._jc, length._jc) - else: - raise TypeError("Unexpected type: %s" % type(startPos)) - return Column(jc, self.sql_ctx) - - __getslice__ = substr - - # order - asc = _unary_op("asc") - desc = _unary_op("desc") - - isNull = _unary_op("isNull", "True if the current expression is null.") - isNotNull = _unary_op("isNotNull", "True if the current expression is not null.") - - def alias(self, alias): - """Return a alias for this column - - >>> df.age.alias("age2").collect() - [Row(age2=2), Row(age2=5)] - """ - return Column(getattr(self._jc, "as")(alias), self.sql_ctx) - - def cast(self, dataType): - """ Convert the column into type `dataType` - - >>> df.select(df.age.cast("string").alias('ages')).collect() - [Row(ages=u'2'), Row(ages=u'5')] - >>> df.select(df.age.cast(StringType()).alias('ages')).collect() - [Row(ages=u'2'), Row(ages=u'5')] - """ - if self.sql_ctx is None: - sc = SparkContext._active_spark_context - ssql_ctx = sc._jvm.SQLContext(sc._jsc.sc()) - else: - ssql_ctx = self.sql_ctx._ssql_ctx - if isinstance(dataType, basestring): - jc = self._jc.cast(dataType) - elif isinstance(dataType, DataType): - jdt = ssql_ctx.parseDataType(dataType.json()) - jc = self._jc.cast(jdt) - return Column(jc, self.sql_ctx) - - -def _aggregate_func(name, doc=""): - """ Create a function for aggregator by name""" - def _(col): - sc = SparkContext._active_spark_context - jc = getattr(sc._jvm.Dsl, name)(_to_java_column(col)) - return Column(jc) - _.__name__ = name - _.__doc__ = doc - return staticmethod(_) - - -class UserDefinedFunction(object): - def __init__(self, func, returnType): - self.func = func - self.returnType = returnType - self._broadcast = None - self._judf = self._create_judf() - - def _create_judf(self): - f = self.func # put it in closure `func` - func = lambda _, it: imap(lambda x: f(*x), it) - ser = AutoBatchedSerializer(PickleSerializer()) - command = (func, None, ser, ser) - sc = SparkContext._active_spark_context - pickled_command, broadcast_vars, env, includes = _prepare_for_python_RDD(sc, command, self) - ssql_ctx = sc._jvm.SQLContext(sc._jsc.sc()) - jdt = ssql_ctx.parseDataType(self.returnType.json()) - judf = sc._jvm.UserDefinedPythonFunction(f.__name__, bytearray(pickled_command), env, - includes, sc.pythonExec, broadcast_vars, - sc._javaAccumulator, jdt) - return judf - - def __del__(self): - if self._broadcast is not None: - self._broadcast.unpersist() - self._broadcast = None - - def __call__(self, *cols): - sc = SparkContext._active_spark_context - jcols = ListConverter().convert([_to_java_column(c) for c in cols], - sc._gateway._gateway_client) - jc = self._judf.apply(sc._jvm.PythonUtils.toSeq(jcols)) - return Column(jc) - - -class Dsl(object): - """ - A collections of builtin aggregators - """ - DSLS = { - 'lit': 'Creates a :class:`Column` of literal value.', - 'col': 'Returns a :class:`Column` based on the given column name.', - 'column': 'Returns a :class:`Column` based on the given column name.', - 'upper': 'Converts a string expression to upper case.', - 'lower': 'Converts a string expression to upper case.', - 'sqrt': 'Computes the square root of the specified float value.', - 'abs': 'Computes the absolutle value.', - - 'max': 'Aggregate function: returns the maximum value of the expression in a group.', - 'min': 'Aggregate function: returns the minimum value of the expression in a group.', - 'first': 'Aggregate function: returns the first value in a group.', - 'last': 'Aggregate function: returns the last value in a group.', - 'count': 'Aggregate function: returns the number of items in a group.', - 'sum': 'Aggregate function: returns the sum of all values in the expression.', - 'avg': 'Aggregate function: returns the average of the values in a group.', - 'mean': 'Aggregate function: returns the average of the values in a group.', - 'sumDistinct': 'Aggregate function: returns the sum of distinct values in the expression.', - } - - for _name, _doc in DSLS.items(): - locals()[_name] = _aggregate_func(_name, _doc) - del _name, _doc - - @staticmethod - def countDistinct(col, *cols): - """ Return a new Column for distinct count of (col, *cols) - - >>> from pyspark.sql import Dsl - >>> df.agg(Dsl.countDistinct(df.age, df.name).alias('c')).collect() - [Row(c=2)] - - >>> df.agg(Dsl.countDistinct("age", "name").alias('c')).collect() - [Row(c=2)] - """ - sc = SparkContext._active_spark_context - jcols = ListConverter().convert([_to_java_column(c) for c in cols], - sc._gateway._gateway_client) - jc = sc._jvm.Dsl.countDistinct(_to_java_column(col), - sc._jvm.PythonUtils.toSeq(jcols)) - return Column(jc) - - @staticmethod - def approxCountDistinct(col, rsd=None): - """ Return a new Column for approxiate distinct count of (col, *cols) - - >>> from pyspark.sql import Dsl - >>> df.agg(Dsl.approxCountDistinct(df.age).alias('c')).collect() - [Row(c=2)] - """ - sc = SparkContext._active_spark_context - if rsd is None: - jc = sc._jvm.Dsl.approxCountDistinct(_to_java_column(col)) - else: - jc = sc._jvm.Dsl.approxCountDistinct(_to_java_column(col), rsd) - return Column(jc) - - @staticmethod - def udf(f, returnType=StringType()): - """Create a user defined function (UDF) - - >>> slen = Dsl.udf(lambda s: len(s), IntegerType()) - >>> df.select(slen(df.name).alias('slen')).collect() - [Row(slen=5), Row(slen=3)] - """ - return UserDefinedFunction(f, returnType) - - -def _test(): - import doctest - from pyspark.context import SparkContext - # let doctest run in pyspark.sql, so DataTypes can be picklable - import pyspark.sql - from pyspark.sql import Row, SQLContext - from pyspark.sql_tests import ExamplePoint, ExamplePointUDT - globs = pyspark.sql.__dict__.copy() - sc = SparkContext('local[4]', 'PythonTest') - globs['sc'] = sc - globs['sqlCtx'] = sqlCtx = SQLContext(sc) - globs['rdd'] = sc.parallelize( - [Row(field1=1, field2="row1"), - Row(field1=2, field2="row2"), - Row(field1=3, field2="row3")] - ) - rdd2 = sc.parallelize([Row(name='Alice', age=2), Row(name='Bob', age=5)]) - rdd3 = sc.parallelize([Row(name='Tom', height=80), Row(name='Bob', height=85)]) - globs['df'] = sqlCtx.inferSchema(rdd2) - globs['df2'] = sqlCtx.inferSchema(rdd3) - globs['ExamplePoint'] = ExamplePoint - globs['ExamplePointUDT'] = ExamplePointUDT - jsonStrings = [ - '{"field1": 1, "field2": "row1", "field3":{"field4":11}}', - '{"field1" : 2, "field3":{"field4":22, "field5": [10, 11]},' - '"field6":[{"field7": "row2"}]}', - '{"field1" : null, "field2": "row3", ' - '"field3":{"field4":33, "field5": []}}' - ] - globs['jsonStrings'] = jsonStrings - globs['json'] = sc.parallelize(jsonStrings) - (failure_count, test_count) = doctest.testmod( - pyspark.sql, globs=globs, optionflags=doctest.ELLIPSIS) - globs['sc'].stop() - if failure_count: - exit(-1) - - -if __name__ == "__main__": - _test() diff --git a/python/pyspark/sql/__init__.py b/python/pyspark/sql/__init__.py new file mode 100644 index 0000000000000..0a5ba00393aab --- /dev/null +++ b/python/pyspark/sql/__init__.py @@ -0,0 +1,42 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one or more +# contributor license agreements. See the NOTICE file distributed with +# this work for additional information regarding copyright ownership. +# The ASF licenses this file to You under the Apache License, Version 2.0 +# (the "License"); you may not use this file except in compliance with +# the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +""" +public classes of Spark SQL: + + - L{SQLContext} + Main entry point for SQL functionality. + - L{DataFrame} + A Resilient Distributed Dataset (RDD) with Schema information for the data contained. In + addition to normal RDD operations, DataFrames also support SQL. + - L{GroupedData} + - L{Column} + Column is a DataFrame with a single column. + - L{Row} + A Row of data returned by a Spark SQL query. + - L{HiveContext} + Main entry point for accessing data stored in Apache Hive.. +""" + +from pyspark.sql.context import SQLContext, HiveContext +from pyspark.sql.types import Row +from pyspark.sql.dataframe import DataFrame, GroupedData, Column, Dsl, SchemaRDD + +__all__ = [ + 'SQLContext', 'HiveContext', 'DataFrame', 'GroupedData', 'Column', 'Row', + 'Dsl', +] diff --git a/python/pyspark/sql/context.py b/python/pyspark/sql/context.py new file mode 100644 index 0000000000000..49f016a9cf2e9 --- /dev/null +++ b/python/pyspark/sql/context.py @@ -0,0 +1,642 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one or more +# contributor license agreements. See the NOTICE file distributed with +# this work for additional information regarding copyright ownership. +# The ASF licenses this file to You under the Apache License, Version 2.0 +# (the "License"); you may not use this file except in compliance with +# the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import warnings +import json +from array import array +from itertools import imap + +from py4j.protocol import Py4JError + +from pyspark.rdd import _prepare_for_python_RDD +from pyspark.serializers import AutoBatchedSerializer, PickleSerializer +from pyspark.sql.types import StringType, StructType, _verify_type, \ + _infer_schema, _has_nulltype, _merge_type, _create_converter, _python_to_sql_converter +from pyspark.sql.dataframe import DataFrame + +__all__ = ["SQLContext", "HiveContext"] + + +class SQLContext(object): + + """Main entry point for Spark SQL functionality. + + A SQLContext can be used create L{DataFrame}, register L{DataFrame} as + tables, execute SQL over tables, cache tables, and read parquet files. + """ + + def __init__(self, sparkContext, sqlContext=None): + """Create a new SQLContext. + + :param sparkContext: The SparkContext to wrap. + :param sqlContext: An optional JVM Scala SQLContext. If set, we do not instatiate a new + SQLContext in the JVM, instead we make all calls to this object. + + >>> df = sqlCtx.inferSchema(rdd) + >>> sqlCtx.inferSchema(df) # doctest: +IGNORE_EXCEPTION_DETAIL + Traceback (most recent call last): + ... + TypeError:... + + >>> bad_rdd = sc.parallelize([1,2,3]) + >>> sqlCtx.inferSchema(bad_rdd) # doctest: +IGNORE_EXCEPTION_DETAIL + Traceback (most recent call last): + ... + ValueError:... + + >>> from datetime import datetime + >>> allTypes = sc.parallelize([Row(i=1, s="string", d=1.0, l=1L, + ... b=True, list=[1, 2, 3], dict={"s": 0}, row=Row(a=1), + ... time=datetime(2014, 8, 1, 14, 1, 5))]) + >>> df = sqlCtx.inferSchema(allTypes) + >>> df.registerTempTable("allTypes") + >>> sqlCtx.sql('select i+1, d+1, not b, list[1], dict["s"], time, row.a ' + ... 'from allTypes where b and i > 0').collect() + [Row(c0=2, c1=2.0, c2=False, c3=2, c4=0...8, 1, 14, 1, 5), a=1)] + >>> df.map(lambda x: (x.i, x.s, x.d, x.l, x.b, x.time, + ... x.row.a, x.list)).collect() + [(1, u'string', 1.0, 1, True, ...(2014, 8, 1, 14, 1, 5), 1, [1, 2, 3])] + """ + self._sc = sparkContext + self._jsc = self._sc._jsc + self._jvm = self._sc._jvm + self._scala_SQLContext = sqlContext + + @property + def _ssql_ctx(self): + """Accessor for the JVM Spark SQL context. + + Subclasses can override this property to provide their own + JVM Contexts. + """ + if self._scala_SQLContext is None: + self._scala_SQLContext = self._jvm.SQLContext(self._jsc.sc()) + return self._scala_SQLContext + + def registerFunction(self, name, f, returnType=StringType()): + """Registers a lambda function as a UDF so it can be used in SQL statements. + + In addition to a name and the function itself, the return type can be optionally specified. + When the return type is not given it default to a string and conversion will automatically + be done. For any other return type, the produced object must match the specified type. + + >>> sqlCtx.registerFunction("stringLengthString", lambda x: len(x)) + >>> sqlCtx.sql("SELECT stringLengthString('test')").collect() + [Row(c0=u'4')] + >>> from pyspark.sql.types import IntegerType + >>> sqlCtx.registerFunction("stringLengthInt", lambda x: len(x), IntegerType()) + >>> sqlCtx.sql("SELECT stringLengthInt('test')").collect() + [Row(c0=4)] + """ + func = lambda _, it: imap(lambda x: f(*x), it) + ser = AutoBatchedSerializer(PickleSerializer()) + command = (func, None, ser, ser) + pickled_cmd, bvars, env, includes = _prepare_for_python_RDD(self._sc, command, self) + self._ssql_ctx.udf().registerPython(name, + bytearray(pickled_cmd), + env, + includes, + self._sc.pythonExec, + bvars, + self._sc._javaAccumulator, + returnType.json()) + + def inferSchema(self, rdd, samplingRatio=None): + """Infer and apply a schema to an RDD of L{Row}. + + When samplingRatio is specified, the schema is inferred by looking + at the types of each row in the sampled dataset. Otherwise, the + first 100 rows of the RDD are inspected. Nested collections are + supported, which can include array, dict, list, Row, tuple, + namedtuple, or object. + + Each row could be L{pyspark.sql.Row} object or namedtuple or objects. + Using top level dicts is deprecated, as dict is used to represent Maps. + + If a single column has multiple distinct inferred types, it may cause + runtime exceptions. + + >>> rdd = sc.parallelize( + ... [Row(field1=1, field2="row1"), + ... Row(field1=2, field2="row2"), + ... Row(field1=3, field2="row3")]) + >>> df = sqlCtx.inferSchema(rdd) + >>> df.collect()[0] + Row(field1=1, field2=u'row1') + + >>> NestedRow = Row("f1", "f2") + >>> nestedRdd1 = sc.parallelize([ + ... NestedRow(array('i', [1, 2]), {"row1": 1.0}), + ... NestedRow(array('i', [2, 3]), {"row2": 2.0})]) + >>> df = sqlCtx.inferSchema(nestedRdd1) + >>> df.collect() + [Row(f1=[1, 2], f2={u'row1': 1.0}), ..., f2={u'row2': 2.0})] + + >>> nestedRdd2 = sc.parallelize([ + ... NestedRow([[1, 2], [2, 3]], [1, 2]), + ... NestedRow([[2, 3], [3, 4]], [2, 3])]) + >>> df = sqlCtx.inferSchema(nestedRdd2) + >>> df.collect() + [Row(f1=[[1, 2], [2, 3]], f2=[1, 2]), ..., f2=[2, 3])] + + >>> from collections import namedtuple + >>> CustomRow = namedtuple('CustomRow', 'field1 field2') + >>> rdd = sc.parallelize( + ... [CustomRow(field1=1, field2="row1"), + ... CustomRow(field1=2, field2="row2"), + ... CustomRow(field1=3, field2="row3")]) + >>> df = sqlCtx.inferSchema(rdd) + >>> df.collect()[0] + Row(field1=1, field2=u'row1') + """ + + if isinstance(rdd, DataFrame): + raise TypeError("Cannot apply schema to DataFrame") + + first = rdd.first() + if not first: + raise ValueError("The first row in RDD is empty, " + "can not infer schema") + if type(first) is dict: + warnings.warn("Using RDD of dict to inferSchema is deprecated," + "please use pyspark.sql.Row instead") + + if samplingRatio is None: + schema = _infer_schema(first) + if _has_nulltype(schema): + for row in rdd.take(100)[1:]: + schema = _merge_type(schema, _infer_schema(row)) + if not _has_nulltype(schema): + break + else: + warnings.warn("Some of types cannot be determined by the " + "first 100 rows, please try again with sampling") + else: + if samplingRatio > 0.99: + rdd = rdd.sample(False, float(samplingRatio)) + schema = rdd.map(_infer_schema).reduce(_merge_type) + + converter = _create_converter(schema) + rdd = rdd.map(converter) + return self.applySchema(rdd, schema) + + def applySchema(self, rdd, schema): + """ + Applies the given schema to the given RDD of L{tuple} or L{list}. + + These tuples or lists can contain complex nested structures like + lists, maps or nested rows. + + The schema should be a StructType. + + It is important that the schema matches the types of the objects + in each row or exceptions could be thrown at runtime. + + >>> from pyspark.sql.types import * + >>> rdd2 = sc.parallelize([(1, "row1"), (2, "row2"), (3, "row3")]) + >>> schema = StructType([StructField("field1", IntegerType(), False), + ... StructField("field2", StringType(), False)]) + >>> df = sqlCtx.applySchema(rdd2, schema) + >>> sqlCtx.registerRDDAsTable(df, "table1") + >>> df2 = sqlCtx.sql("SELECT * from table1") + >>> df2.collect() + [Row(field1=1, field2=u'row1'),..., Row(field1=3, field2=u'row3')] + + >>> from datetime import date, datetime + >>> rdd = sc.parallelize([(127, -128L, -32768, 32767, 2147483647L, 1.0, + ... date(2010, 1, 1), + ... datetime(2010, 1, 1, 1, 1, 1), + ... {"a": 1}, (2,), [1, 2, 3], None)]) + >>> schema = StructType([ + ... StructField("byte1", ByteType(), False), + ... StructField("byte2", ByteType(), False), + ... StructField("short1", ShortType(), False), + ... StructField("short2", ShortType(), False), + ... StructField("int", IntegerType(), False), + ... StructField("float", FloatType(), False), + ... StructField("date", DateType(), False), + ... StructField("time", TimestampType(), False), + ... StructField("map", + ... MapType(StringType(), IntegerType(), False), False), + ... StructField("struct", + ... StructType([StructField("b", ShortType(), False)]), False), + ... StructField("list", ArrayType(ByteType(), False), False), + ... StructField("null", DoubleType(), True)]) + >>> df = sqlCtx.applySchema(rdd, schema) + >>> results = df.map( + ... lambda x: (x.byte1, x.byte2, x.short1, x.short2, x.int, x.float, x.date, + ... x.time, x.map["a"], x.struct.b, x.list, x.null)) + >>> results.collect()[0] # doctest: +NORMALIZE_WHITESPACE + (127, -128, -32768, 32767, 2147483647, 1.0, datetime.date(2010, 1, 1), + datetime.datetime(2010, 1, 1, 1, 1, 1), 1, 2, [1, 2, 3], None) + + >>> df.registerTempTable("table2") + >>> sqlCtx.sql( + ... "SELECT byte1 - 1 AS byte1, byte2 + 1 AS byte2, " + + ... "short1 + 1 AS short1, short2 - 1 AS short2, int - 1 AS int, " + + ... "float + 1.5 as float FROM table2").collect() + [Row(byte1=126, byte2=-127, short1=-32767, short2=32766, int=2147483646, float=2.5)] + + >>> from pyspark.sql.types import _parse_schema_abstract, _infer_schema_type + >>> rdd = sc.parallelize([(127, -32768, 1.0, + ... datetime(2010, 1, 1, 1, 1, 1), + ... {"a": 1}, (2,), [1, 2, 3])]) + >>> abstract = "byte short float time map{} struct(b) list[]" + >>> schema = _parse_schema_abstract(abstract) + >>> typedSchema = _infer_schema_type(rdd.first(), schema) + >>> df = sqlCtx.applySchema(rdd, typedSchema) + >>> df.collect() + [Row(byte=127, short=-32768, float=1.0, time=..., list=[1, 2, 3])] + """ + + if isinstance(rdd, DataFrame): + raise TypeError("Cannot apply schema to DataFrame") + + if not isinstance(schema, StructType): + raise TypeError("schema should be StructType") + + # take the first few rows to verify schema + rows = rdd.take(10) + # Row() cannot been deserialized by Pyrolite + if rows and isinstance(rows[0], tuple) and rows[0].__class__.__name__ == 'Row': + rdd = rdd.map(tuple) + rows = rdd.take(10) + + for row in rows: + _verify_type(row, schema) + + # convert python objects to sql data + converter = _python_to_sql_converter(schema) + rdd = rdd.map(converter) + + jrdd = self._jvm.SerDeUtil.toJavaArray(rdd._to_java_object_rdd()) + df = self._ssql_ctx.applySchemaToPythonRDD(jrdd.rdd(), schema.json()) + return DataFrame(df, self) + + def registerRDDAsTable(self, rdd, tableName): + """Registers the given RDD as a temporary table in the catalog. + + Temporary tables exist only during the lifetime of this instance of + SQLContext. + + >>> df = sqlCtx.inferSchema(rdd) + >>> sqlCtx.registerRDDAsTable(df, "table1") + """ + if (rdd.__class__ is DataFrame): + df = rdd._jdf + self._ssql_ctx.registerRDDAsTable(df, tableName) + else: + raise ValueError("Can only register DataFrame as table") + + def parquetFile(self, *paths): + """Loads a Parquet file, returning the result as a L{DataFrame}. + + >>> import tempfile, shutil + >>> parquetFile = tempfile.mkdtemp() + >>> shutil.rmtree(parquetFile) + >>> df = sqlCtx.inferSchema(rdd) + >>> df.saveAsParquetFile(parquetFile) + >>> df2 = sqlCtx.parquetFile(parquetFile) + >>> sorted(df.collect()) == sorted(df2.collect()) + True + """ + gateway = self._sc._gateway + jpath = paths[0] + jpaths = gateway.new_array(gateway.jvm.java.lang.String, len(paths) - 1) + for i in range(1, len(paths)): + jpaths[i] = paths[i] + jdf = self._ssql_ctx.parquetFile(jpath, jpaths) + return DataFrame(jdf, self) + + def jsonFile(self, path, schema=None, samplingRatio=1.0): + """ + Loads a text file storing one JSON object per line as a + L{DataFrame}. + + If the schema is provided, applies the given schema to this + JSON dataset. + + Otherwise, it samples the dataset with ratio `samplingRatio` to + determine the schema. + + >>> import tempfile, shutil + >>> jsonFile = tempfile.mkdtemp() + >>> shutil.rmtree(jsonFile) + >>> ofn = open(jsonFile, 'w') + >>> for json in jsonStrings: + ... print>>ofn, json + >>> ofn.close() + >>> df1 = sqlCtx.jsonFile(jsonFile) + >>> sqlCtx.registerRDDAsTable(df1, "table1") + >>> df2 = sqlCtx.sql( + ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " + ... "field6 as f4 from table1") + >>> for r in df2.collect(): + ... print r + Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) + Row(f1=2, f2=None, f3=Row(field4=22,..., f4=[Row(field7=u'row2')]) + Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) + + >>> df3 = sqlCtx.jsonFile(jsonFile, df1.schema()) + >>> sqlCtx.registerRDDAsTable(df3, "table2") + >>> df4 = sqlCtx.sql( + ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " + ... "field6 as f4 from table2") + >>> for r in df4.collect(): + ... print r + Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) + Row(f1=2, f2=None, f3=Row(field4=22,..., f4=[Row(field7=u'row2')]) + Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) + + >>> from pyspark.sql.types import * + >>> schema = StructType([ + ... StructField("field2", StringType(), True), + ... StructField("field3", + ... StructType([ + ... StructField("field5", + ... ArrayType(IntegerType(), False), True)]), False)]) + >>> df5 = sqlCtx.jsonFile(jsonFile, schema) + >>> sqlCtx.registerRDDAsTable(df5, "table3") + >>> df6 = sqlCtx.sql( + ... "SELECT field2 AS f1, field3.field5 as f2, " + ... "field3.field5[0] as f3 from table3") + >>> df6.collect() + [Row(f1=u'row1', f2=None, f3=None)...Row(f1=u'row3', f2=[], f3=None)] + """ + if schema is None: + df = self._ssql_ctx.jsonFile(path, samplingRatio) + else: + scala_datatype = self._ssql_ctx.parseDataType(schema.json()) + df = self._ssql_ctx.jsonFile(path, scala_datatype) + return DataFrame(df, self) + + def jsonRDD(self, rdd, schema=None, samplingRatio=1.0): + """Loads an RDD storing one JSON object per string as a L{DataFrame}. + + If the schema is provided, applies the given schema to this + JSON dataset. + + Otherwise, it samples the dataset with ratio `samplingRatio` to + determine the schema. + + >>> df1 = sqlCtx.jsonRDD(json) + >>> sqlCtx.registerRDDAsTable(df1, "table1") + >>> df2 = sqlCtx.sql( + ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " + ... "field6 as f4 from table1") + >>> for r in df2.collect(): + ... print r + Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) + Row(f1=2, f2=None, f3=Row(field4=22..., f4=[Row(field7=u'row2')]) + Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) + + >>> df3 = sqlCtx.jsonRDD(json, df1.schema()) + >>> sqlCtx.registerRDDAsTable(df3, "table2") + >>> df4 = sqlCtx.sql( + ... "SELECT field1 AS f1, field2 as f2, field3 as f3, " + ... "field6 as f4 from table2") + >>> for r in df4.collect(): + ... print r + Row(f1=1, f2=u'row1', f3=Row(field4=11, field5=None), f4=None) + Row(f1=2, f2=None, f3=Row(field4=22..., f4=[Row(field7=u'row2')]) + Row(f1=None, f2=u'row3', f3=Row(field4=33, field5=[]), f4=None) + + >>> from pyspark.sql.types import * + >>> schema = StructType([ + ... StructField("field2", StringType(), True), + ... StructField("field3", + ... StructType([ + ... StructField("field5", + ... ArrayType(IntegerType(), False), True)]), False)]) + >>> df5 = sqlCtx.jsonRDD(json, schema) + >>> sqlCtx.registerRDDAsTable(df5, "table3") + >>> df6 = sqlCtx.sql( + ... "SELECT field2 AS f1, field3.field5 as f2, " + ... "field3.field5[0] as f3 from table3") + >>> df6.collect() + [Row(f1=u'row1', f2=None,...Row(f1=u'row3', f2=[], f3=None)] + + >>> sqlCtx.jsonRDD(sc.parallelize(['{}', + ... '{"key0": {"key1": "value1"}}'])).collect() + [Row(key0=None), Row(key0=Row(key1=u'value1'))] + >>> sqlCtx.jsonRDD(sc.parallelize(['{"key0": null}', + ... '{"key0": {"key1": "value1"}}'])).collect() + [Row(key0=None), Row(key0=Row(key1=u'value1'))] + """ + + def func(iterator): + for x in iterator: + if not isinstance(x, basestring): + x = unicode(x) + if isinstance(x, unicode): + x = x.encode("utf-8") + yield x + keyed = rdd.mapPartitions(func) + keyed._bypass_serializer = True + jrdd = keyed._jrdd.map(self._jvm.BytesToString()) + if schema is None: + df = self._ssql_ctx.jsonRDD(jrdd.rdd(), samplingRatio) + else: + scala_datatype = self._ssql_ctx.parseDataType(schema.json()) + df = self._ssql_ctx.jsonRDD(jrdd.rdd(), scala_datatype) + return DataFrame(df, self) + + def sql(self, sqlQuery): + """Return a L{DataFrame} representing the result of the given query. + + >>> df = sqlCtx.inferSchema(rdd) + >>> sqlCtx.registerRDDAsTable(df, "table1") + >>> df2 = sqlCtx.sql("SELECT field1 AS f1, field2 as f2 from table1") + >>> df2.collect() + [Row(f1=1, f2=u'row1'), Row(f1=2, f2=u'row2'), Row(f1=3, f2=u'row3')] + """ + return DataFrame(self._ssql_ctx.sql(sqlQuery), self) + + def table(self, tableName): + """Returns the specified table as a L{DataFrame}. + + >>> df = sqlCtx.inferSchema(rdd) + >>> sqlCtx.registerRDDAsTable(df, "table1") + >>> df2 = sqlCtx.table("table1") + >>> sorted(df.collect()) == sorted(df2.collect()) + True + """ + return DataFrame(self._ssql_ctx.table(tableName), self) + + def cacheTable(self, tableName): + """Caches the specified table in-memory.""" + self._ssql_ctx.cacheTable(tableName) + + def uncacheTable(self, tableName): + """Removes the specified table from the in-memory cache.""" + self._ssql_ctx.uncacheTable(tableName) + + +class HiveContext(SQLContext): + + """A variant of Spark SQL that integrates with data stored in Hive. + + Configuration for Hive is read from hive-site.xml on the classpath. + It supports running both SQL and HiveQL commands. + """ + + def __init__(self, sparkContext, hiveContext=None): + """Create a new HiveContext. + + :param sparkContext: The SparkContext to wrap. + :param hiveContext: An optional JVM Scala HiveContext. If set, we do not instatiate a new + HiveContext in the JVM, instead we make all calls to this object. + """ + SQLContext.__init__(self, sparkContext) + + if hiveContext: + self._scala_HiveContext = hiveContext + + @property + def _ssql_ctx(self): + try: + if not hasattr(self, '_scala_HiveContext'): + self._scala_HiveContext = self._get_hive_ctx() + return self._scala_HiveContext + except Py4JError as e: + raise Exception("You must build Spark with Hive. " + "Export 'SPARK_HIVE=true' and run " + "build/sbt assembly", e) + + def _get_hive_ctx(self): + return self._jvm.HiveContext(self._jsc.sc()) + + +def _create_row(fields, values): + row = Row(*values) + row.__FIELDS__ = fields + return row + + +class Row(tuple): + + """ + A row in L{DataFrame}. The fields in it can be accessed like attributes. + + Row can be used to create a row object by using named arguments, + the fields will be sorted by names. + + >>> row = Row(name="Alice", age=11) + >>> row + Row(age=11, name='Alice') + >>> row.name, row.age + ('Alice', 11) + + Row also can be used to create another Row like class, then it + could be used to create Row objects, such as + + >>> Person = Row("name", "age") + >>> Person + + >>> Person("Alice", 11) + Row(name='Alice', age=11) + """ + + def __new__(self, *args, **kwargs): + if args and kwargs: + raise ValueError("Can not use both args " + "and kwargs to create Row") + if args: + # create row class or objects + return tuple.__new__(self, args) + + elif kwargs: + # create row objects + names = sorted(kwargs.keys()) + values = tuple(kwargs[n] for n in names) + row = tuple.__new__(self, values) + row.__FIELDS__ = names + return row + + else: + raise ValueError("No args or kwargs") + + def asDict(self): + """ + Return as an dict + """ + if not hasattr(self, "__FIELDS__"): + raise TypeError("Cannot convert a Row class into dict") + return dict(zip(self.__FIELDS__, self)) + + # let obect acs like class + def __call__(self, *args): + """create new Row object""" + return _create_row(self, args) + + def __getattr__(self, item): + if item.startswith("__"): + raise AttributeError(item) + try: + # it will be slow when it has many fields, + # but this will not be used in normal cases + idx = self.__FIELDS__.index(item) + return self[idx] + except IndexError: + raise AttributeError(item) + + def __reduce__(self): + if hasattr(self, "__FIELDS__"): + return (_create_row, (self.__FIELDS__, tuple(self))) + else: + return tuple.__reduce__(self) + + def __repr__(self): + if hasattr(self, "__FIELDS__"): + return "Row(%s)" % ", ".join("%s=%r" % (k, v) + for k, v in zip(self.__FIELDS__, self)) + else: + return "" % ", ".join(self) + + +def _test(): + import doctest + from pyspark.context import SparkContext + from pyspark.sql import Row, SQLContext + import pyspark.sql.context + globs = pyspark.sql.context.__dict__.copy() + sc = SparkContext('local[4]', 'PythonTest') + globs['sc'] = sc + globs['sqlCtx'] = sqlCtx = SQLContext(sc) + globs['rdd'] = sc.parallelize( + [Row(field1=1, field2="row1"), + Row(field1=2, field2="row2"), + Row(field1=3, field2="row3")] + ) + jsonStrings = [ + '{"field1": 1, "field2": "row1", "field3":{"field4":11}}', + '{"field1" : 2, "field3":{"field4":22, "field5": [10, 11]},' + '"field6":[{"field7": "row2"}]}', + '{"field1" : null, "field2": "row3", ' + '"field3":{"field4":33, "field5": []}}' + ] + globs['jsonStrings'] = jsonStrings + globs['json'] = sc.parallelize(jsonStrings) + (failure_count, test_count) = doctest.testmod( + pyspark.sql.context, globs=globs, optionflags=doctest.ELLIPSIS) + globs['sc'].stop() + if failure_count: + exit(-1) + + +if __name__ == "__main__": + _test() diff --git a/python/pyspark/sql/dataframe.py b/python/pyspark/sql/dataframe.py new file mode 100644 index 0000000000000..04be65fe241c4 --- /dev/null +++ b/python/pyspark/sql/dataframe.py @@ -0,0 +1,974 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one or more +# contributor license agreements. See the NOTICE file distributed with +# this work for additional information regarding copyright ownership. +# The ASF licenses this file to You under the Apache License, Version 2.0 +# (the "License"); you may not use this file except in compliance with +# the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import sys +import itertools +import warnings +import random +import os +from tempfile import NamedTemporaryFile +from itertools import imap + +from py4j.java_collections import ListConverter, MapConverter + +from pyspark.context import SparkContext +from pyspark.rdd import RDD, _prepare_for_python_RDD +from pyspark.serializers import BatchedSerializer, AutoBatchedSerializer, PickleSerializer, \ + UTF8Deserializer +from pyspark.storagelevel import StorageLevel +from pyspark.traceback_utils import SCCallSiteSync +from pyspark.sql.types import * +from pyspark.sql.types import _create_cls, _parse_datatype_json_string + + +__all__ = ["DataFrame", "GroupedData", "Column", "Dsl", "SchemaRDD"] + + +class DataFrame(object): + + """A collection of rows that have the same columns. + + A :class:`DataFrame` is equivalent to a relational table in Spark SQL, + and can be created using various functions in :class:`SQLContext`:: + + people = sqlContext.parquetFile("...") + + Once created, it can be manipulated using the various domain-specific-language + (DSL) functions defined in: :class:`DataFrame`, :class:`Column`. + + To select a column from the data frame, use the apply method:: + + ageCol = people.age + + Note that the :class:`Column` type can also be manipulated + through its various functions:: + + # The following creates a new column that increases everybody's age by 10. + people.age + 10 + + + A more concrete example:: + + # To create DataFrame using SQLContext + people = sqlContext.parquetFile("...") + department = sqlContext.parquetFile("...") + + people.filter(people.age > 30).join(department, people.deptId == department.id)) \ + .groupBy(department.name, "gender").agg({"salary": "avg", "age": "max"}) + """ + + def __init__(self, jdf, sql_ctx): + self._jdf = jdf + self.sql_ctx = sql_ctx + self._sc = sql_ctx and sql_ctx._sc + self.is_cached = False + + @property + def rdd(self): + """ + Return the content of the :class:`DataFrame` as an :class:`RDD` + of :class:`Row` s. + """ + if not hasattr(self, '_lazy_rdd'): + jrdd = self._jdf.javaToPython() + rdd = RDD(jrdd, self.sql_ctx._sc, BatchedSerializer(PickleSerializer())) + schema = self.schema() + + def applySchema(it): + cls = _create_cls(schema) + return itertools.imap(cls, it) + + self._lazy_rdd = rdd.mapPartitions(applySchema) + + return self._lazy_rdd + + def toJSON(self, use_unicode=False): + """Convert a DataFrame into a MappedRDD of JSON documents; one document per row. + + >>> df.toJSON().first() + '{"age":2,"name":"Alice"}' + """ + rdd = self._jdf.toJSON() + return RDD(rdd.toJavaRDD(), self._sc, UTF8Deserializer(use_unicode)) + + def saveAsParquetFile(self, path): + """Save the contents as a Parquet file, preserving the schema. + + Files that are written out using this method can be read back in as + a DataFrame using the L{SQLContext.parquetFile} method. + + >>> import tempfile, shutil + >>> parquetFile = tempfile.mkdtemp() + >>> shutil.rmtree(parquetFile) + >>> df.saveAsParquetFile(parquetFile) + >>> df2 = sqlCtx.parquetFile(parquetFile) + >>> sorted(df2.collect()) == sorted(df.collect()) + True + """ + self._jdf.saveAsParquetFile(path) + + def registerTempTable(self, name): + """Registers this RDD as a temporary table using the given name. + + The lifetime of this temporary table is tied to the L{SQLContext} + that was used to create this DataFrame. + + >>> df.registerTempTable("people") + >>> df2 = sqlCtx.sql("select * from people") + >>> sorted(df.collect()) == sorted(df2.collect()) + True + """ + self._jdf.registerTempTable(name) + + def registerAsTable(self, name): + """DEPRECATED: use registerTempTable() instead""" + warnings.warn("Use registerTempTable instead of registerAsTable.", DeprecationWarning) + self.registerTempTable(name) + + def insertInto(self, tableName, overwrite=False): + """Inserts the contents of this DataFrame into the specified table. + + Optionally overwriting any existing data. + """ + self._jdf.insertInto(tableName, overwrite) + + def saveAsTable(self, tableName): + """Creates a new table with the contents of this DataFrame.""" + self._jdf.saveAsTable(tableName) + + def schema(self): + """Returns the schema of this DataFrame (represented by + a L{StructType}). + + >>> df.schema() + StructType(List(StructField(age,IntegerType,true),StructField(name,StringType,true))) + """ + return _parse_datatype_json_string(self._jdf.schema().json()) + + def printSchema(self): + """Prints out the schema in the tree format. + + >>> df.printSchema() + root + |-- age: integer (nullable = true) + |-- name: string (nullable = true) + + """ + print (self._jdf.schema().treeString()) + + def count(self): + """Return the number of elements in this RDD. + + Unlike the base RDD implementation of count, this implementation + leverages the query optimizer to compute the count on the DataFrame, + which supports features such as filter pushdown. + + >>> df.count() + 2L + """ + return self._jdf.count() + + def collect(self): + """Return a list that contains all of the rows. + + Each object in the list is a Row, the fields can be accessed as + attributes. + + >>> df.collect() + [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] + """ + with SCCallSiteSync(self._sc) as css: + bytesInJava = self._jdf.javaToPython().collect().iterator() + tempFile = NamedTemporaryFile(delete=False, dir=self._sc._temp_dir) + tempFile.close() + self._sc._writeToFile(bytesInJava, tempFile.name) + # Read the data into Python and deserialize it: + with open(tempFile.name, 'rb') as tempFile: + rs = list(BatchedSerializer(PickleSerializer()).load_stream(tempFile)) + os.unlink(tempFile.name) + cls = _create_cls(self.schema()) + return [cls(r) for r in rs] + + def limit(self, num): + """Limit the result count to the number specified. + + >>> df.limit(1).collect() + [Row(age=2, name=u'Alice')] + >>> df.limit(0).collect() + [] + """ + jdf = self._jdf.limit(num) + return DataFrame(jdf, self.sql_ctx) + + def take(self, num): + """Take the first num rows of the RDD. + + Each object in the list is a Row, the fields can be accessed as + attributes. + + >>> df.take(2) + [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] + """ + return self.limit(num).collect() + + def map(self, f): + """ Return a new RDD by applying a function to each Row, it's a + shorthand for df.rdd.map() + + >>> df.map(lambda p: p.name).collect() + [u'Alice', u'Bob'] + """ + return self.rdd.map(f) + + def mapPartitions(self, f, preservesPartitioning=False): + """ + Return a new RDD by applying a function to each partition. + + >>> rdd = sc.parallelize([1, 2, 3, 4], 4) + >>> def f(iterator): yield 1 + >>> rdd.mapPartitions(f).sum() + 4 + """ + return self.rdd.mapPartitions(f, preservesPartitioning) + + def cache(self): + """ Persist with the default storage level (C{MEMORY_ONLY_SER}). + """ + self.is_cached = True + self._jdf.cache() + return self + + def persist(self, storageLevel=StorageLevel.MEMORY_ONLY_SER): + """ Set the storage level to persist its values across operations + after the first time it is computed. This can only be used to assign + a new storage level if the RDD does not have a storage level set yet. + If no storage level is specified defaults to (C{MEMORY_ONLY_SER}). + """ + self.is_cached = True + javaStorageLevel = self._sc._getJavaStorageLevel(storageLevel) + self._jdf.persist(javaStorageLevel) + return self + + def unpersist(self, blocking=True): + """ Mark it as non-persistent, and remove all blocks for it from + memory and disk. + """ + self.is_cached = False + self._jdf.unpersist(blocking) + return self + + # def coalesce(self, numPartitions, shuffle=False): + # rdd = self._jdf.coalesce(numPartitions, shuffle, None) + # return DataFrame(rdd, self.sql_ctx) + + def repartition(self, numPartitions): + """ Return a new :class:`DataFrame` that has exactly `numPartitions` + partitions. + """ + rdd = self._jdf.repartition(numPartitions, None) + return DataFrame(rdd, self.sql_ctx) + + def sample(self, withReplacement, fraction, seed=None): + """ + Return a sampled subset of this DataFrame. + + >>> df.sample(False, 0.5, 97).count() + 1L + """ + assert fraction >= 0.0, "Negative fraction value: %s" % fraction + seed = seed if seed is not None else random.randint(0, sys.maxint) + rdd = self._jdf.sample(withReplacement, fraction, long(seed)) + return DataFrame(rdd, self.sql_ctx) + + # def takeSample(self, withReplacement, num, seed=None): + # """Return a fixed-size sampled subset of this DataFrame. + # + # >>> df = sqlCtx.inferSchema(rdd) + # >>> df.takeSample(False, 2, 97) + # [Row(field1=3, field2=u'row3'), Row(field1=1, field2=u'row1')] + # """ + # seed = seed if seed is not None else random.randint(0, sys.maxint) + # with SCCallSiteSync(self.context) as css: + # bytesInJava = self._jdf \ + # .takeSampleToPython(withReplacement, num, long(seed)) \ + # .iterator() + # cls = _create_cls(self.schema()) + # return map(cls, self._collect_iterator_through_file(bytesInJava)) + + @property + def dtypes(self): + """Return all column names and their data types as a list. + + >>> df.dtypes + [('age', 'integer'), ('name', 'string')] + """ + return [(str(f.name), f.dataType.jsonValue()) for f in self.schema().fields] + + @property + def columns(self): + """ Return all column names as a list. + + >>> df.columns + [u'age', u'name'] + """ + return [f.name for f in self.schema().fields] + + def join(self, other, joinExprs=None, joinType=None): + """ + Join with another DataFrame, using the given join expression. + The following performs a full outer join between `df1` and `df2`:: + + :param other: Right side of the join + :param joinExprs: Join expression + :param joinType: One of `inner`, `outer`, `left_outer`, `right_outer`, `semijoin`. + + >>> df.join(df2, df.name == df2.name, 'outer').select(df.name, df2.height).collect() + [Row(name=None, height=80), Row(name=u'Bob', height=85), Row(name=u'Alice', height=None)] + """ + + if joinExprs is None: + jdf = self._jdf.join(other._jdf) + else: + assert isinstance(joinExprs, Column), "joinExprs should be Column" + if joinType is None: + jdf = self._jdf.join(other._jdf, joinExprs._jc) + else: + assert isinstance(joinType, basestring), "joinType should be basestring" + jdf = self._jdf.join(other._jdf, joinExprs._jc, joinType) + return DataFrame(jdf, self.sql_ctx) + + def sort(self, *cols): + """ Return a new :class:`DataFrame` sorted by the specified column. + + :param cols: The columns or expressions used for sorting + + >>> df.sort(df.age.desc()).collect() + [Row(age=5, name=u'Bob'), Row(age=2, name=u'Alice')] + >>> df.sortBy(df.age.desc()).collect() + [Row(age=5, name=u'Bob'), Row(age=2, name=u'Alice')] + """ + if not cols: + raise ValueError("should sort by at least one column") + jcols = ListConverter().convert([_to_java_column(c) for c in cols], + self._sc._gateway._gateway_client) + jdf = self._jdf.sort(self._sc._jvm.PythonUtils.toSeq(jcols)) + return DataFrame(jdf, self.sql_ctx) + + sortBy = sort + + def head(self, n=None): + """ Return the first `n` rows or the first row if n is None. + + >>> df.head() + Row(age=2, name=u'Alice') + >>> df.head(1) + [Row(age=2, name=u'Alice')] + """ + if n is None: + rs = self.head(1) + return rs[0] if rs else None + return self.take(n) + + def first(self): + """ Return the first row. + + >>> df.first() + Row(age=2, name=u'Alice') + """ + return self.head() + + def __getitem__(self, item): + """ Return the column by given name + + >>> df['age'].collect() + [Row(age=2), Row(age=5)] + >>> df[ ["name", "age"]].collect() + [Row(name=u'Alice', age=2), Row(name=u'Bob', age=5)] + >>> df[ df.age > 3 ].collect() + [Row(age=5, name=u'Bob')] + """ + if isinstance(item, basestring): + jc = self._jdf.apply(item) + return Column(jc, self.sql_ctx) + elif isinstance(item, Column): + return self.filter(item) + elif isinstance(item, list): + return self.select(*item) + else: + raise IndexError("unexpected index: %s" % item) + + def __getattr__(self, name): + """ Return the column by given name + + >>> df.age.collect() + [Row(age=2), Row(age=5)] + """ + if name.startswith("__"): + raise AttributeError(name) + jc = self._jdf.apply(name) + return Column(jc, self.sql_ctx) + + def select(self, *cols): + """ Selecting a set of expressions. + + >>> df.select().collect() + [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] + >>> df.select('*').collect() + [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')] + >>> df.select('name', 'age').collect() + [Row(name=u'Alice', age=2), Row(name=u'Bob', age=5)] + >>> df.select(df.name, (df.age + 10).alias('age')).collect() + [Row(name=u'Alice', age=12), Row(name=u'Bob', age=15)] + """ + if not cols: + cols = ["*"] + jcols = ListConverter().convert([_to_java_column(c) for c in cols], + self._sc._gateway._gateway_client) + jdf = self._jdf.select(self.sql_ctx._sc._jvm.PythonUtils.toSeq(jcols)) + return DataFrame(jdf, self.sql_ctx) + + def selectExpr(self, *expr): + """ + Selects a set of SQL expressions. This is a variant of + `select` that accepts SQL expressions. + + >>> df.selectExpr("age * 2", "abs(age)").collect() + [Row((age * 2)=4, Abs(age)=2), Row((age * 2)=10, Abs(age)=5)] + """ + jexpr = ListConverter().convert(expr, self._sc._gateway._gateway_client) + jdf = self._jdf.selectExpr(self._sc._jvm.PythonUtils.toSeq(jexpr)) + return DataFrame(jdf, self.sql_ctx) + + def filter(self, condition): + """ Filtering rows using the given condition, which could be + Column expression or string of SQL expression. + + where() is an alias for filter(). + + >>> df.filter(df.age > 3).collect() + [Row(age=5, name=u'Bob')] + >>> df.where(df.age == 2).collect() + [Row(age=2, name=u'Alice')] + + >>> df.filter("age > 3").collect() + [Row(age=5, name=u'Bob')] + >>> df.where("age = 2").collect() + [Row(age=2, name=u'Alice')] + """ + if isinstance(condition, basestring): + jdf = self._jdf.filter(condition) + elif isinstance(condition, Column): + jdf = self._jdf.filter(condition._jc) + else: + raise TypeError("condition should be string or Column") + return DataFrame(jdf, self.sql_ctx) + + where = filter + + def groupBy(self, *cols): + """ Group the :class:`DataFrame` using the specified columns, + so we can run aggregation on them. See :class:`GroupedData` + for all the available aggregate functions. + + >>> df.groupBy().avg().collect() + [Row(AVG(age#0)=3.5)] + >>> df.groupBy('name').agg({'age': 'mean'}).collect() + [Row(name=u'Bob', AVG(age#0)=5.0), Row(name=u'Alice', AVG(age#0)=2.0)] + >>> df.groupBy(df.name).avg().collect() + [Row(name=u'Bob', AVG(age#0)=5.0), Row(name=u'Alice', AVG(age#0)=2.0)] + """ + jcols = ListConverter().convert([_to_java_column(c) for c in cols], + self._sc._gateway._gateway_client) + jdf = self._jdf.groupBy(self.sql_ctx._sc._jvm.PythonUtils.toSeq(jcols)) + return GroupedData(jdf, self.sql_ctx) + + def agg(self, *exprs): + """ Aggregate on the entire :class:`DataFrame` without groups + (shorthand for df.groupBy.agg()). + + >>> df.agg({"age": "max"}).collect() + [Row(MAX(age#0)=5)] + >>> from pyspark.sql import Dsl + >>> df.agg(Dsl.min(df.age)).collect() + [Row(MIN(age#0)=2)] + """ + return self.groupBy().agg(*exprs) + + def unionAll(self, other): + """ Return a new DataFrame containing union of rows in this + frame and another frame. + + This is equivalent to `UNION ALL` in SQL. + """ + return DataFrame(self._jdf.unionAll(other._jdf), self.sql_ctx) + + def intersect(self, other): + """ Return a new :class:`DataFrame` containing rows only in + both this frame and another frame. + + This is equivalent to `INTERSECT` in SQL. + """ + return DataFrame(self._jdf.intersect(other._jdf), self.sql_ctx) + + def subtract(self, other): + """ Return a new :class:`DataFrame` containing rows in this frame + but not in another frame. + + This is equivalent to `EXCEPT` in SQL. + """ + return DataFrame(getattr(self._jdf, "except")(other._jdf), self.sql_ctx) + + def addColumn(self, colName, col): + """ Return a new :class:`DataFrame` by adding a column. + + >>> df.addColumn('age2', df.age + 2).collect() + [Row(age=2, name=u'Alice', age2=4), Row(age=5, name=u'Bob', age2=7)] + """ + return self.select('*', col.alias(colName)) + + def to_pandas(self): + """ + Collect all the rows and return a `pandas.DataFrame`. + + >>> df.to_pandas() # doctest: +SKIP + age name + 0 2 Alice + 1 5 Bob + """ + import pandas as pd + return pd.DataFrame.from_records(self.collect(), columns=self.columns) + + +# Having SchemaRDD for backward compatibility (for docs) +class SchemaRDD(DataFrame): + """ + SchemaRDD is deprecated, please use DataFrame + """ + + +def dfapi(f): + def _api(self): + name = f.__name__ + jdf = getattr(self._jdf, name)() + return DataFrame(jdf, self.sql_ctx) + _api.__name__ = f.__name__ + _api.__doc__ = f.__doc__ + return _api + + +class GroupedData(object): + + """ + A set of methods for aggregations on a :class:`DataFrame`, + created by DataFrame.groupBy(). + """ + + def __init__(self, jdf, sql_ctx): + self._jdf = jdf + self.sql_ctx = sql_ctx + + def agg(self, *exprs): + """ Compute aggregates by specifying a map from column name + to aggregate methods. + + The available aggregate methods are `avg`, `max`, `min`, + `sum`, `count`. + + :param exprs: list or aggregate columns or a map from column + name to aggregate methods. + + >>> gdf = df.groupBy(df.name) + >>> gdf.agg({"age": "max"}).collect() + [Row(name=u'Bob', MAX(age#0)=5), Row(name=u'Alice', MAX(age#0)=2)] + >>> from pyspark.sql import Dsl + >>> gdf.agg(Dsl.min(df.age)).collect() + [Row(MIN(age#0)=5), Row(MIN(age#0)=2)] + """ + assert exprs, "exprs should not be empty" + if len(exprs) == 1 and isinstance(exprs[0], dict): + jmap = MapConverter().convert(exprs[0], + self.sql_ctx._sc._gateway._gateway_client) + jdf = self._jdf.agg(jmap) + else: + # Columns + assert all(isinstance(c, Column) for c in exprs), "all exprs should be Column" + jcols = ListConverter().convert([c._jc for c in exprs[1:]], + self.sql_ctx._sc._gateway._gateway_client) + jdf = self._jdf.agg(exprs[0]._jc, self.sql_ctx._sc._jvm.PythonUtils.toSeq(jcols)) + return DataFrame(jdf, self.sql_ctx) + + @dfapi + def count(self): + """ Count the number of rows for each group. + + >>> df.groupBy(df.age).count().collect() + [Row(age=2, count=1), Row(age=5, count=1)] + """ + + @dfapi + def mean(self): + """Compute the average value for each numeric columns + for each group. This is an alias for `avg`.""" + + @dfapi + def avg(self): + """Compute the average value for each numeric columns + for each group.""" + + @dfapi + def max(self): + """Compute the max value for each numeric columns for + each group. """ + + @dfapi + def min(self): + """Compute the min value for each numeric column for + each group.""" + + @dfapi + def sum(self): + """Compute the sum for each numeric columns for each + group.""" + + +def _create_column_from_literal(literal): + sc = SparkContext._active_spark_context + return sc._jvm.Dsl.lit(literal) + + +def _create_column_from_name(name): + sc = SparkContext._active_spark_context + return sc._jvm.Dsl.col(name) + + +def _to_java_column(col): + if isinstance(col, Column): + jcol = col._jc + else: + jcol = _create_column_from_name(col) + return jcol + + +def _unary_op(name, doc="unary operator"): + """ Create a method for given unary operator """ + def _(self): + jc = getattr(self._jc, name)() + return Column(jc, self.sql_ctx) + _.__doc__ = doc + return _ + + +def _dsl_op(name, doc=''): + def _(self): + jc = getattr(self._sc._jvm.Dsl, name)(self._jc) + return Column(jc, self.sql_ctx) + _.__doc__ = doc + return _ + + +def _bin_op(name, doc="binary operator"): + """ Create a method for given binary operator + """ + def _(self, other): + jc = other._jc if isinstance(other, Column) else other + njc = getattr(self._jc, name)(jc) + return Column(njc, self.sql_ctx) + _.__doc__ = doc + return _ + + +def _reverse_op(name, doc="binary operator"): + """ Create a method for binary operator (this object is on right side) + """ + def _(self, other): + jother = _create_column_from_literal(other) + jc = getattr(jother, name)(self._jc) + return Column(jc, self.sql_ctx) + _.__doc__ = doc + return _ + + +class Column(DataFrame): + + """ + A column in a DataFrame. + + `Column` instances can be created by:: + + # 1. Select a column out of a DataFrame + df.colName + df["colName"] + + # 2. Create from an expression + df.colName + 1 + 1 / df.colName + """ + + def __init__(self, jc, sql_ctx=None): + self._jc = jc + super(Column, self).__init__(jc, sql_ctx) + + # arithmetic operators + __neg__ = _dsl_op("negate") + __add__ = _bin_op("plus") + __sub__ = _bin_op("minus") + __mul__ = _bin_op("multiply") + __div__ = _bin_op("divide") + __mod__ = _bin_op("mod") + __radd__ = _bin_op("plus") + __rsub__ = _reverse_op("minus") + __rmul__ = _bin_op("multiply") + __rdiv__ = _reverse_op("divide") + __rmod__ = _reverse_op("mod") + + # logistic operators + __eq__ = _bin_op("equalTo") + __ne__ = _bin_op("notEqual") + __lt__ = _bin_op("lt") + __le__ = _bin_op("leq") + __ge__ = _bin_op("geq") + __gt__ = _bin_op("gt") + + # `and`, `or`, `not` cannot be overloaded in Python, + # so use bitwise operators as boolean operators + __and__ = _bin_op('and') + __or__ = _bin_op('or') + __invert__ = _dsl_op('not') + __rand__ = _bin_op("and") + __ror__ = _bin_op("or") + + # container operators + __contains__ = _bin_op("contains") + __getitem__ = _bin_op("getItem") + getField = _bin_op("getField", "An expression that gets a field by name in a StructField.") + + # string methods + rlike = _bin_op("rlike") + like = _bin_op("like") + startswith = _bin_op("startsWith") + endswith = _bin_op("endsWith") + + def substr(self, startPos, length): + """ + Return a Column which is a substring of the column + + :param startPos: start position (int or Column) + :param length: length of the substring (int or Column) + + >>> df.name.substr(1, 3).collect() + [Row(col=u'Ali'), Row(col=u'Bob')] + """ + if type(startPos) != type(length): + raise TypeError("Can not mix the type") + if isinstance(startPos, (int, long)): + jc = self._jc.substr(startPos, length) + elif isinstance(startPos, Column): + jc = self._jc.substr(startPos._jc, length._jc) + else: + raise TypeError("Unexpected type: %s" % type(startPos)) + return Column(jc, self.sql_ctx) + + __getslice__ = substr + + # order + asc = _unary_op("asc") + desc = _unary_op("desc") + + isNull = _unary_op("isNull", "True if the current expression is null.") + isNotNull = _unary_op("isNotNull", "True if the current expression is not null.") + + def alias(self, alias): + """Return a alias for this column + + >>> df.age.alias("age2").collect() + [Row(age2=2), Row(age2=5)] + """ + return Column(getattr(self._jc, "as")(alias), self.sql_ctx) + + def cast(self, dataType): + """ Convert the column into type `dataType` + + >>> df.select(df.age.cast("string").alias('ages')).collect() + [Row(ages=u'2'), Row(ages=u'5')] + >>> df.select(df.age.cast(StringType()).alias('ages')).collect() + [Row(ages=u'2'), Row(ages=u'5')] + """ + if self.sql_ctx is None: + sc = SparkContext._active_spark_context + ssql_ctx = sc._jvm.SQLContext(sc._jsc.sc()) + else: + ssql_ctx = self.sql_ctx._ssql_ctx + if isinstance(dataType, basestring): + jc = self._jc.cast(dataType) + elif isinstance(dataType, DataType): + jdt = ssql_ctx.parseDataType(dataType.json()) + jc = self._jc.cast(jdt) + return Column(jc, self.sql_ctx) + + def to_pandas(self): + """ + Return a pandas.Series from the column + + >>> df.age.to_pandas() # doctest: +SKIP + 0 2 + 1 5 + dtype: int64 + """ + import pandas as pd + data = [c for c, in self.collect()] + return pd.Series(data) + + +def _aggregate_func(name, doc=""): + """ Create a function for aggregator by name""" + def _(col): + sc = SparkContext._active_spark_context + jc = getattr(sc._jvm.Dsl, name)(_to_java_column(col)) + return Column(jc) + _.__name__ = name + _.__doc__ = doc + return staticmethod(_) + + +class UserDefinedFunction(object): + def __init__(self, func, returnType): + self.func = func + self.returnType = returnType + self._broadcast = None + self._judf = self._create_judf() + + def _create_judf(self): + f = self.func # put it in closure `func` + func = lambda _, it: imap(lambda x: f(*x), it) + ser = AutoBatchedSerializer(PickleSerializer()) + command = (func, None, ser, ser) + sc = SparkContext._active_spark_context + pickled_command, broadcast_vars, env, includes = _prepare_for_python_RDD(sc, command, self) + ssql_ctx = sc._jvm.SQLContext(sc._jsc.sc()) + jdt = ssql_ctx.parseDataType(self.returnType.json()) + judf = sc._jvm.UserDefinedPythonFunction(f.__name__, bytearray(pickled_command), env, + includes, sc.pythonExec, broadcast_vars, + sc._javaAccumulator, jdt) + return judf + + def __del__(self): + if self._broadcast is not None: + self._broadcast.unpersist() + self._broadcast = None + + def __call__(self, *cols): + sc = SparkContext._active_spark_context + jcols = ListConverter().convert([_to_java_column(c) for c in cols], + sc._gateway._gateway_client) + jc = self._judf.apply(sc._jvm.PythonUtils.toSeq(jcols)) + return Column(jc) + + +class Dsl(object): + """ + A collections of builtin aggregators + """ + DSLS = { + 'lit': 'Creates a :class:`Column` of literal value.', + 'col': 'Returns a :class:`Column` based on the given column name.', + 'column': 'Returns a :class:`Column` based on the given column name.', + 'upper': 'Converts a string expression to upper case.', + 'lower': 'Converts a string expression to upper case.', + 'sqrt': 'Computes the square root of the specified float value.', + 'abs': 'Computes the absolutle value.', + + 'max': 'Aggregate function: returns the maximum value of the expression in a group.', + 'min': 'Aggregate function: returns the minimum value of the expression in a group.', + 'first': 'Aggregate function: returns the first value in a group.', + 'last': 'Aggregate function: returns the last value in a group.', + 'count': 'Aggregate function: returns the number of items in a group.', + 'sum': 'Aggregate function: returns the sum of all values in the expression.', + 'avg': 'Aggregate function: returns the average of the values in a group.', + 'mean': 'Aggregate function: returns the average of the values in a group.', + 'sumDistinct': 'Aggregate function: returns the sum of distinct values in the expression.', + } + + for _name, _doc in DSLS.items(): + locals()[_name] = _aggregate_func(_name, _doc) + del _name, _doc + + @staticmethod + def countDistinct(col, *cols): + """ Return a new Column for distinct count of (col, *cols) + + >>> from pyspark.sql import Dsl + >>> df.agg(Dsl.countDistinct(df.age, df.name).alias('c')).collect() + [Row(c=2)] + + >>> df.agg(Dsl.countDistinct("age", "name").alias('c')).collect() + [Row(c=2)] + """ + sc = SparkContext._active_spark_context + jcols = ListConverter().convert([_to_java_column(c) for c in cols], + sc._gateway._gateway_client) + jc = sc._jvm.Dsl.countDistinct(_to_java_column(col), + sc._jvm.PythonUtils.toSeq(jcols)) + return Column(jc) + + @staticmethod + def approxCountDistinct(col, rsd=None): + """ Return a new Column for approxiate distinct count of (col, *cols) + + >>> from pyspark.sql import Dsl + >>> df.agg(Dsl.approxCountDistinct(df.age).alias('c')).collect() + [Row(c=2)] + """ + sc = SparkContext._active_spark_context + if rsd is None: + jc = sc._jvm.Dsl.approxCountDistinct(_to_java_column(col)) + else: + jc = sc._jvm.Dsl.approxCountDistinct(_to_java_column(col), rsd) + return Column(jc) + + @staticmethod + def udf(f, returnType=StringType()): + """Create a user defined function (UDF) + + >>> slen = Dsl.udf(lambda s: len(s), IntegerType()) + >>> df.select(slen(df.name).alias('slen')).collect() + [Row(slen=5), Row(slen=3)] + """ + return UserDefinedFunction(f, returnType) + + +def _test(): + import doctest + from pyspark.context import SparkContext + from pyspark.sql import Row, SQLContext + import pyspark.sql.dataframe + globs = pyspark.sql.dataframe.__dict__.copy() + sc = SparkContext('local[4]', 'PythonTest') + globs['sc'] = sc + globs['sqlCtx'] = sqlCtx = SQLContext(sc) + rdd2 = sc.parallelize([Row(name='Alice', age=2), Row(name='Bob', age=5)]) + rdd3 = sc.parallelize([Row(name='Tom', height=80), Row(name='Bob', height=85)]) + globs['df'] = sqlCtx.inferSchema(rdd2) + globs['df2'] = sqlCtx.inferSchema(rdd3) + (failure_count, test_count) = doctest.testmod( + pyspark.sql.dataframe, globs=globs, optionflags=doctest.ELLIPSIS) + globs['sc'].stop() + if failure_count: + exit(-1) + + +if __name__ == "__main__": + _test() diff --git a/python/pyspark/sql_tests.py b/python/pyspark/sql/tests.py similarity index 96% rename from python/pyspark/sql_tests.py rename to python/pyspark/sql/tests.py index d314f46e8d2d5..d25c6365ed067 100644 --- a/python/pyspark/sql_tests.py +++ b/python/pyspark/sql/tests.py @@ -34,8 +34,10 @@ else: import unittest -from pyspark.sql import SQLContext, IntegerType, Row, ArrayType, StructType, StructField, \ - UserDefinedType, DoubleType + +from pyspark.sql import SQLContext, Column +from pyspark.sql.types import IntegerType, Row, ArrayType, StructType, StructField, \ + UserDefinedType, DoubleType, LongType from pyspark.tests import ReusedPySparkTestCase @@ -220,7 +222,7 @@ def test_convert_row_to_dict(self): self.assertEqual(1.0, row.asDict()['d']['key'].c) def test_infer_schema_with_udt(self): - from pyspark.sql_tests import ExamplePoint, ExamplePointUDT + from pyspark.sql.tests import ExamplePoint, ExamplePointUDT row = Row(label=1.0, point=ExamplePoint(1.0, 2.0)) rdd = self.sc.parallelize([row]) df = self.sqlCtx.inferSchema(rdd) @@ -232,7 +234,7 @@ def test_infer_schema_with_udt(self): self.assertEqual(point, ExamplePoint(1.0, 2.0)) def test_apply_schema_with_udt(self): - from pyspark.sql_tests import ExamplePoint, ExamplePointUDT + from pyspark.sql.tests import ExamplePoint, ExamplePointUDT row = (1.0, ExamplePoint(1.0, 2.0)) rdd = self.sc.parallelize([row]) schema = StructType([StructField("label", DoubleType(), False), @@ -242,7 +244,7 @@ def test_apply_schema_with_udt(self): self.assertEquals(point, ExamplePoint(1.0, 2.0)) def test_parquet_with_udt(self): - from pyspark.sql_tests import ExamplePoint + from pyspark.sql.tests import ExamplePoint row = Row(label=1.0, point=ExamplePoint(1.0, 2.0)) rdd = self.sc.parallelize([row]) df0 = self.sqlCtx.inferSchema(rdd) @@ -253,7 +255,6 @@ def test_parquet_with_udt(self): self.assertEquals(point, ExamplePoint(1.0, 2.0)) def test_column_operators(self): - from pyspark.sql import Column, LongType ci = self.df.key cs = self.df.value c = ci == cs diff --git a/python/pyspark/sql/types.py b/python/pyspark/sql/types.py new file mode 100644 index 0000000000000..41afefe48ee5e --- /dev/null +++ b/python/pyspark/sql/types.py @@ -0,0 +1,1279 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one or more +# contributor license agreements. See the NOTICE file distributed with +# this work for additional information regarding copyright ownership. +# The ASF licenses this file to You under the Apache License, Version 2.0 +# (the "License"); you may not use this file except in compliance with +# the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import decimal +import datetime +import keyword +import warnings +import json +import re +from array import array +from operator import itemgetter + + +__all__ = [ + "DataType", "NullType", "StringType", "BinaryType", "BooleanType", "DateType", + "TimestampType", "DecimalType", "DoubleType", "FloatType", "ByteType", "IntegerType", + "LongType", "ShortType", "ArrayType", "MapType", "StructField", "StructType", ] + + +class DataType(object): + + """Spark SQL DataType""" + + def __repr__(self): + return self.__class__.__name__ + + def __hash__(self): + return hash(str(self)) + + def __eq__(self, other): + return (isinstance(other, self.__class__) and + self.__dict__ == other.__dict__) + + def __ne__(self, other): + return not self.__eq__(other) + + @classmethod + def typeName(cls): + return cls.__name__[:-4].lower() + + def jsonValue(self): + return self.typeName() + + def json(self): + return json.dumps(self.jsonValue(), + separators=(',', ':'), + sort_keys=True) + + +class PrimitiveTypeSingleton(type): + + """Metaclass for PrimitiveType""" + + _instances = {} + + def __call__(cls): + if cls not in cls._instances: + cls._instances[cls] = super(PrimitiveTypeSingleton, cls).__call__() + return cls._instances[cls] + + +class PrimitiveType(DataType): + + """Spark SQL PrimitiveType""" + + __metaclass__ = PrimitiveTypeSingleton + + def __eq__(self, other): + # because they should be the same object + return self is other + + +class NullType(PrimitiveType): + + """Spark SQL NullType + + The data type representing None, used for the types which has not + been inferred. + """ + + +class StringType(PrimitiveType): + + """Spark SQL StringType + + The data type representing string values. + """ + + +class BinaryType(PrimitiveType): + + """Spark SQL BinaryType + + The data type representing bytearray values. + """ + + +class BooleanType(PrimitiveType): + + """Spark SQL BooleanType + + The data type representing bool values. + """ + + +class DateType(PrimitiveType): + + """Spark SQL DateType + + The data type representing datetime.date values. + """ + + +class TimestampType(PrimitiveType): + + """Spark SQL TimestampType + + The data type representing datetime.datetime values. + """ + + +class DecimalType(DataType): + + """Spark SQL DecimalType + + The data type representing decimal.Decimal values. + """ + + def __init__(self, precision=None, scale=None): + self.precision = precision + self.scale = scale + self.hasPrecisionInfo = precision is not None + + def jsonValue(self): + if self.hasPrecisionInfo: + return "decimal(%d,%d)" % (self.precision, self.scale) + else: + return "decimal" + + def __repr__(self): + if self.hasPrecisionInfo: + return "DecimalType(%d,%d)" % (self.precision, self.scale) + else: + return "DecimalType()" + + +class DoubleType(PrimitiveType): + + """Spark SQL DoubleType + + The data type representing float values. + """ + + +class FloatType(PrimitiveType): + + """Spark SQL FloatType + + The data type representing single precision floating-point values. + """ + + +class ByteType(PrimitiveType): + + """Spark SQL ByteType + + The data type representing int values with 1 singed byte. + """ + + +class IntegerType(PrimitiveType): + + """Spark SQL IntegerType + + The data type representing int values. + """ + + +class LongType(PrimitiveType): + + """Spark SQL LongType + + The data type representing long values. If the any value is + beyond the range of [-9223372036854775808, 9223372036854775807], + please use DecimalType. + """ + + +class ShortType(PrimitiveType): + + """Spark SQL ShortType + + The data type representing int values with 2 signed bytes. + """ + + +class ArrayType(DataType): + + """Spark SQL ArrayType + + The data type representing list values. An ArrayType object + comprises two fields, elementType (a DataType) and containsNull (a bool). + The field of elementType is used to specify the type of array elements. + The field of containsNull is used to specify if the array has None values. + + """ + + def __init__(self, elementType, containsNull=True): + """Creates an ArrayType + + :param elementType: the data type of elements. + :param containsNull: indicates whether the list contains None values. + + >>> ArrayType(StringType) == ArrayType(StringType, True) + True + >>> ArrayType(StringType, False) == ArrayType(StringType) + False + """ + self.elementType = elementType + self.containsNull = containsNull + + def __repr__(self): + return "ArrayType(%s,%s)" % (self.elementType, + str(self.containsNull).lower()) + + def jsonValue(self): + return {"type": self.typeName(), + "elementType": self.elementType.jsonValue(), + "containsNull": self.containsNull} + + @classmethod + def fromJson(cls, json): + return ArrayType(_parse_datatype_json_value(json["elementType"]), + json["containsNull"]) + + +class MapType(DataType): + + """Spark SQL MapType + + The data type representing dict values. A MapType object comprises + three fields, keyType (a DataType), valueType (a DataType) and + valueContainsNull (a bool). + + The field of keyType is used to specify the type of keys in the map. + The field of valueType is used to specify the type of values in the map. + The field of valueContainsNull is used to specify if values of this + map has None values. + + For values of a MapType column, keys are not allowed to have None values. + + """ + + def __init__(self, keyType, valueType, valueContainsNull=True): + """Creates a MapType + :param keyType: the data type of keys. + :param valueType: the data type of values. + :param valueContainsNull: indicates whether values contains + null values. + + >>> (MapType(StringType, IntegerType) + ... == MapType(StringType, IntegerType, True)) + True + >>> (MapType(StringType, IntegerType, False) + ... == MapType(StringType, FloatType)) + False + """ + self.keyType = keyType + self.valueType = valueType + self.valueContainsNull = valueContainsNull + + def __repr__(self): + return "MapType(%s,%s,%s)" % (self.keyType, self.valueType, + str(self.valueContainsNull).lower()) + + def jsonValue(self): + return {"type": self.typeName(), + "keyType": self.keyType.jsonValue(), + "valueType": self.valueType.jsonValue(), + "valueContainsNull": self.valueContainsNull} + + @classmethod + def fromJson(cls, json): + return MapType(_parse_datatype_json_value(json["keyType"]), + _parse_datatype_json_value(json["valueType"]), + json["valueContainsNull"]) + + +class StructField(DataType): + + """Spark SQL StructField + + Represents a field in a StructType. + A StructField object comprises three fields, name (a string), + dataType (a DataType) and nullable (a bool). The field of name + is the name of a StructField. The field of dataType specifies + the data type of a StructField. + + The field of nullable specifies if values of a StructField can + contain None values. + + """ + + def __init__(self, name, dataType, nullable=True, metadata=None): + """Creates a StructField + :param name: the name of this field. + :param dataType: the data type of this field. + :param nullable: indicates whether values of this field + can be null. + :param metadata: metadata of this field, which is a map from string + to simple type that can be serialized to JSON + automatically + + >>> (StructField("f1", StringType, True) + ... == StructField("f1", StringType, True)) + True + >>> (StructField("f1", StringType, True) + ... == StructField("f2", StringType, True)) + False + """ + self.name = name + self.dataType = dataType + self.nullable = nullable + self.metadata = metadata or {} + + def __repr__(self): + return "StructField(%s,%s,%s)" % (self.name, self.dataType, + str(self.nullable).lower()) + + def jsonValue(self): + return {"name": self.name, + "type": self.dataType.jsonValue(), + "nullable": self.nullable, + "metadata": self.metadata} + + @classmethod + def fromJson(cls, json): + return StructField(json["name"], + _parse_datatype_json_value(json["type"]), + json["nullable"], + json["metadata"]) + + +class StructType(DataType): + + """Spark SQL StructType + + The data type representing rows. + A StructType object comprises a list of L{StructField}. + + """ + + def __init__(self, fields): + """Creates a StructType + + >>> struct1 = StructType([StructField("f1", StringType, True)]) + >>> struct2 = StructType([StructField("f1", StringType, True)]) + >>> struct1 == struct2 + True + >>> struct1 = StructType([StructField("f1", StringType, True)]) + >>> struct2 = StructType([StructField("f1", StringType, True), + ... [StructField("f2", IntegerType, False)]]) + >>> struct1 == struct2 + False + """ + self.fields = fields + + def __repr__(self): + return ("StructType(List(%s))" % + ",".join(str(field) for field in self.fields)) + + def jsonValue(self): + return {"type": self.typeName(), + "fields": [f.jsonValue() for f in self.fields]} + + @classmethod + def fromJson(cls, json): + return StructType([StructField.fromJson(f) for f in json["fields"]]) + + +class UserDefinedType(DataType): + """ + .. note:: WARN: Spark Internal Use Only + SQL User-Defined Type (UDT). + """ + + @classmethod + def typeName(cls): + return cls.__name__.lower() + + @classmethod + def sqlType(cls): + """ + Underlying SQL storage type for this UDT. + """ + raise NotImplementedError("UDT must implement sqlType().") + + @classmethod + def module(cls): + """ + The Python module of the UDT. + """ + raise NotImplementedError("UDT must implement module().") + + @classmethod + def scalaUDT(cls): + """ + The class name of the paired Scala UDT. + """ + raise NotImplementedError("UDT must have a paired Scala UDT.") + + def serialize(self, obj): + """ + Converts the a user-type object into a SQL datum. + """ + raise NotImplementedError("UDT must implement serialize().") + + def deserialize(self, datum): + """ + Converts a SQL datum into a user-type object. + """ + raise NotImplementedError("UDT must implement deserialize().") + + def json(self): + return json.dumps(self.jsonValue(), separators=(',', ':'), sort_keys=True) + + def jsonValue(self): + schema = { + "type": "udt", + "class": self.scalaUDT(), + "pyClass": "%s.%s" % (self.module(), type(self).__name__), + "sqlType": self.sqlType().jsonValue() + } + return schema + + @classmethod + def fromJson(cls, json): + pyUDT = json["pyClass"] + split = pyUDT.rfind(".") + pyModule = pyUDT[:split] + pyClass = pyUDT[split+1:] + m = __import__(pyModule, globals(), locals(), [pyClass], -1) + UDT = getattr(m, pyClass) + return UDT() + + def __eq__(self, other): + return type(self) == type(other) + + +_all_primitive_types = dict((v.typeName(), v) + for v in globals().itervalues() + if type(v) is PrimitiveTypeSingleton and + v.__base__ == PrimitiveType) + + +_all_complex_types = dict((v.typeName(), v) + for v in [ArrayType, MapType, StructType]) + + +def _parse_datatype_json_string(json_string): + """Parses the given data type JSON string. + >>> def check_datatype(datatype): + ... scala_datatype = sqlCtx._ssql_ctx.parseDataType(datatype.json()) + ... python_datatype = _parse_datatype_json_string(scala_datatype.json()) + ... return datatype == python_datatype + >>> all(check_datatype(cls()) for cls in _all_primitive_types.values()) + True + >>> # Simple ArrayType. + >>> simple_arraytype = ArrayType(StringType(), True) + >>> check_datatype(simple_arraytype) + True + >>> # Simple MapType. + >>> simple_maptype = MapType(StringType(), LongType()) + >>> check_datatype(simple_maptype) + True + >>> # Simple StructType. + >>> simple_structtype = StructType([ + ... StructField("a", DecimalType(), False), + ... StructField("b", BooleanType(), True), + ... StructField("c", LongType(), True), + ... StructField("d", BinaryType(), False)]) + >>> check_datatype(simple_structtype) + True + >>> # Complex StructType. + >>> complex_structtype = StructType([ + ... StructField("simpleArray", simple_arraytype, True), + ... StructField("simpleMap", simple_maptype, True), + ... StructField("simpleStruct", simple_structtype, True), + ... StructField("boolean", BooleanType(), False), + ... StructField("withMeta", DoubleType(), False, {"name": "age"})]) + >>> check_datatype(complex_structtype) + True + >>> # Complex ArrayType. + >>> complex_arraytype = ArrayType(complex_structtype, True) + >>> check_datatype(complex_arraytype) + True + >>> # Complex MapType. + >>> complex_maptype = MapType(complex_structtype, + ... complex_arraytype, False) + >>> check_datatype(complex_maptype) + True + >>> check_datatype(ExamplePointUDT()) + True + >>> structtype_with_udt = StructType([StructField("label", DoubleType(), False), + ... StructField("point", ExamplePointUDT(), False)]) + >>> check_datatype(structtype_with_udt) + True + """ + return _parse_datatype_json_value(json.loads(json_string)) + + +_FIXED_DECIMAL = re.compile("decimal\\((\\d+),(\\d+)\\)") + + +def _parse_datatype_json_value(json_value): + if type(json_value) is unicode: + if json_value in _all_primitive_types.keys(): + return _all_primitive_types[json_value]() + elif json_value == u'decimal': + return DecimalType() + elif _FIXED_DECIMAL.match(json_value): + m = _FIXED_DECIMAL.match(json_value) + return DecimalType(int(m.group(1)), int(m.group(2))) + else: + raise ValueError("Could not parse datatype: %s" % json_value) + else: + tpe = json_value["type"] + if tpe in _all_complex_types: + return _all_complex_types[tpe].fromJson(json_value) + elif tpe == 'udt': + return UserDefinedType.fromJson(json_value) + else: + raise ValueError("not supported type: %s" % tpe) + + +# Mapping Python types to Spark SQL DataType +_type_mappings = { + type(None): NullType, + bool: BooleanType, + int: IntegerType, + long: LongType, + float: DoubleType, + str: StringType, + unicode: StringType, + bytearray: BinaryType, + decimal.Decimal: DecimalType, + datetime.date: DateType, + datetime.datetime: TimestampType, + datetime.time: TimestampType, +} + + +def _infer_type(obj): + """Infer the DataType from obj + + >>> p = ExamplePoint(1.0, 2.0) + >>> _infer_type(p) + ExamplePointUDT + """ + if obj is None: + raise ValueError("Can not infer type for None") + + if hasattr(obj, '__UDT__'): + return obj.__UDT__ + + dataType = _type_mappings.get(type(obj)) + if dataType is not None: + return dataType() + + if isinstance(obj, dict): + for key, value in obj.iteritems(): + if key is not None and value is not None: + return MapType(_infer_type(key), _infer_type(value), True) + else: + return MapType(NullType(), NullType(), True) + elif isinstance(obj, (list, array)): + for v in obj: + if v is not None: + return ArrayType(_infer_type(obj[0]), True) + else: + return ArrayType(NullType(), True) + else: + try: + return _infer_schema(obj) + except ValueError: + raise ValueError("not supported type: %s" % type(obj)) + + +def _infer_schema(row): + """Infer the schema from dict/namedtuple/object""" + if isinstance(row, dict): + items = sorted(row.items()) + + elif isinstance(row, tuple): + if hasattr(row, "_fields"): # namedtuple + items = zip(row._fields, tuple(row)) + elif hasattr(row, "__FIELDS__"): # Row + items = zip(row.__FIELDS__, tuple(row)) + elif all(isinstance(x, tuple) and len(x) == 2 for x in row): + items = row + else: + raise ValueError("Can't infer schema from tuple") + + elif hasattr(row, "__dict__"): # object + items = sorted(row.__dict__.items()) + + else: + raise ValueError("Can not infer schema for type: %s" % type(row)) + + fields = [StructField(k, _infer_type(v), True) for k, v in items] + return StructType(fields) + + +def _need_python_to_sql_conversion(dataType): + """ + Checks whether we need python to sql conversion for the given type. + For now, only UDTs need this conversion. + + >>> _need_python_to_sql_conversion(DoubleType()) + False + >>> schema0 = StructType([StructField("indices", ArrayType(IntegerType(), False), False), + ... StructField("values", ArrayType(DoubleType(), False), False)]) + >>> _need_python_to_sql_conversion(schema0) + False + >>> _need_python_to_sql_conversion(ExamplePointUDT()) + True + >>> schema1 = ArrayType(ExamplePointUDT(), False) + >>> _need_python_to_sql_conversion(schema1) + True + >>> schema2 = StructType([StructField("label", DoubleType(), False), + ... StructField("point", ExamplePointUDT(), False)]) + >>> _need_python_to_sql_conversion(schema2) + True + """ + if isinstance(dataType, StructType): + return any([_need_python_to_sql_conversion(f.dataType) for f in dataType.fields]) + elif isinstance(dataType, ArrayType): + return _need_python_to_sql_conversion(dataType.elementType) + elif isinstance(dataType, MapType): + return _need_python_to_sql_conversion(dataType.keyType) or \ + _need_python_to_sql_conversion(dataType.valueType) + elif isinstance(dataType, UserDefinedType): + return True + else: + return False + + +def _python_to_sql_converter(dataType): + """ + Returns a converter that converts a Python object into a SQL datum for the given type. + + >>> conv = _python_to_sql_converter(DoubleType()) + >>> conv(1.0) + 1.0 + >>> conv = _python_to_sql_converter(ArrayType(DoubleType(), False)) + >>> conv([1.0, 2.0]) + [1.0, 2.0] + >>> conv = _python_to_sql_converter(ExamplePointUDT()) + >>> conv(ExamplePoint(1.0, 2.0)) + [1.0, 2.0] + >>> schema = StructType([StructField("label", DoubleType(), False), + ... StructField("point", ExamplePointUDT(), False)]) + >>> conv = _python_to_sql_converter(schema) + >>> conv((1.0, ExamplePoint(1.0, 2.0))) + (1.0, [1.0, 2.0]) + """ + if not _need_python_to_sql_conversion(dataType): + return lambda x: x + + if isinstance(dataType, StructType): + names, types = zip(*[(f.name, f.dataType) for f in dataType.fields]) + converters = map(_python_to_sql_converter, types) + + def converter(obj): + if isinstance(obj, dict): + return tuple(c(obj.get(n)) for n, c in zip(names, converters)) + elif isinstance(obj, tuple): + if hasattr(obj, "_fields") or hasattr(obj, "__FIELDS__"): + return tuple(c(v) for c, v in zip(converters, obj)) + elif all(isinstance(x, tuple) and len(x) == 2 for x in obj): # k-v pairs + d = dict(obj) + return tuple(c(d.get(n)) for n, c in zip(names, converters)) + else: + return tuple(c(v) for c, v in zip(converters, obj)) + else: + raise ValueError("Unexpected tuple %r with type %r" % (obj, dataType)) + return converter + elif isinstance(dataType, ArrayType): + element_converter = _python_to_sql_converter(dataType.elementType) + return lambda a: [element_converter(v) for v in a] + elif isinstance(dataType, MapType): + key_converter = _python_to_sql_converter(dataType.keyType) + value_converter = _python_to_sql_converter(dataType.valueType) + return lambda m: dict([(key_converter(k), value_converter(v)) for k, v in m.items()]) + elif isinstance(dataType, UserDefinedType): + return lambda obj: dataType.serialize(obj) + else: + raise ValueError("Unexpected type %r" % dataType) + + +def _has_nulltype(dt): + """ Return whether there is NullType in `dt` or not """ + if isinstance(dt, StructType): + return any(_has_nulltype(f.dataType) for f in dt.fields) + elif isinstance(dt, ArrayType): + return _has_nulltype((dt.elementType)) + elif isinstance(dt, MapType): + return _has_nulltype(dt.keyType) or _has_nulltype(dt.valueType) + else: + return isinstance(dt, NullType) + + +def _merge_type(a, b): + if isinstance(a, NullType): + return b + elif isinstance(b, NullType): + return a + elif type(a) is not type(b): + # TODO: type cast (such as int -> long) + raise TypeError("Can not merge type %s and %s" % (a, b)) + + # same type + if isinstance(a, StructType): + nfs = dict((f.name, f.dataType) for f in b.fields) + fields = [StructField(f.name, _merge_type(f.dataType, nfs.get(f.name, NullType()))) + for f in a.fields] + names = set([f.name for f in fields]) + for n in nfs: + if n not in names: + fields.append(StructField(n, nfs[n])) + return StructType(fields) + + elif isinstance(a, ArrayType): + return ArrayType(_merge_type(a.elementType, b.elementType), True) + + elif isinstance(a, MapType): + return MapType(_merge_type(a.keyType, b.keyType), + _merge_type(a.valueType, b.valueType), + True) + else: + return a + + +def _create_converter(dataType): + """Create an converter to drop the names of fields in obj """ + if isinstance(dataType, ArrayType): + conv = _create_converter(dataType.elementType) + return lambda row: map(conv, row) + + elif isinstance(dataType, MapType): + kconv = _create_converter(dataType.keyType) + vconv = _create_converter(dataType.valueType) + return lambda row: dict((kconv(k), vconv(v)) for k, v in row.iteritems()) + + elif isinstance(dataType, NullType): + return lambda x: None + + elif not isinstance(dataType, StructType): + return lambda x: x + + # dataType must be StructType + names = [f.name for f in dataType.fields] + converters = [_create_converter(f.dataType) for f in dataType.fields] + + def convert_struct(obj): + if obj is None: + return + + if isinstance(obj, tuple): + if hasattr(obj, "_fields"): + d = dict(zip(obj._fields, obj)) + elif hasattr(obj, "__FIELDS__"): + d = dict(zip(obj.__FIELDS__, obj)) + elif all(isinstance(x, tuple) and len(x) == 2 for x in obj): + d = dict(obj) + else: + raise ValueError("unexpected tuple: %s" % str(obj)) + + elif isinstance(obj, dict): + d = obj + elif hasattr(obj, "__dict__"): # object + d = obj.__dict__ + else: + raise ValueError("Unexpected obj: %s" % obj) + + return tuple([conv(d.get(name)) for name, conv in zip(names, converters)]) + + return convert_struct + + +_BRACKETS = {'(': ')', '[': ']', '{': '}'} + + +def _split_schema_abstract(s): + """ + split the schema abstract into fields + + >>> _split_schema_abstract("a b c") + ['a', 'b', 'c'] + >>> _split_schema_abstract("a(a b)") + ['a(a b)'] + >>> _split_schema_abstract("a b[] c{a b}") + ['a', 'b[]', 'c{a b}'] + >>> _split_schema_abstract(" ") + [] + """ + + r = [] + w = '' + brackets = [] + for c in s: + if c == ' ' and not brackets: + if w: + r.append(w) + w = '' + else: + w += c + if c in _BRACKETS: + brackets.append(c) + elif c in _BRACKETS.values(): + if not brackets or c != _BRACKETS[brackets.pop()]: + raise ValueError("unexpected " + c) + + if brackets: + raise ValueError("brackets not closed: %s" % brackets) + if w: + r.append(w) + return r + + +def _parse_field_abstract(s): + """ + Parse a field in schema abstract + + >>> _parse_field_abstract("a") + StructField(a,None,true) + >>> _parse_field_abstract("b(c d)") + StructField(b,StructType(...c,None,true),StructField(d... + >>> _parse_field_abstract("a[]") + StructField(a,ArrayType(None,true),true) + >>> _parse_field_abstract("a{[]}") + StructField(a,MapType(None,ArrayType(None,true),true),true) + """ + if set(_BRACKETS.keys()) & set(s): + idx = min((s.index(c) for c in _BRACKETS if c in s)) + name = s[:idx] + return StructField(name, _parse_schema_abstract(s[idx:]), True) + else: + return StructField(s, None, True) + + +def _parse_schema_abstract(s): + """ + parse abstract into schema + + >>> _parse_schema_abstract("a b c") + StructType...a...b...c... + >>> _parse_schema_abstract("a[b c] b{}") + StructType...a,ArrayType...b...c...b,MapType... + >>> _parse_schema_abstract("c{} d{a b}") + StructType...c,MapType...d,MapType...a...b... + >>> _parse_schema_abstract("a b(t)").fields[1] + StructField(b,StructType(List(StructField(t,None,true))),true) + """ + s = s.strip() + if not s: + return + + elif s.startswith('('): + return _parse_schema_abstract(s[1:-1]) + + elif s.startswith('['): + return ArrayType(_parse_schema_abstract(s[1:-1]), True) + + elif s.startswith('{'): + return MapType(None, _parse_schema_abstract(s[1:-1])) + + parts = _split_schema_abstract(s) + fields = [_parse_field_abstract(p) for p in parts] + return StructType(fields) + + +def _infer_schema_type(obj, dataType): + """ + Fill the dataType with types inferred from obj + + >>> schema = _parse_schema_abstract("a b c d") + >>> row = (1, 1.0, "str", datetime.date(2014, 10, 10)) + >>> _infer_schema_type(row, schema) + StructType...IntegerType...DoubleType...StringType...DateType... + >>> row = [[1], {"key": (1, 2.0)}] + >>> schema = _parse_schema_abstract("a[] b{c d}") + >>> _infer_schema_type(row, schema) + StructType...a,ArrayType...b,MapType(StringType,...c,IntegerType... + """ + if dataType is None: + return _infer_type(obj) + + if not obj: + return NullType() + + if isinstance(dataType, ArrayType): + eType = _infer_schema_type(obj[0], dataType.elementType) + return ArrayType(eType, True) + + elif isinstance(dataType, MapType): + k, v = obj.iteritems().next() + return MapType(_infer_schema_type(k, dataType.keyType), + _infer_schema_type(v, dataType.valueType)) + + elif isinstance(dataType, StructType): + fs = dataType.fields + assert len(fs) == len(obj), \ + "Obj(%s) have different length with fields(%s)" % (obj, fs) + fields = [StructField(f.name, _infer_schema_type(o, f.dataType), True) + for o, f in zip(obj, fs)] + return StructType(fields) + + else: + raise ValueError("Unexpected dataType: %s" % dataType) + + +_acceptable_types = { + BooleanType: (bool,), + ByteType: (int, long), + ShortType: (int, long), + IntegerType: (int, long), + LongType: (int, long), + FloatType: (float,), + DoubleType: (float,), + DecimalType: (decimal.Decimal,), + StringType: (str, unicode), + BinaryType: (bytearray,), + DateType: (datetime.date,), + TimestampType: (datetime.datetime,), + ArrayType: (list, tuple, array), + MapType: (dict,), + StructType: (tuple, list), +} + + +def _verify_type(obj, dataType): + """ + Verify the type of obj against dataType, raise an exception if + they do not match. + + >>> _verify_type(None, StructType([])) + >>> _verify_type("", StringType()) + >>> _verify_type(0, IntegerType()) + >>> _verify_type(range(3), ArrayType(ShortType())) + >>> _verify_type(set(), ArrayType(StringType())) # doctest: +IGNORE_EXCEPTION_DETAIL + Traceback (most recent call last): + ... + TypeError:... + >>> _verify_type({}, MapType(StringType(), IntegerType())) + >>> _verify_type((), StructType([])) + >>> _verify_type([], StructType([])) + >>> _verify_type([1], StructType([])) # doctest: +IGNORE_EXCEPTION_DETAIL + Traceback (most recent call last): + ... + ValueError:... + >>> _verify_type(ExamplePoint(1.0, 2.0), ExamplePointUDT()) + >>> _verify_type([1.0, 2.0], ExamplePointUDT()) # doctest: +IGNORE_EXCEPTION_DETAIL + Traceback (most recent call last): + ... + ValueError:... + """ + # all objects are nullable + if obj is None: + return + + if isinstance(dataType, UserDefinedType): + if not (hasattr(obj, '__UDT__') and obj.__UDT__ == dataType): + raise ValueError("%r is not an instance of type %r" % (obj, dataType)) + _verify_type(dataType.serialize(obj), dataType.sqlType()) + return + + _type = type(dataType) + assert _type in _acceptable_types, "unkown datatype: %s" % dataType + + # subclass of them can not be deserialized in JVM + if type(obj) not in _acceptable_types[_type]: + raise TypeError("%s can not accept object in type %s" + % (dataType, type(obj))) + + if isinstance(dataType, ArrayType): + for i in obj: + _verify_type(i, dataType.elementType) + + elif isinstance(dataType, MapType): + for k, v in obj.iteritems(): + _verify_type(k, dataType.keyType) + _verify_type(v, dataType.valueType) + + elif isinstance(dataType, StructType): + if len(obj) != len(dataType.fields): + raise ValueError("Length of object (%d) does not match with" + "length of fields (%d)" % (len(obj), len(dataType.fields))) + for v, f in zip(obj, dataType.fields): + _verify_type(v, f.dataType) + + +_cached_cls = {} + + +def _restore_object(dataType, obj): + """ Restore object during unpickling. """ + # use id(dataType) as key to speed up lookup in dict + # Because of batched pickling, dataType will be the + # same object in most cases. + k = id(dataType) + cls = _cached_cls.get(k) + if cls is None: + # use dataType as key to avoid create multiple class + cls = _cached_cls.get(dataType) + if cls is None: + cls = _create_cls(dataType) + _cached_cls[dataType] = cls + _cached_cls[k] = cls + return cls(obj) + + +def _create_object(cls, v): + """ Create an customized object with class `cls`. """ + # datetime.date would be deserialized as datetime.datetime + # from java type, so we need to set it back. + if cls is datetime.date and isinstance(v, datetime.datetime): + return v.date() + return cls(v) if v is not None else v + + +def _create_getter(dt, i): + """ Create a getter for item `i` with schema """ + cls = _create_cls(dt) + + def getter(self): + return _create_object(cls, self[i]) + + return getter + + +def _has_struct_or_date(dt): + """Return whether `dt` is or has StructType/DateType in it""" + if isinstance(dt, StructType): + return True + elif isinstance(dt, ArrayType): + return _has_struct_or_date(dt.elementType) + elif isinstance(dt, MapType): + return _has_struct_or_date(dt.keyType) or _has_struct_or_date(dt.valueType) + elif isinstance(dt, DateType): + return True + elif isinstance(dt, UserDefinedType): + return True + return False + + +def _create_properties(fields): + """Create properties according to fields""" + ps = {} + for i, f in enumerate(fields): + name = f.name + if (name.startswith("__") and name.endswith("__") + or keyword.iskeyword(name)): + warnings.warn("field name %s can not be accessed in Python," + "use position to access it instead" % name) + if _has_struct_or_date(f.dataType): + # delay creating object until accessing it + getter = _create_getter(f.dataType, i) + else: + getter = itemgetter(i) + ps[name] = property(getter) + return ps + + +def _create_cls(dataType): + """ + Create an class by dataType + + The created class is similar to namedtuple, but can have nested schema. + + >>> schema = _parse_schema_abstract("a b c") + >>> row = (1, 1.0, "str") + >>> schema = _infer_schema_type(row, schema) + >>> obj = _create_cls(schema)(row) + >>> import pickle + >>> pickle.loads(pickle.dumps(obj)) + Row(a=1, b=1.0, c='str') + + >>> row = [[1], {"key": (1, 2.0)}] + >>> schema = _parse_schema_abstract("a[] b{c d}") + >>> schema = _infer_schema_type(row, schema) + >>> obj = _create_cls(schema)(row) + >>> pickle.loads(pickle.dumps(obj)) + Row(a=[1], b={'key': Row(c=1, d=2.0)}) + >>> pickle.loads(pickle.dumps(obj.a)) + [1] + >>> pickle.loads(pickle.dumps(obj.b)) + {'key': Row(c=1, d=2.0)} + """ + + if isinstance(dataType, ArrayType): + cls = _create_cls(dataType.elementType) + + def List(l): + if l is None: + return + return [_create_object(cls, v) for v in l] + + return List + + elif isinstance(dataType, MapType): + kcls = _create_cls(dataType.keyType) + vcls = _create_cls(dataType.valueType) + + def Dict(d): + if d is None: + return + return dict((_create_object(kcls, k), _create_object(vcls, v)) for k, v in d.items()) + + return Dict + + elif isinstance(dataType, DateType): + return datetime.date + + elif isinstance(dataType, UserDefinedType): + return lambda datum: dataType.deserialize(datum) + + elif not isinstance(dataType, StructType): + # no wrapper for primitive types + return lambda x: x + + class Row(tuple): + + """ Row in DataFrame """ + __DATATYPE__ = dataType + __FIELDS__ = tuple(f.name for f in dataType.fields) + __slots__ = () + + # create property for fast access + locals().update(_create_properties(dataType.fields)) + + def asDict(self): + """ Return as a dict """ + return dict((n, getattr(self, n)) for n in self.__FIELDS__) + + def __repr__(self): + # call collect __repr__ for nested objects + return ("Row(%s)" % ", ".join("%s=%r" % (n, getattr(self, n)) + for n in self.__FIELDS__)) + + def __reduce__(self): + return (_restore_object, (self.__DATATYPE__, tuple(self))) + + return Row + + +def _create_row(fields, values): + row = Row(*values) + row.__FIELDS__ = fields + return row + + +class Row(tuple): + + """ + A row in L{DataFrame}. The fields in it can be accessed like attributes. + + Row can be used to create a row object by using named arguments, + the fields will be sorted by names. + + >>> row = Row(name="Alice", age=11) + >>> row + Row(age=11, name='Alice') + >>> row.name, row.age + ('Alice', 11) + + Row also can be used to create another Row like class, then it + could be used to create Row objects, such as + + >>> Person = Row("name", "age") + >>> Person + + >>> Person("Alice", 11) + Row(name='Alice', age=11) + """ + + def __new__(self, *args, **kwargs): + if args and kwargs: + raise ValueError("Can not use both args " + "and kwargs to create Row") + if args: + # create row class or objects + return tuple.__new__(self, args) + + elif kwargs: + # create row objects + names = sorted(kwargs.keys()) + values = tuple(kwargs[n] for n in names) + row = tuple.__new__(self, values) + row.__FIELDS__ = names + return row + + else: + raise ValueError("No args or kwargs") + + def asDict(self): + """ + Return as an dict + """ + if not hasattr(self, "__FIELDS__"): + raise TypeError("Cannot convert a Row class into dict") + return dict(zip(self.__FIELDS__, self)) + + # let obect acs like class + def __call__(self, *args): + """create new Row object""" + return _create_row(self, args) + + def __getattr__(self, item): + if item.startswith("__"): + raise AttributeError(item) + try: + # it will be slow when it has many fields, + # but this will not be used in normal cases + idx = self.__FIELDS__.index(item) + return self[idx] + except IndexError: + raise AttributeError(item) + + def __reduce__(self): + if hasattr(self, "__FIELDS__"): + return (_create_row, (self.__FIELDS__, tuple(self))) + else: + return tuple.__reduce__(self) + + def __repr__(self): + if hasattr(self, "__FIELDS__"): + return "Row(%s)" % ", ".join("%s=%r" % (k, v) + for k, v in zip(self.__FIELDS__, self)) + else: + return "" % ", ".join(self) + + +def _test(): + import doctest + from pyspark.context import SparkContext + # let doctest run in pyspark.sql.types, so DataTypes can be picklable + import pyspark.sql.types + from pyspark.sql import Row, SQLContext + from pyspark.sql.tests import ExamplePoint, ExamplePointUDT + globs = pyspark.sql.types.__dict__.copy() + sc = SparkContext('local[4]', 'PythonTest') + globs['sc'] = sc + globs['sqlCtx'] = sqlCtx = SQLContext(sc) + globs['ExamplePoint'] = ExamplePoint + globs['ExamplePointUDT'] = ExamplePointUDT + (failure_count, test_count) = doctest.testmod( + pyspark.sql.types, globs=globs, optionflags=doctest.ELLIPSIS) + globs['sc'].stop() + if failure_count: + exit(-1) + + +if __name__ == "__main__": + _test() diff --git a/python/run-tests b/python/run-tests index 649a2c44d187b..58a26dd8ff088 100755 --- a/python/run-tests +++ b/python/run-tests @@ -64,8 +64,10 @@ function run_core_tests() { function run_sql_tests() { echo "Run sql tests ..." - run_test "pyspark/sql.py" - run_test "pyspark/sql_tests.py" + run_test "pyspark/sql/types.py" + run_test "pyspark/sql/context.py" + run_test "pyspark/sql/dataframe.py" + run_test "pyspark/sql/tests.py" } function run_mllib_tests() { diff --git a/repl/pom.xml b/repl/pom.xml index bd39b90fd8714..3d4adf8fd5b03 100644 --- a/repl/pom.xml +++ b/repl/pom.xml @@ -66,7 +66,6 @@ org.apache.spark spark-sql_${scala.binary.version} ${project.version} - test org.scala-lang diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala index 7d1f319e721fe..3f0d77ad6322a 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala @@ -23,8 +23,7 @@ import org.apache.spark.sql.catalyst.errors.TreeNodeException import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.plans.logical._ import org.apache.spark.sql.catalyst.rules._ -import org.apache.spark.sql.types.StructType -import org.apache.spark.sql.types.IntegerType +import org.apache.spark.sql.types.{ArrayType, StructField, StructType, IntegerType} /** * A trivial [[Analyzer]] with an [[EmptyCatalog]] and [[EmptyFunctionRegistry]]. Used for testing @@ -314,21 +313,28 @@ class Analyzer(catalog: Catalog, * desired fields are found. */ protected def resolveGetField(expr: Expression, fieldName: String): Expression = { + def findField(fields: Array[StructField]): Int = { + val checkField = (f: StructField) => resolver(f.name, fieldName) + val ordinal = fields.indexWhere(checkField) + if (ordinal == -1) { + throw new AnalysisException( + s"No such struct field $fieldName in ${fields.map(_.name).mkString(", ")}") + } else if (fields.indexWhere(checkField, ordinal + 1) != -1) { + throw new AnalysisException( + s"Ambiguous reference to fields ${fields.filter(checkField).mkString(", ")}") + } else { + ordinal + } + } expr.dataType match { case StructType(fields) => - val actualField = fields.filter(f => resolver(f.name, fieldName)) - if (actualField.length == 0) { - throw new AnalysisException( - s"No such struct field $fieldName in ${fields.map(_.name).mkString(", ")}") - } else if (actualField.length == 1) { - val field = actualField(0) - GetField(expr, field, fields.indexOf(field)) - } else { - throw new AnalysisException( - s"Ambiguous reference to fields ${actualField.mkString(", ")}") - } + val ordinal = findField(fields) + StructGetField(expr, fields(ordinal), ordinal) + case ArrayType(StructType(fields), containsNull) => + val ordinal = findField(fields) + ArrayGetField(expr, fields(ordinal), ordinal, containsNull) case otherType => - throw new AnalysisException(s"GetField is not valid on columns of type $otherType") + throw new AnalysisException(s"GetField is not valid on fields of type $otherType") } } } diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala index cf14992ef835c..c32a4b886eb82 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala @@ -17,6 +17,7 @@ package org.apache.spark.sql.catalyst.expressions +import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute import org.apache.spark.sql.catalyst.errors.TreeNodeException import org.apache.spark.sql.catalyst.trees import org.apache.spark.sql.catalyst.trees.TreeNode @@ -66,6 +67,17 @@ abstract class Expression extends TreeNode[Expression] { */ def childrenResolved = !children.exists(!_.resolved) + /** + * Returns a string representation of this expression that does not have developer centric + * debugging information like the expression id. + */ + def prettyString: String = { + transform { + case a: AttributeReference => PrettyAttribute(a.name) + case u: UnresolvedAttribute => PrettyAttribute(u.name) + }.toString + } + /** * A set of helper functions that return the correct descendant of `scala.math.Numeric[T]` type * and do any casting necessary of child evaluation. diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala index 66e2e5c4bafce..68051a2a2007e 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypes.scala @@ -70,22 +70,48 @@ case class GetItem(child: Expression, ordinal: Expression) extends Expression { } } + +trait GetField extends UnaryExpression { + self: Product => + + type EvaluatedType = Any + override def foldable = child.foldable + override def toString = s"$child.${field.name}" + + def field: StructField +} + /** * Returns the value of fields in the Struct `child`. */ -case class GetField(child: Expression, field: StructField, ordinal: Int) extends UnaryExpression { - type EvaluatedType = Any +case class StructGetField(child: Expression, field: StructField, ordinal: Int) extends GetField { def dataType = field.dataType override def nullable = child.nullable || field.nullable - override def foldable = child.foldable override def eval(input: Row): Any = { val baseValue = child.eval(input).asInstanceOf[Row] if (baseValue == null) null else baseValue(ordinal) } +} - override def toString = s"$child.${field.name}" +/** + * Returns the array of value of fields in the Array of Struct `child`. + */ +case class ArrayGetField(child: Expression, field: StructField, ordinal: Int, containsNull: Boolean) + extends GetField { + + def dataType = ArrayType(field.dataType, containsNull) + override def nullable = child.nullable + + override def eval(input: Row): Any = { + val baseValue = child.eval(input).asInstanceOf[Seq[Row]] + if (baseValue == null) null else { + baseValue.map { row => + if (row == null) null else row(ordinal) + } + } + } } /** diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/namedExpressions.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/namedExpressions.scala index e6ab1fd8d7939..7f122e9d55734 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/namedExpressions.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/namedExpressions.scala @@ -190,6 +190,26 @@ case class AttributeReference( override def toString: String = s"$name#${exprId.id}$typeSuffix" } +/** + * A place holder used when printing expressions without debugging information such as the + * expression id or the unresolved indicator. + */ +case class PrettyAttribute(name: String) extends Attribute with trees.LeafNode[Expression] { + type EvaluatedType = Any + + override def toString = name + + override def withNullability(newNullability: Boolean): Attribute = ??? + override def newInstance(): Attribute = ??? + override def withQualifiers(newQualifiers: Seq[String]): Attribute = ??? + override def withName(newName: String): Attribute = ??? + override def qualifiers: Seq[String] = ??? + override def exprId: ExprId = ??? + override def eval(input: Row): EvaluatedType = ??? + override def nullable: Boolean = ??? + override def dataType: DataType = ??? +} + object VirtualColumn { val groupingIdName = "grouping__id" def newGroupingId = AttributeReference(groupingIdName, IntegerType, false)() diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala index 8c8f2896eb99b..0da081ed1a6e2 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala @@ -50,7 +50,10 @@ object DefaultOptimizer extends Optimizer { CombineFilters, PushPredicateThroughProject, PushPredicateThroughJoin, - ColumnPruning) :: Nil + PushPredicateThroughGenerate, + ColumnPruning) :: + Batch("LocalRelation", FixedPoint(100), + ConvertToLocalRelation) :: Nil } /** @@ -206,7 +209,8 @@ object NullPropagation extends Rule[LogicalPlan] { case e @ IsNotNull(c) if !c.nullable => Literal(true, BooleanType) case e @ GetItem(Literal(null, _), _) => Literal(null, e.dataType) case e @ GetItem(_, Literal(null, _)) => Literal(null, e.dataType) - case e @ GetField(Literal(null, _), _, _) => Literal(null, e.dataType) + case e @ StructGetField(Literal(null, _), _, _) => Literal(null, e.dataType) + case e @ ArrayGetField(Literal(null, _), _, _, _) => Literal(null, e.dataType) case e @ EqualNullSafe(Literal(null, _), r) => IsNull(r) case e @ EqualNullSafe(l, Literal(null, _)) => IsNull(l) case e @ Count(expr) if !expr.nullable => Count(Literal(1)) @@ -453,6 +457,30 @@ object PushPredicateThroughProject extends Rule[LogicalPlan] { } } +/** + * Push [[Filter]] operators through [[Generate]] operators. Parts of the predicate that reference + * attributes generated in [[Generate]] will remain above, and the rest should be pushed beneath. + */ +object PushPredicateThroughGenerate extends Rule[LogicalPlan] with PredicateHelper { + + def apply(plan: LogicalPlan): LogicalPlan = plan transform { + case filter @ Filter(condition, + generate @ Generate(generator, join, outer, alias, grandChild)) => + // Predicates that reference attributes produced by the `Generate` operator cannot + // be pushed below the operator. + val (pushDown, stayUp) = splitConjunctivePredicates(condition).partition { + conjunct => conjunct.references subsetOf grandChild.outputSet + } + if (pushDown.nonEmpty) { + val pushDownPredicate = pushDown.reduce(And) + val withPushdown = generate.copy(child = Filter(pushDownPredicate, grandChild)) + stayUp.reduceOption(And).map(Filter(_, withPushdown)).getOrElse(withPushdown) + } else { + filter + } + } +} + /** * Pushes down [[Filter]] operators where the `condition` can be * evaluated using only the attributes of the left or right side of a join. Other @@ -610,3 +638,17 @@ object DecimalAggregates extends Rule[LogicalPlan] { DecimalType(prec + 4, scale + 4)) } } + +/** + * Converts local operations (i.e. ones that don't require data exchange) on LocalRelation to + * another LocalRelation. + * + * This is relatively simple as it currently handles only a single case: Project. + */ +object ConvertToLocalRelation extends Rule[LogicalPlan] { + def apply(plan: LogicalPlan): LogicalPlan = plan transform { + case Project(projectList, LocalRelation(output, data)) => + val projection = new InterpretedProjection(projectList, output) + LocalRelation(projectList.map(_.toAttribute), data.map(projection)) + } +} diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala index 8d30528328946..7cf4b81274906 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala @@ -29,12 +29,15 @@ import org.apache.spark.sql.catalyst.trees /** * Estimates of various statistics. The default estimation logic simply lazily multiplies the * corresponding statistic produced by the children. To override this behavior, override - * `statistics` and assign it an overriden version of `Statistics`. + * `statistics` and assign it an overridden version of `Statistics`. * - * '''NOTE''': concrete and/or overriden versions of statistics fields should pay attention to the + * '''NOTE''': concrete and/or overridden versions of statistics fields should pay attention to the * performance of the implementations. The reason is that estimations might get triggered in * performance-critical processes, such as query plan planning. * + * Note that we are using a BigInt here since it is easy to overflow a 64-bit integer in + * cardinality estimation (e.g. cartesian joins). + * * @param sizeInBytes Physical size in bytes. For leaf operators this defaults to 1, otherwise it * defaults to the product of children's `sizeInBytes`. */ diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/types/dataTypes.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/types/dataTypes.scala index 91efe320546a7..2abb1caee9cd9 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/types/dataTypes.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/types/dataTypes.scala @@ -240,10 +240,16 @@ abstract class DataType { * @group dataType */ @DeveloperApi -case object NullType extends DataType { +class NullType private() extends DataType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "NullType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. override def defaultSize: Int = 1 } +case object NullType extends NullType + + protected[sql] object NativeType { val all = Seq( IntegerType, BooleanType, LongType, DoubleType, FloatType, ShortType, ByteType, StringType) @@ -292,7 +298,10 @@ protected[sql] abstract class NativeType extends DataType { * @group dataType */ @DeveloperApi -case object StringType extends NativeType with PrimitiveType { +class StringType private() extends NativeType with PrimitiveType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "StringType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = String @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val ordering = implicitly[Ordering[JvmType]] @@ -303,6 +312,8 @@ case object StringType extends NativeType with PrimitiveType { override def defaultSize: Int = 4096 } +case object StringType extends StringType + /** * :: DeveloperApi :: @@ -313,7 +324,10 @@ case object StringType extends NativeType with PrimitiveType { * @group dataType */ @DeveloperApi -case object BinaryType extends NativeType with PrimitiveType { +class BinaryType private() extends NativeType with PrimitiveType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "BinaryType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Array[Byte] @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val ordering = new Ordering[JvmType] { @@ -332,6 +346,8 @@ case object BinaryType extends NativeType with PrimitiveType { override def defaultSize: Int = 4096 } +case object BinaryType extends BinaryType + /** * :: DeveloperApi :: @@ -341,7 +357,10 @@ case object BinaryType extends NativeType with PrimitiveType { *@group dataType */ @DeveloperApi -case object BooleanType extends NativeType with PrimitiveType { +class BooleanType private() extends NativeType with PrimitiveType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "BooleanType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Boolean @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val ordering = implicitly[Ordering[JvmType]] @@ -352,6 +371,8 @@ case object BooleanType extends NativeType with PrimitiveType { override def defaultSize: Int = 1 } +case object BooleanType extends BooleanType + /** * :: DeveloperApi :: @@ -362,7 +383,10 @@ case object BooleanType extends NativeType with PrimitiveType { * @group dataType */ @DeveloperApi -case object TimestampType extends NativeType { +class TimestampType private() extends NativeType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "TimestampType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Timestamp @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } @@ -377,6 +401,8 @@ case object TimestampType extends NativeType { override def defaultSize: Int = 12 } +case object TimestampType extends TimestampType + /** * :: DeveloperApi :: @@ -387,7 +413,10 @@ case object TimestampType extends NativeType { * @group dataType */ @DeveloperApi -case object DateType extends NativeType { +class DateType private() extends NativeType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "DateType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Int @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } @@ -400,6 +429,8 @@ case object DateType extends NativeType { override def defaultSize: Int = 4 } +case object DateType extends DateType + abstract class NumericType extends NativeType with PrimitiveType { // Unfortunately we can't get this implicitly as that breaks Spark Serialization. In order for @@ -438,7 +469,10 @@ protected[sql] sealed abstract class IntegralType extends NumericType { * @group dataType */ @DeveloperApi -case object LongType extends IntegralType { +class LongType private() extends IntegralType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "LongType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Long @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val numeric = implicitly[Numeric[Long]] @@ -453,6 +487,8 @@ case object LongType extends IntegralType { override def simpleString = "bigint" } +case object LongType extends LongType + /** * :: DeveloperApi :: @@ -462,7 +498,10 @@ case object LongType extends IntegralType { * @group dataType */ @DeveloperApi -case object IntegerType extends IntegralType { +class IntegerType private() extends IntegralType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "IntegerType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Int @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val numeric = implicitly[Numeric[Int]] @@ -477,6 +516,8 @@ case object IntegerType extends IntegralType { override def simpleString = "int" } +case object IntegerType extends IntegerType + /** * :: DeveloperApi :: @@ -486,7 +527,10 @@ case object IntegerType extends IntegralType { * @group dataType */ @DeveloperApi -case object ShortType extends IntegralType { +class ShortType private() extends IntegralType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "ShortType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Short @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val numeric = implicitly[Numeric[Short]] @@ -501,6 +545,8 @@ case object ShortType extends IntegralType { override def simpleString = "smallint" } +case object ShortType extends ShortType + /** * :: DeveloperApi :: @@ -510,7 +556,10 @@ case object ShortType extends IntegralType { * @group dataType */ @DeveloperApi -case object ByteType extends IntegralType { +class ByteType private() extends IntegralType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "ByteType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Byte @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val numeric = implicitly[Numeric[Byte]] @@ -525,6 +574,8 @@ case object ByteType extends IntegralType { override def simpleString = "tinyint" } +case object ByteType extends ByteType + /** Matcher for any expressions that evaluate to [[FractionalType]]s */ protected[sql] object FractionalType { @@ -630,7 +681,10 @@ object DecimalType { * @group dataType */ @DeveloperApi -case object DoubleType extends FractionalType { +class DoubleType private() extends FractionalType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "DoubleType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Double @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val numeric = implicitly[Numeric[Double]] @@ -644,6 +698,8 @@ case object DoubleType extends FractionalType { override def defaultSize: Int = 8 } +case object DoubleType extends DoubleType + /** * :: DeveloperApi :: @@ -653,7 +709,10 @@ case object DoubleType extends FractionalType { * @group dataType */ @DeveloperApi -case object FloatType extends FractionalType { +class FloatType private() extends FractionalType { + // The companion object and this class is separated so the companion object also subclasses + // this type. Otherwise, the companion object would be of type "FloatType$" in byte code. + // Defined with a private constructor so the companion object is the only possible instantiation. private[sql] type JvmType = Float @transient private[sql] lazy val tag = ScalaReflectionLock.synchronized { typeTag[JvmType] } private[sql] val numeric = implicitly[Numeric[Float]] @@ -667,6 +726,8 @@ case object FloatType extends FractionalType { override def defaultSize: Int = 4 } +case object FloatType extends FloatType + object ArrayType { /** Construct a [[ArrayType]] object with the given element type. The `containsNull` is true. */ diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisSuite.scala index 60060bf02913b..f011a5ff15ea9 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisSuite.scala @@ -19,8 +19,8 @@ package org.apache.spark.sql.catalyst.analysis import org.scalatest.{BeforeAndAfter, FunSuite} +import org.apache.spark.sql.AnalysisException import org.apache.spark.sql.catalyst.expressions.{Alias, AttributeReference} -import org.apache.spark.sql.catalyst.errors.TreeNodeException import org.apache.spark.sql.catalyst.plans.logical._ import org.apache.spark.sql.types._ @@ -69,12 +69,12 @@ class AnalysisSuite extends FunSuite with BeforeAndAfter { UnresolvedRelation(Seq("TaBlE"), Some("TbL")))) === Project(testRelation.output, testRelation)) - val e = intercept[TreeNodeException[_]] { + val e = intercept[AnalysisException] { caseSensitiveAnalyze( Project(Seq(UnresolvedAttribute("tBl.a")), UnresolvedRelation(Seq("TaBlE"), Some("TbL")))) } - assert(e.getMessage().toLowerCase.contains("unresolved")) + assert(e.getMessage().toLowerCase.contains("cannot resolve")) assert( caseInsensitiveAnalyze( @@ -109,10 +109,10 @@ class AnalysisSuite extends FunSuite with BeforeAndAfter { } test("throw errors for unresolved attributes during analysis") { - val e = intercept[TreeNodeException[_]] { + val e = intercept[AnalysisException] { caseSensitiveAnalyze(Project(Seq(UnresolvedAttribute("abcd")), testRelation)) } - assert(e.getMessage().toLowerCase.contains("unresolved attribute")) + assert(e.getMessage().toLowerCase.contains("cannot resolve")) } test("throw errors for unresolved plans during analysis") { @@ -120,10 +120,10 @@ class AnalysisSuite extends FunSuite with BeforeAndAfter { override lazy val resolved = false override def output = Nil } - val e = intercept[TreeNodeException[_]] { + val e = intercept[AnalysisException] { caseSensitiveAnalyze(UnresolvedTestPlan()) } - assert(e.getMessage().toLowerCase.contains("unresolved plan")) + assert(e.getMessage().toLowerCase.contains("unresolved")) } test("divide should be casted into fractional types") { diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala index 7cf6c80194f6c..dcfd8b28cb02a 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala @@ -851,7 +851,7 @@ class ExpressionEvaluationSuite extends FunSuite { expr.dataType match { case StructType(fields) => val field = fields.find(_.name == fieldName).get - GetField(expr, field, fields.indexOf(field)) + StructGetField(expr, field, fields.indexOf(field)) } } diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConvertToLocalRelationSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConvertToLocalRelationSuite.scala new file mode 100644 index 0000000000000..cf42d43823399 --- /dev/null +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ConvertToLocalRelationSuite.scala @@ -0,0 +1,57 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.optimizer + +import org.apache.spark.sql.Row +import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute +import org.apache.spark.sql.catalyst.dsl.plans._ +import org.apache.spark.sql.catalyst.dsl.expressions._ +import org.apache.spark.sql.catalyst.plans.PlanTest +import org.apache.spark.sql.catalyst.plans.logical.{LocalRelation, LogicalPlan} +import org.apache.spark.sql.catalyst.rules.RuleExecutor + + +class ConvertToLocalRelationSuite extends PlanTest { + + object Optimize extends RuleExecutor[LogicalPlan] { + val batches = + Batch("LocalRelation", FixedPoint(100), + ConvertToLocalRelation) :: Nil + } + + test("Project on LocalRelation should be turned into a single LocalRelation") { + val testRelation = LocalRelation( + LocalRelation('a.int, 'b.int).output, + Row(1, 2) :: + Row(4, 5) :: Nil) + + val correctAnswer = LocalRelation( + LocalRelation('a1.int, 'b1.int).output, + Row(1, 3) :: + Row(4, 6) :: Nil) + + val projectOnLocal = testRelation.select( + UnresolvedAttribute("a").as("a1"), + (UnresolvedAttribute("b") + 1).as("b1")) + + val optimized = Optimize(projectOnLocal.analyze) + + comparePlans(optimized, correctAnswer) + } + +} diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/FilterPushdownSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/FilterPushdownSuite.scala index ebb123c1f909e..1158b5dfc6147 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/FilterPushdownSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/FilterPushdownSuite.scala @@ -19,11 +19,13 @@ package org.apache.spark.sql.catalyst.optimizer import org.apache.spark.sql.catalyst.analysis import org.apache.spark.sql.catalyst.analysis.EliminateAnalysisOperators +import org.apache.spark.sql.catalyst.expressions.Explode import org.apache.spark.sql.catalyst.plans.logical._ import org.apache.spark.sql.catalyst.plans.{PlanTest, LeftOuter, RightOuter} import org.apache.spark.sql.catalyst.rules._ import org.apache.spark.sql.catalyst.dsl.plans._ import org.apache.spark.sql.catalyst.dsl.expressions._ +import org.apache.spark.sql.types.IntegerType class FilterPushdownSuite extends PlanTest { @@ -34,7 +36,8 @@ class FilterPushdownSuite extends PlanTest { Batch("Filter Pushdown", Once, CombineFilters, PushPredicateThroughProject, - PushPredicateThroughJoin) :: Nil + PushPredicateThroughJoin, + PushPredicateThroughGenerate) :: Nil } val testRelation = LocalRelation('a.int, 'b.int, 'c.int) @@ -411,4 +414,62 @@ class FilterPushdownSuite extends PlanTest { comparePlans(optimized, analysis.EliminateAnalysisOperators(correctAnswer)) } + + val testRelationWithArrayType = LocalRelation('a.int, 'b.int, 'c_arr.array(IntegerType)) + + test("generate: predicate referenced no generated column") { + val originalQuery = { + testRelationWithArrayType + .generate(Explode(Seq("c"), 'c_arr), true, false, Some("arr")) + .where(('b >= 5) && ('a > 6)) + } + val optimized = Optimize(originalQuery.analyze) + val correctAnswer = { + testRelationWithArrayType + .where(('b >= 5) && ('a > 6)) + .generate(Explode(Seq("c"), 'c_arr), true, false, Some("arr")).analyze + } + + comparePlans(optimized, correctAnswer) + } + + test("generate: part of conjuncts referenced generated column") { + val generator = Explode(Seq("c"), 'c_arr) + val originalQuery = { + testRelationWithArrayType + .generate(generator, true, false, Some("arr")) + .where(('b >= 5) && ('c > 6)) + } + val optimized = Optimize(originalQuery.analyze) + val referenceResult = { + testRelationWithArrayType + .where('b >= 5) + .generate(generator, true, false, Some("arr")) + .where('c > 6).analyze + } + + // Since newly generated columns get different ids every time being analyzed + // e.g. comparePlans(originalQuery.analyze, originalQuery.analyze) fails. + // So we check operators manually here. + // Filter("c" > 6) + assertResult(classOf[Filter])(optimized.getClass) + assertResult(1)(optimized.asInstanceOf[Filter].condition.references.size) + assertResult("c"){ + optimized.asInstanceOf[Filter].condition.references.toSeq(0).name + } + + // the rest part + comparePlans(optimized.children(0), referenceResult.children(0)) + } + + test("generate: all conjuncts referenced generated column") { + val originalQuery = { + testRelationWithArrayType + .generate(Explode(Seq("c"), 'c_arr), true, false, Some("arr")) + .where(('c > 6) || ('b > 5)).analyze + } + val optimized = Optimize(originalQuery) + + comparePlans(optimized, originalQuery) + } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/Column.scala b/sql/core/src/main/scala/org/apache/spark/sql/Column.scala index 878b2b0556de7..1011bf0bb5ef4 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/Column.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/Column.scala @@ -549,6 +549,15 @@ trait Column extends DataFrame { */ override def as(alias: String): Column = exprToColumn(Alias(expr, alias)()) + /** + * Gives the column an alias. + * {{{ + * // Renames colA to colB in select output. + * df.select($"colA".as('colB)) + * }}} + */ + override def as(alias: Symbol): Column = exprToColumn(Alias(expr, alias.name)()) + /** * Casts the column to a different data type. * {{{ diff --git a/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala b/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala index 8ad6526f872e5..04e0d09947492 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/DataFrame.scala @@ -27,6 +27,8 @@ import org.apache.spark.sql.catalyst.plans.logical._ import org.apache.spark.sql.types.StructType import org.apache.spark.util.Utils +import scala.util.control.NonFatal + private[sql] object DataFrame { def apply(sqlContext: SQLContext, logicalPlan: LogicalPlan): DataFrame = { @@ -92,6 +94,12 @@ trait DataFrame extends RDDApi[Row] { */ def toDataFrame: DataFrame = this + override def toString = + try schema.map(f => s"${f.name}: ${f.dataType.simpleString}").mkString("[", ", ", "]") catch { + case NonFatal(e) => + s"Invalid tree; ${e.getMessage}:\n$queryExecution" + } + /** * Returns a new [[DataFrame]] with columns renamed. This can be quite convenient in conversion * from a RDD of tuples into a [[DataFrame]] with meaningful names. For example: @@ -102,7 +110,7 @@ trait DataFrame extends RDDApi[Row] { * }}} */ @scala.annotation.varargs - def toDataFrame(colName: String, colNames: String*): DataFrame + def toDataFrame(colNames: String*): DataFrame /** Returns the schema of this [[DataFrame]]. */ def schema: StructType @@ -116,6 +124,25 @@ trait DataFrame extends RDDApi[Row] { /** Prints the schema to the console in a nice tree format. */ def printSchema(): Unit + /** + * Returns true if the `collect` and `take` methods can be run locally + * (without any Spark executors). + */ + def isLocal: Boolean + + /** + * Displays the [[DataFrame]] in a tabular form. For example: + * {{{ + * year month AVG('Adj Close) MAX('Adj Close) + * 1980 12 0.503218 0.595103 + * 1981 01 0.523289 0.570307 + * 1982 02 0.436504 0.475256 + * 1983 03 0.410516 0.442194 + * 1984 04 0.450090 0.483521 + * }}} + */ + def show(): Unit + /** * Cartesian join with another [[DataFrame]]. * @@ -137,7 +164,7 @@ trait DataFrame extends RDDApi[Row] { def join(right: DataFrame, joinExprs: Column): DataFrame /** - * Join with another [[DataFrame]], usin g the given join expression. The following performs + * Join with another [[DataFrame]], using the given join expression. The following performs * a full outer join between `df1` and `df2`. * * {{{ @@ -214,7 +241,12 @@ trait DataFrame extends RDDApi[Row] { /** * Returns a new [[DataFrame]] with an alias set. */ - def as(name: String): DataFrame + def as(alias: String): DataFrame + + /** + * (Scala-specific) Returns a new [[DataFrame]] with an alias set. + */ + def as(alias: Symbol): DataFrame /** * Selects a set of expressions. @@ -497,6 +529,9 @@ trait DataFrame extends RDDApi[Row] { */ override def repartition(numPartitions: Int): DataFrame + /** Returns a new [[DataFrame]] that contains only the unique rows from this [[DataFrame]]. */ + override def distinct: DataFrame + override def persist(): this.type override def persist(newLevel: StorageLevel): this.type diff --git a/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala b/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala index 789bcf6184b3e..1ee16ad5161c8 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/DataFrameImpl.scala @@ -90,14 +90,13 @@ private[sql] class DataFrameImpl protected[sql]( } } - override def toDataFrame(colName: String, colNames: String*): DataFrame = { - val newNames = colName +: colNames - require(schema.size == newNames.size, + override def toDataFrame(colNames: String*): DataFrame = { + require(schema.size == colNames.size, "The number of columns doesn't match.\n" + "Old column names: " + schema.fields.map(_.name).mkString(", ") + "\n" + - "New column names: " + newNames.mkString(", ")) + "New column names: " + colNames.mkString(", ")) - val newCols = schema.fieldNames.zip(newNames).map { case (oldName, newName) => + val newCols = schema.fieldNames.zip(colNames).map { case (oldName, newName) => apply(oldName).as(newName) } select(newCols :_*) @@ -113,6 +112,38 @@ private[sql] class DataFrameImpl protected[sql]( override def printSchema(): Unit = println(schema.treeString) + override def isLocal: Boolean = { + logicalPlan.isInstanceOf[LocalRelation] + } + + override def show(): Unit = { + val data = take(20) + val numCols = schema.fieldNames.length + + // For cells that are beyond 20 characters, replace it with the first 17 and "..." + val rows: Seq[Seq[String]] = schema.fieldNames.toSeq +: data.map { row => + row.toSeq.map { cell => + val str = if (cell == null) "null" else cell.toString + if (str.length > 20) str.substring(0, 17) + "..." else str + } : Seq[String] + } + + // Compute the width of each column + val colWidths = Array.fill(numCols)(0) + for (row <- rows) { + for ((cell, i) <- row.zipWithIndex) { + colWidths(i) = math.max(colWidths(i), cell.length) + } + } + + // Pad the cells and print them + println(rows.map { row => + row.zipWithIndex.map { case (cell, i) => + String.format(s"%-${colWidths(i)}s", cell) + }.mkString(" ") + }.mkString("\n")) + } + override def join(right: DataFrame): DataFrame = { Join(logicalPlan, right.logicalPlan, joinType = Inner, None) } @@ -165,16 +196,16 @@ private[sql] class DataFrameImpl protected[sql]( }.toSeq :_*) } - override def as(name: String): DataFrame = Subquery(name, logicalPlan) + override def as(alias: String): DataFrame = Subquery(alias, logicalPlan) + + override def as(alias: Symbol): DataFrame = Subquery(alias.name, logicalPlan) override def select(cols: Column*): DataFrame = { - val exprs = cols.zipWithIndex.map { - case (Column(expr: NamedExpression), _) => - expr - case (Column(expr: Expression), _) => - Alias(expr, expr.toString)() + val namedExpressions = cols.map { + case Column(expr: NamedExpression) => expr + case Column(expr: Expression) => Alias(expr, expr.prettyString)() } - Project(exprs.toSeq, logicalPlan) + Project(namedExpressions.toSeq, logicalPlan) } override def select(col: String, cols: String*): DataFrame = { @@ -184,7 +215,19 @@ private[sql] class DataFrameImpl protected[sql]( override def selectExpr(exprs: String*): DataFrame = { select(exprs.map { expr => Column(new SqlParser().parseExpression(expr)) - } :_*) + }: _*) + } + + override def addColumn(colName: String, col: Column): DataFrame = { + select(Column("*"), col.as(colName)) + } + + override def renameColumn(existingName: String, newName: String): DataFrame = { + val colNames = schema.map { field => + val name = field.name + if (name == existingName) Column(name).as(newName) else Column(name) + } + select(colNames :_*) } override def filter(condition: Column): DataFrame = { @@ -233,18 +276,8 @@ private[sql] class DataFrameImpl protected[sql]( } ///////////////////////////////////////////////////////////////////////////// - - override def addColumn(colName: String, col: Column): DataFrame = { - select(Column("*"), col.as(colName)) - } - - override def renameColumn(existingName: String, newName: String): DataFrame = { - val colNames = schema.map { field => - val name = field.name - if (name == existingName) Column(name).as(newName) else Column(name) - } - select(colNames :_*) - } + // RDD API + ///////////////////////////////////////////////////////////////////////////// override def head(n: Int): Array[Row] = limit(n).collect() @@ -276,6 +309,8 @@ private[sql] class DataFrameImpl protected[sql]( sqlContext.applySchema(rdd.repartition(numPartitions), schema) } + override def distinct: DataFrame = Distinct(logicalPlan) + override def persist(): this.type = { sqlContext.cacheManager.cacheQuery(this) this diff --git a/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala b/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala index 6043fb4dee01d..ce0557b88196f 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/IncomputableColumn.scala @@ -40,6 +40,8 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten throw new UnsupportedOperationException("Cannot run this method on an UncomputableColumn") } + override def toString = expr.prettyString + override def isComputable: Boolean = false override val sqlContext: SQLContext = null @@ -48,7 +50,7 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten protected[sql] override def logicalPlan: LogicalPlan = err() - override def toDataFrame(colName: String, colNames: String*): DataFrame = err() + override def toDataFrame(colNames: String*): DataFrame = err() override def schema: StructType = err() @@ -58,6 +60,10 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten override def printSchema(): Unit = err() + override def show(): Unit = err() + + override def isLocal: Boolean = false + override def join(right: DataFrame): DataFrame = err() override def join(right: DataFrame, joinExprs: Column): DataFrame = err() @@ -82,6 +88,10 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten override def selectExpr(exprs: String*): DataFrame = err() + override def addColumn(colName: String, col: Column): DataFrame = err() + + override def renameColumn(existingName: String, newName: String): DataFrame = err() + override def filter(condition: Column): DataFrame = err() override def filter(conditionExpr: String): DataFrame = err() @@ -106,10 +116,6 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten ///////////////////////////////////////////////////////////////////////////// - override def addColumn(colName: String, col: Column): DataFrame = err() - - override def renameColumn(existingName: String, newName: String): DataFrame = err() - override def head(n: Int): Array[Row] = err() override def head(): Row = err() @@ -136,6 +142,8 @@ private[sql] class IncomputableColumn(protected[sql] val expr: Expression) exten override def repartition(numPartitions: Int): DataFrame = err() + override def distinct: DataFrame = err() + override def persist(): this.type = err() override def persist(newLevel: StorageLevel): this.type = err() diff --git a/sql/core/src/main/scala/org/apache/spark/sql/RDDApi.scala b/sql/core/src/main/scala/org/apache/spark/sql/RDDApi.scala index 38e6382f171d5..df866fd1ad8ad 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/RDDApi.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/RDDApi.scala @@ -60,4 +60,6 @@ private[sql] trait RDDApi[T] { def first(): T def repartition(numPartitions: Int): DataFrame + + def distinct: DataFrame } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala b/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala index bf3990671029e..97e3777f933e4 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala @@ -32,7 +32,7 @@ import org.apache.spark.sql.catalyst.ScalaReflection import org.apache.spark.sql.catalyst.analysis._ import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.optimizer.{DefaultOptimizer, Optimizer} -import org.apache.spark.sql.catalyst.plans.logical.{LocalRelation, LogicalPlan} +import org.apache.spark.sql.catalyst.plans.logical.{LocalRelation, LogicalPlan, NoRelation} import org.apache.spark.sql.catalyst.rules.RuleExecutor import org.apache.spark.sql.execution._ import org.apache.spark.sql.jdbc.{JDBCPartition, JDBCPartitioningInfo, JDBCRelation} @@ -130,6 +130,9 @@ class SQLContext(@transient val sparkContext: SparkContext) */ val experimental: ExperimentalMethods = new ExperimentalMethods(this) + /** Returns a [[DataFrame]] with no rows or columns. */ + lazy val emptyDataFrame = DataFrame(this, NoRelation) + /** * A collection of methods for registering user-defined functions (UDF). * diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala index 81bcf5a6f32dd..edf8a5be64ff1 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala @@ -342,11 +342,6 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] { ExecutedCommand( RunnableDescribeCommand(resultPlan, resultPlan.output, isExtended)) :: Nil - case LogicalDescribeCommand(table, isExtended) => - val resultPlan = self.sqlContext.executePlan(table).executedPlan - ExecutedCommand( - RunnableDescribeCommand(resultPlan, resultPlan.output, isExtended)) :: Nil - case _ => Nil } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala index 66aed5d5113d1..4dc506c21ab9e 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala @@ -17,9 +17,6 @@ package org.apache.spark.sql.execution -import scala.collection.mutable.ArrayBuffer -import scala.reflect.runtime.universe.TypeTag - import org.apache.spark.{SparkEnv, HashPartitioner, SparkConf} import org.apache.spark.annotation.DeveloperApi import org.apache.spark.rdd.{RDD, ShuffledRDD} @@ -40,7 +37,7 @@ case class Project(projectList: Seq[NamedExpression], child: SparkPlan) extends @transient lazy val buildProjection = newMutableProjection(projectList, child.output) - def execute() = child.execute().mapPartitions { iter => + override def execute() = child.execute().mapPartitions { iter => val resuableProjection = buildProjection() iter.map(resuableProjection) } @@ -55,7 +52,7 @@ case class Filter(condition: Expression, child: SparkPlan) extends UnaryNode { @transient lazy val conditionEvaluator = newPredicate(condition, child.output) - def execute() = child.execute().mapPartitions { iter => + override def execute() = child.execute().mapPartitions { iter => iter.filter(conditionEvaluator) } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/debug/package.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/debug/package.scala index 5cc67cdd13944..acef49aabfe70 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/execution/debug/package.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/debug/package.scala @@ -22,7 +22,7 @@ import scala.collection.mutable.HashSet import org.apache.spark.{AccumulatorParam, Accumulator, SparkContext} import org.apache.spark.annotation.DeveloperApi import org.apache.spark.SparkContext._ -import org.apache.spark.sql.{DataFrame, Row} +import org.apache.spark.sql.{SQLConf, SQLContext, DataFrame, Row} import org.apache.spark.sql.catalyst.trees.TreeNodeRef import org.apache.spark.sql.types._ @@ -37,6 +37,15 @@ import org.apache.spark.sql.types._ */ package object debug { + /** + * Augments [[SQLContext]] with debug methods. + */ + implicit class DebugSQLContext(sqlContext: SQLContext) { + def debug() = { + sqlContext.setConf(SQLConf.DATAFRAME_EAGER_ANALYSIS, "false") + } + } + /** * :: DeveloperApi :: * Augments [[DataFrame]]s with debug methods. diff --git a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRDD.scala b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRDD.scala index a2f94675fb5a3..87304ce2496b4 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRDD.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRDD.scala @@ -17,13 +17,10 @@ package org.apache.spark.sql.jdbc -import java.sql.{Connection, DatabaseMetaData, DriverManager, ResultSet, ResultSetMetaData, SQLException} -import scala.collection.mutable.ArrayBuffer +import java.sql.{Connection, DriverManager, ResultSet, ResultSetMetaData, SQLException} import org.apache.spark.{Logging, Partition, SparkContext, TaskContext} import org.apache.spark.rdd.RDD -import org.apache.spark.util.NextIterator -import org.apache.spark.sql.catalyst.analysis.HiveTypeCoercion import org.apache.spark.sql.catalyst.expressions.{Row, SpecificMutableRow} import org.apache.spark.sql.types._ import org.apache.spark.sql.sources._ @@ -100,7 +97,7 @@ private[sql] object JDBCRDD extends Logging { try { val rsmd = rs.getMetaData val ncols = rsmd.getColumnCount - var fields = new Array[StructField](ncols); + val fields = new Array[StructField](ncols) var i = 0 while (i < ncols) { val columnName = rsmd.getColumnName(i + 1) @@ -176,23 +173,27 @@ private[sql] object JDBCRDD extends Logging { * * @return An RDD representing "SELECT requiredColumns FROM fqTable". */ - def scanTable(sc: SparkContext, - schema: StructType, - driver: String, - url: String, - fqTable: String, - requiredColumns: Array[String], - filters: Array[Filter], - parts: Array[Partition]): RDD[Row] = { + def scanTable( + sc: SparkContext, + schema: StructType, + driver: String, + url: String, + fqTable: String, + requiredColumns: Array[String], + filters: Array[Filter], + parts: Array[Partition]): RDD[Row] = { + val prunedSchema = pruneSchema(schema, requiredColumns) - return new JDBCRDD(sc, - getConnector(driver, url), - prunedSchema, - fqTable, - requiredColumns, - filters, - parts) + return new + JDBCRDD( + sc, + getConnector(driver, url), + prunedSchema, + fqTable, + requiredColumns, + filters, + parts) } } @@ -369,21 +370,21 @@ private[sql] class JDBCRDD( def close() { if (closed) return try { - if (null != rs && ! rs.isClosed()) { + if (null != rs) { rs.close() } } catch { case e: Exception => logWarning("Exception closing resultset", e) } try { - if (null != stmt && ! stmt.isClosed()) { + if (null != stmt) { stmt.close() } } catch { case e: Exception => logWarning("Exception closing statement", e) } try { - if (null != conn && ! conn.isClosed()) { + if (null != conn) { conn.close() } logInfo("closed connection") @@ -412,6 +413,5 @@ private[sql] class JDBCRDD( gotNext = false nextValue } - } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRelation.scala index e09125e406ba2..66ad38eb7c45b 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRelation.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JDBCRelation.scala @@ -96,7 +96,8 @@ private[sql] class DefaultSource extends RelationProvider { if (driver != null) Class.forName(driver) - if ( partitionColumn != null + if ( + partitionColumn != null && (lowerBound == null || upperBound == null || numPartitions == null)) { sys.error("Partitioning incompletely specified") } @@ -104,30 +105,34 @@ private[sql] class DefaultSource extends RelationProvider { val partitionInfo = if (partitionColumn == null) { null } else { - JDBCPartitioningInfo(partitionColumn, - lowerBound.toLong, upperBound.toLong, - numPartitions.toInt) + JDBCPartitioningInfo( + partitionColumn, + lowerBound.toLong, + upperBound.toLong, + numPartitions.toInt) } val parts = JDBCRelation.columnPartition(partitionInfo) JDBCRelation(url, table, parts)(sqlContext) } } -private[sql] case class JDBCRelation(url: String, - table: String, - parts: Array[Partition])( - @transient val sqlContext: SQLContext) - extends PrunedFilteredScan { +private[sql] case class JDBCRelation( + url: String, + table: String, + parts: Array[Partition])(@transient val sqlContext: SQLContext) extends PrunedFilteredScan { override val schema = JDBCRDD.resolveTable(url, table) override def buildScan(requiredColumns: Array[String], filters: Array[Filter]) = { val driver: String = DriverManager.getDriver(url).getClass.getCanonicalName - JDBCRDD.scanTable(sqlContext.sparkContext, - schema, - driver, url, - table, - requiredColumns, filters, - parts) + JDBCRDD.scanTable( + sqlContext.sparkContext, + schema, + driver, + url, + table, + requiredColumns, + filters, + parts) } } diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala index a640ba57e0885..5eecc303ef72b 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala @@ -87,13 +87,13 @@ trait CreatableRelationProvider { /** * ::DeveloperApi:: - * Represents a collection of tuples with a known schema. Classes that extend BaseRelation must - * be able to produce the schema of their data in the form of a [[StructType]] Concrete + * Represents a collection of tuples with a known schema. Classes that extend BaseRelation must + * be able to produce the schema of their data in the form of a [[StructType]]. Concrete * implementation should inherit from one of the descendant `Scan` classes, which define various * abstract methods for execution. * * BaseRelations must also define a equality function that only returns true when the two - * instances will return the same data. This equality function is used when determining when + * instances will return the same data. This equality function is used when determining when * it is safe to substitute cached results for a given relation. */ @DeveloperApi @@ -102,13 +102,16 @@ abstract class BaseRelation { def schema: StructType /** - * Returns an estimated size of this relation in bytes. This information is used by the planner + * Returns an estimated size of this relation in bytes. This information is used by the planner * to decided when it is safe to broadcast a relation and can be overridden by sources that * know the size ahead of time. By default, the system will assume that tables are too - * large to broadcast. This method will be called multiple times during query planning + * large to broadcast. This method will be called multiple times during query planning * and thus should not perform expensive operations for each invocation. + * + * Note that it is always better to overestimate size than underestimate, because underestimation + * could lead to execution plans that are suboptimal (i.e. broadcasting a very large table). */ - def sizeInBytes = sqlContext.conf.defaultSizeInBytes + def sizeInBytes: Long = sqlContext.conf.defaultSizeInBytes } /** diff --git a/sql/core/src/main/scala/org/apache/spark/sql/test/ExamplePointUDT.scala b/sql/core/src/main/scala/org/apache/spark/sql/test/ExamplePointUDT.scala index e6f622e87f7a4..eb045e37bf5a9 100644 --- a/sql/core/src/main/scala/org/apache/spark/sql/test/ExamplePointUDT.scala +++ b/sql/core/src/main/scala/org/apache/spark/sql/test/ExamplePointUDT.scala @@ -37,7 +37,7 @@ private[sql] class ExamplePointUDT extends UserDefinedType[ExamplePoint] { override def sqlType: DataType = ArrayType(DoubleType, false) - override def pyUDT: String = "pyspark.sql_tests.ExamplePointUDT" + override def pyUDT: String = "pyspark.sql.tests.ExamplePointUDT" override def serialize(obj: Any): Seq[Double] = { obj match { diff --git a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java index e5588938ea162..a21a15409080c 100644 --- a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java +++ b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaAPISuite.java @@ -19,6 +19,7 @@ import java.io.Serializable; +import org.apache.spark.sql.test.TestSQLContext$; import org.junit.After; import org.junit.Before; import org.junit.Test; @@ -37,14 +38,12 @@ public class JavaAPISuite implements Serializable { @Before public void setUp() { - sc = new JavaSparkContext("local", "JavaAPISuite"); - sqlContext = new SQLContext(sc); + sqlContext = TestSQLContext$.MODULE$; + sc = new JavaSparkContext(sqlContext.sparkContext()); } @After public void tearDown() { - sc.stop(); - sc = null; } @SuppressWarnings("unchecked") diff --git a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java index 8510bac499092..2e6e977fdc752 100644 --- a/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java +++ b/sql/core/src/test/java/org/apache/spark/sql/api/java/JavaApplySchemaSuite.java @@ -22,6 +22,7 @@ import java.util.Arrays; import java.util.List; +import org.apache.spark.sql.test.TestSQLContext$; import org.junit.After; import org.junit.Assert; import org.junit.Before; @@ -42,13 +43,12 @@ public class JavaApplySchemaSuite implements Serializable { @Before public void setUp() { - javaCtx = new JavaSparkContext("local", "JavaApplySchemaSuite"); - javaSqlCtx = new SQLContext(javaCtx); + javaSqlCtx = TestSQLContext$.MODULE$; + javaCtx = new JavaSparkContext(javaSqlCtx.sparkContext()); } @After public void tearDown() { - javaCtx.stop(); javaCtx = null; javaSqlCtx = null; } diff --git a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala index 5aa3db720c886..02623f73c7f76 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala @@ -17,6 +17,8 @@ package org.apache.spark.sql +import org.apache.spark.sql.TestData._ + import scala.language.postfixOps import org.apache.spark.sql.Dsl._ @@ -53,6 +55,33 @@ class DataFrameSuite extends QueryTest { TestSQLContext.setConf(SQLConf.DATAFRAME_EAGER_ANALYSIS, oldSetting.toString) } + test("dataframe toString") { + assert(testData.toString === "[key: int, value: string]") + assert(testData("key").toString === "[key: int]") + } + + test("incomputable toString") { + assert($"test".toString === "test") + } + + test("invalid plan toString, debug mode") { + val oldSetting = TestSQLContext.conf.dataFrameEagerAnalysis + TestSQLContext.setConf(SQLConf.DATAFRAME_EAGER_ANALYSIS, "true") + + // Turn on debug mode so we can see invalid query plans. + import org.apache.spark.sql.execution.debug._ + TestSQLContext.debug() + + val badPlan = testData.select('badColumn) + + assert(badPlan.toString contains badPlan.queryExecution.toString, + "toString on bad query plans should include the query execution but was:\n" + + badPlan.toString) + + // Set the flag back to original value before this test. + TestSQLContext.setConf(SQLConf.DATAFRAME_EAGER_ANALYSIS, oldSetting.toString) + } + test("table scan") { checkAnswer( testData, diff --git a/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala index 11502edf972e9..55fd0b0892fa1 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala @@ -589,7 +589,7 @@ class SQLQuerySuite extends QueryTest with BeforeAndAfterAll { ("1" :: "2" :: "3" :: "4" :: "A" :: "B" :: "C" :: "D" :: "E" :: "F" :: Nil).map(Row(_))) // Column type mismatches where a coercion is not possible, in this case between integer // and array types, trigger a TreeNodeException. - intercept[TreeNodeException[_]] { + intercept[AnalysisException] { sql("SELECT data FROM arrayData UNION SELECT 1 FROM arrayData").collect() } } diff --git a/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala index 926ba68828ee8..7870cf9b0a868 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/json/JsonSuite.scala @@ -342,21 +342,19 @@ class JsonSuite extends QueryTest { ) } - ignore("Complex field and type inferring (Ignored)") { + test("GetField operation on complex data type") { val jsonDF = jsonRDD(complexFieldAndType1) jsonDF.registerTempTable("jsonTable") - // Right now, "field1" and "field2" are treated as aliases. We should fix it. checkAnswer( sql("select arrayOfStruct[0].field1, arrayOfStruct[0].field2 from jsonTable"), Row(true, "str1") ) - // Right now, the analyzer cannot resolve arrayOfStruct.field1 and arrayOfStruct.field2. // Getting all values of a specific field from an array of structs. checkAnswer( sql("select arrayOfStruct.field1, arrayOfStruct.field2 from jsonTable"), - Row(Seq(true, false), Seq("str1", null)) + Row(Seq(true, false, null), Seq("str1", null, null)) ) } diff --git a/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala index 48c7598343e55..cba06835f9a61 100644 --- a/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala +++ b/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetQuerySuite.scala @@ -35,7 +35,6 @@ class ParquetQuerySuite extends QueryTest with ParquetTest { } } - // TODO Re-enable this after data source insertion API is merged test(s"$prefix: appending") { val data = (0 until 10).map(i => (i, i.toString)) withParquetTable(data, "t") { diff --git a/sql/hive/compatibility/src/test/scala/org/apache/spark/sql/hive/execution/HiveCompatibilitySuite.scala b/sql/hive/compatibility/src/test/scala/org/apache/spark/sql/hive/execution/HiveCompatibilitySuite.scala index a6266f611c219..e443e5bd5f54d 100644 --- a/sql/hive/compatibility/src/test/scala/org/apache/spark/sql/hive/execution/HiveCompatibilitySuite.scala +++ b/sql/hive/compatibility/src/test/scala/org/apache/spark/sql/hive/execution/HiveCompatibilitySuite.scala @@ -518,6 +518,7 @@ class HiveCompatibilitySuite extends HiveQueryFileTest with BeforeAndAfter { "inputddl2", "inputddl3", "inputddl4", + "inputddl5", "inputddl6", "inputddl7", "inputddl8", diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala index ad37b7d0e6f59..2c00659496972 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala @@ -424,6 +424,11 @@ class HiveContext(sc: SparkContext) extends SQLContext(sc) { /** Extends QueryExecution with hive specific features. */ protected[sql] class QueryExecution(logicalPlan: LogicalPlan) extends super.QueryExecution(logicalPlan) { + // Like what we do in runHive, makes sure the session represented by the + // `sessionState` field is activated. + if (SessionState.get() != sessionState) { + SessionState.start(sessionState) + } /** * Returns the result as a hive compatible sequence of strings. For native commands, the diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala index 2a4b88092179f..8618301ba84d6 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala @@ -18,6 +18,7 @@ package org.apache.spark.sql.hive import java.sql.Date + import scala.collection.mutable.ArrayBuffer import org.apache.hadoop.hive.conf.HiveConf @@ -76,6 +77,7 @@ private[hive] object HiveQl { "TOK_REVOKE", "TOK_SHOW_GRANT", "TOK_SHOW_ROLE_GRANT", + "TOK_SHOW_SET_ROLE", "TOK_CREATEFUNCTION", "TOK_DROPFUNCTION", @@ -103,6 +105,7 @@ private[hive] object HiveQl { "TOK_CREATEINDEX", "TOK_DROPDATABASE", "TOK_DROPINDEX", + "TOK_DROPTABLE_PROPERTIES", "TOK_MSCK", "TOK_ALTERVIEW_ADDPARTS", @@ -111,6 +114,7 @@ private[hive] object HiveQl { "TOK_ALTERVIEW_PROPERTIES", "TOK_ALTERVIEW_RENAME", "TOK_CREATEVIEW", + "TOK_DROPVIEW_PROPERTIES", "TOK_DROPVIEW", "TOK_EXPORT", @@ -1235,6 +1239,9 @@ https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation%2C+Cube%2C case ast: ASTNode if ast.getType == HiveParser.TOK_DATELITERAL => Literal(Date.valueOf(ast.getText.substring(1, ast.getText.length - 1))) + case ast: ASTNode if ast.getType == HiveParser.TOK_CHARSETLITERAL => + Literal(BaseSemanticAnalyzer.charSetString(ast.getChild(0).getText, ast.getChild(1).getText)) + case a: ASTNode => throw new NotImplementedError( s"""No parse rules for ASTNode type: ${a.getType}, text: ${a.getText} : diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveWriterContainers.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveWriterContainers.scala index aae175e426ade..f136e43acc8f2 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveWriterContainers.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveWriterContainers.scala @@ -30,6 +30,7 @@ import org.apache.hadoop.hive.ql.io.{HiveFileFormatUtils, HiveOutputFormat} import org.apache.hadoop.hive.ql.plan.{PlanUtils, TableDesc} import org.apache.hadoop.io.Writable import org.apache.hadoop.mapred._ +import org.apache.hadoop.hive.common.FileUtils import org.apache.spark.mapred.SparkHadoopMapRedUtil import org.apache.spark.sql.Row @@ -212,9 +213,14 @@ private[spark] class SparkHiveDynamicPartitionWriterContainer( .zip(row.toSeq.takeRight(dynamicPartColNames.length)) .map { case (col, rawVal) => val string = if (rawVal == null) null else String.valueOf(rawVal) - s"/$col=${if (string == null || string.isEmpty) defaultPartName else string}" - } - .mkString + val colString = + if (string == null || string.isEmpty) { + defaultPartName + } else { + FileUtils.escapePathName(string) + } + s"/$col=$colString" + }.mkString def newWriter = { val newFileSinkDesc = new FileSinkDesc( diff --git a/sql/hive/src/test/resources/golden/inputddl5-0-ebbf2aec5f76af7225c2efaf870b8ba7 b/sql/hive/src/test/resources/golden/inputddl5-0-ebbf2aec5f76af7225c2efaf870b8ba7 new file mode 100644 index 0000000000000..e69de29bb2d1d diff --git a/sql/hive/src/test/resources/golden/inputddl5-1-2691407ccdc5c848a4ba2aecb6dbad75 b/sql/hive/src/test/resources/golden/inputddl5-1-2691407ccdc5c848a4ba2aecb6dbad75 new file mode 100644 index 0000000000000..e69de29bb2d1d diff --git a/sql/hive/src/test/resources/golden/inputddl5-2-ca2faacf63dc4785f8bfd2ecc397e69b b/sql/hive/src/test/resources/golden/inputddl5-2-ca2faacf63dc4785f8bfd2ecc397e69b new file mode 100644 index 0000000000000..518a70918b2c7 --- /dev/null +++ b/sql/hive/src/test/resources/golden/inputddl5-2-ca2faacf63dc4785f8bfd2ecc397e69b @@ -0,0 +1 @@ +name string diff --git a/sql/hive/src/test/resources/golden/inputddl5-3-4f28c7412a05cff89c0bd86b65aa7ce b/sql/hive/src/test/resources/golden/inputddl5-3-4f28c7412a05cff89c0bd86b65aa7ce new file mode 100644 index 0000000000000..33398360345d7 --- /dev/null +++ b/sql/hive/src/test/resources/golden/inputddl5-3-4f28c7412a05cff89c0bd86b65aa7ce @@ -0,0 +1 @@ +邵铮 diff --git a/sql/hive/src/test/resources/golden/inputddl5-4-bd7e25cff73f470d2e2336876342b783 b/sql/hive/src/test/resources/golden/inputddl5-4-bd7e25cff73f470d2e2336876342b783 new file mode 100644 index 0000000000000..d00491fd7e5bb --- /dev/null +++ b/sql/hive/src/test/resources/golden/inputddl5-4-bd7e25cff73f470d2e2336876342b783 @@ -0,0 +1 @@ +1 diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/MetastoreDataSourcesSuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/MetastoreDataSourcesSuite.scala index c23575fe96898..036efa84d7c85 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/MetastoreDataSourcesSuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/MetastoreDataSourcesSuite.scala @@ -351,9 +351,6 @@ class MetastoreDataSourcesSuite extends QueryTest with BeforeAndAfterEach { |) """.stripMargin) - new Path("/Users/yhuai/Desktop/whatever") - - val expectedPath = catalog.hiveDefaultTableFilePath("ctasJsonTable") val filesystemPath = new Path(expectedPath) val fs = filesystemPath.getFileSystem(sparkContext.hadoopConfiguration) diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala index 27047ce4b1b0b..405b200d05412 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveQuerySuite.scala @@ -859,6 +859,22 @@ class HiveQuerySuite extends HiveComparisonTest with BeforeAndAfter { } } + test("SPARK-5592: get java.net.URISyntaxException when dynamic partitioning") { + sql(""" + |create table sc as select * + |from (select '2011-01-11', '2011-01-11+14:18:26' from src tablesample (1 rows) + |union all + |select '2011-01-11', '2011-01-11+15:18:26' from src tablesample (1 rows) + |union all + |select '2011-01-11', '2011-01-11+16:18:26' from src tablesample (1 rows) ) s + """.stripMargin) + sql("create table sc_part (key string) partitioned by (ts string) stored as rcfile") + sql("set hive.exec.dynamic.partition=true") + sql("set hive.exec.dynamic.partition.mode=nonstrict") + sql("insert overwrite table sc_part partition(ts) select * from sc") + sql("drop table sc_part") + } + test("Partition spec validation") { sql("DROP TABLE IF EXISTS dp_test") sql("CREATE TABLE dp_test(key INT, value STRING) PARTITIONED BY (dp INT, sp INT)") diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala index eae69af5864aa..e89b4489f15d1 100644 --- a/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala +++ b/sql/hive/src/test/scala/org/apache/spark/sql/parquet/HiveParquetSuite.scala @@ -17,7 +17,7 @@ package org.apache.spark.sql.parquet -import org.apache.spark.sql.QueryTest +import org.apache.spark.sql.{SQLConf, QueryTest} import org.apache.spark.sql.catalyst.expressions.Row import org.apache.spark.sql.hive.test.TestHive @@ -64,8 +64,7 @@ class HiveParquetSuite extends QueryTest with ParquetTest { } } - // TODO Re-enable this after data source insertion API is merged - ignore(s"$prefix: INSERT OVERWRITE TABLE Parquet table") { + test(s"$prefix: INSERT OVERWRITE TABLE Parquet table") { withParquetTable((1 to 10).map(i => (i, s"val_$i")), "t") { withTempPath { file => sql("SELECT * FROM t LIMIT 1").saveAsParquetFile(file.getCanonicalPath) @@ -81,4 +80,12 @@ class HiveParquetSuite extends QueryTest with ParquetTest { } } } + + withSQLConf(SQLConf.PARQUET_USE_DATA_SOURCE_API -> "true") { + run("Parquet data source enabled") + } + + withSQLConf(SQLConf.PARQUET_USE_DATA_SOURCE_API -> "false") { + run("Parquet data source disabled") + } } diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala index 4cc320c5d59b5..a9bf861d160c1 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala @@ -19,9 +19,9 @@ package org.apache.spark.deploy.yarn import scala.util.control.NonFatal -import java.io.IOException +import java.io.{File, IOException} import java.lang.reflect.InvocationTargetException -import java.net.Socket +import java.net.{Socket, URL} import java.util.concurrent.atomic.AtomicReference import akka.actor._ @@ -38,7 +38,8 @@ import org.apache.spark.deploy.{PythonRunner, SparkHadoopUtil} import org.apache.spark.deploy.history.HistoryServer import org.apache.spark.scheduler.cluster.YarnSchedulerBackend import org.apache.spark.scheduler.cluster.CoarseGrainedClusterMessages._ -import org.apache.spark.util.{AkkaUtils, SignalLogger, Utils} +import org.apache.spark.util.{AkkaUtils, ChildFirstURLClassLoader, MutableURLClassLoader, + SignalLogger, Utils} /** * Common application master functionality for Spark on Yarn. @@ -244,7 +245,6 @@ private[spark] class ApplicationMaster( host: String, port: String, isClusterMode: Boolean): Unit = { - val driverUrl = AkkaUtils.address( AkkaUtils.protocol(actorSystem), SparkEnv.driverActorSystemName, @@ -453,12 +453,24 @@ private[spark] class ApplicationMaster( private def startUserApplication(): Thread = { logInfo("Starting the user application in a separate Thread") System.setProperty("spark.executor.instances", args.numExecutors.toString) + + val classpath = Client.getUserClasspath(sparkConf) + val urls = classpath.map { entry => + new URL("file:" + new File(entry.getPath()).getAbsolutePath()) + } + val userClassLoader = + if (Client.isUserClassPathFirst(sparkConf, isDriver = true)) { + new ChildFirstURLClassLoader(urls, Utils.getContextOrSparkClassLoader) + } else { + new MutableURLClassLoader(urls, Utils.getContextOrSparkClassLoader) + } + if (args.primaryPyFile != null && args.primaryPyFile.endsWith(".py")) { System.setProperty("spark.submit.pyFiles", PythonRunner.formatPaths(args.pyFiles).mkString(",")) } - val mainMethod = Class.forName(args.userClass, false, - Thread.currentThread.getContextClassLoader).getMethod("main", classOf[Array[String]]) + val mainMethod = userClassLoader.loadClass(args.userClass) + .getMethod("main", classOf[Array[String]]) val userThread = new Thread { override def run() { @@ -483,6 +495,7 @@ private[spark] class ApplicationMaster( } } } + userThread.setContextClassLoader(userClassLoader) userThread.setName("Driver") userThread.start() userThread diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala index e7005094b5f3c..46d9df93488cb 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala @@ -183,8 +183,7 @@ private[spark] class Client( private[yarn] def copyFileToRemote( destDir: Path, srcPath: Path, - replication: Short, - setPerms: Boolean = false): Path = { + replication: Short): Path = { val destFs = destDir.getFileSystem(hadoopConf) val srcFs = srcPath.getFileSystem(hadoopConf) var destPath = srcPath @@ -193,9 +192,7 @@ private[spark] class Client( logInfo(s"Uploading resource $srcPath -> $destPath") FileUtil.copy(srcFs, srcPath, destFs, destPath, false, hadoopConf) destFs.setReplication(destPath, replication) - if (setPerms) { - destFs.setPermission(destPath, new FsPermission(APP_FILE_PERMISSION)) - } + destFs.setPermission(destPath, new FsPermission(APP_FILE_PERMISSION)) } else { logInfo(s"Source and destination file systems are the same. Not copying $srcPath") } @@ -239,23 +236,22 @@ private[spark] class Client( /** * Copy the given main resource to the distributed cache if the scheme is not "local". * Otherwise, set the corresponding key in our SparkConf to handle it downstream. - * Each resource is represented by a 4-tuple of: + * Each resource is represented by a 3-tuple of: * (1) destination resource name, * (2) local path to the resource, - * (3) Spark property key to set if the scheme is not local, and - * (4) whether to set permissions for this resource + * (3) Spark property key to set if the scheme is not local */ List( - (SPARK_JAR, sparkJar(sparkConf), CONF_SPARK_JAR, false), - (APP_JAR, args.userJar, CONF_SPARK_USER_JAR, true), - ("log4j.properties", oldLog4jConf.orNull, null, false) - ).foreach { case (destName, _localPath, confKey, setPermissions) => + (SPARK_JAR, sparkJar(sparkConf), CONF_SPARK_JAR), + (APP_JAR, args.userJar, CONF_SPARK_USER_JAR), + ("log4j.properties", oldLog4jConf.orNull, null) + ).foreach { case (destName, _localPath, confKey) => val localPath: String = if (_localPath != null) _localPath.trim() else "" if (!localPath.isEmpty()) { val localURI = new URI(localPath) if (localURI.getScheme != LOCAL_SCHEME) { val src = getQualifiedLocalPath(localURI, hadoopConf) - val destPath = copyFileToRemote(dst, src, replication, setPermissions) + val destPath = copyFileToRemote(dst, src, replication) val destFs = FileSystem.get(destPath.toUri(), hadoopConf) distCacheMgr.addResource(destFs, hadoopConf, destPath, localResources, LocalResourceType.FILE, destName, statCache) @@ -435,10 +431,11 @@ private[spark] class Client( // Include driver-specific java options if we are launching a driver if (isClusterMode) { - sparkConf.getOption("spark.driver.extraJavaOptions") + val driverOpts = sparkConf.getOption("spark.driver.extraJavaOptions") .orElse(sys.env.get("SPARK_JAVA_OPTS")) - .map(Utils.splitCommandString).getOrElse(Seq.empty) - .foreach(opts => javaOpts += opts) + driverOpts.foreach { opts => + javaOpts ++= Utils.splitCommandString(opts).map(YarnSparkHadoopUtil.escapeForShell) + } val libraryPaths = Seq(sys.props.get("spark.driver.extraLibraryPath"), sys.props.get("spark.driver.libraryPath")).flatten if (libraryPaths.nonEmpty) { @@ -460,7 +457,7 @@ private[spark] class Client( val msg = s"$amOptsKey is not allowed to alter memory settings (was '$opts')." throw new SparkException(msg) } - javaOpts ++= Utils.splitCommandString(opts) + javaOpts ++= Utils.splitCommandString(opts).map(YarnSparkHadoopUtil.escapeForShell) } } @@ -706,7 +703,7 @@ object Client extends Logging { * Return the path to the given application's staging directory. */ private def getAppStagingDir(appId: ApplicationId): String = { - SPARK_STAGING + Path.SEPARATOR + appId.toString() + Path.SEPARATOR + buildPath(SPARK_STAGING, appId.toString()) } /** @@ -782,7 +779,13 @@ object Client extends Logging { /** * Populate the classpath entry in the given environment map. - * This includes the user jar, Spark jar, and any extra application jars. + * + * User jars are generally not added to the JVM's system classpath; those are handled by the AM + * and executor backend. When the deprecated `spark.yarn.user.classpath.first` is used, user jars + * are included in the system classpath, though. The extra class path and other uploaded files are + * always made available through the system class path. + * + * @param args Client arguments (when starting the AM) or null (when starting executors). */ private[yarn] def populateClasspath( args: ClientArguments, @@ -794,48 +797,38 @@ object Client extends Logging { addClasspathEntry( YarnSparkHadoopUtil.expandEnvironment(Environment.PWD), env ) - - // Normally the users app.jar is last in case conflicts with spark jars if (sparkConf.getBoolean("spark.yarn.user.classpath.first", false)) { - addUserClasspath(args, sparkConf, env) - addFileToClasspath(sparkJar(sparkConf), SPARK_JAR, env) - populateHadoopClasspath(conf, env) - } else { - addFileToClasspath(sparkJar(sparkConf), SPARK_JAR, env) - populateHadoopClasspath(conf, env) - addUserClasspath(args, sparkConf, env) + val userClassPath = + if (args != null) { + getUserClasspath(Option(args.userJar), Option(args.addJars)) + } else { + getUserClasspath(sparkConf) + } + userClassPath.foreach { x => + addFileToClasspath(x, null, env) + } } - - // Append all jar files under the working directory to the classpath. - addClasspathEntry( - YarnSparkHadoopUtil.expandEnvironment(Environment.PWD) + Path.SEPARATOR + "*", env - ) + addFileToClasspath(new URI(sparkJar(sparkConf)), SPARK_JAR, env) + populateHadoopClasspath(conf, env) + sys.env.get(ENV_DIST_CLASSPATH).foreach(addClasspathEntry(_, env)) } /** - * Adds the user jars which have local: URIs (or alternate names, such as APP_JAR) explicitly - * to the classpath. + * Returns a list of URIs representing the user classpath. + * + * @param conf Spark configuration. */ - private def addUserClasspath( - args: ClientArguments, - conf: SparkConf, - env: HashMap[String, String]): Unit = { - - // If `args` is not null, we are launching an AM container. - // Otherwise, we are launching executor containers. - val (mainJar, secondaryJars) = - if (args != null) { - (args.userJar, args.addJars) - } else { - (conf.get(CONF_SPARK_USER_JAR, null), conf.get(CONF_SPARK_YARN_SECONDARY_JARS, null)) - } + def getUserClasspath(conf: SparkConf): Array[URI] = { + getUserClasspath(conf.getOption(CONF_SPARK_USER_JAR), + conf.getOption(CONF_SPARK_YARN_SECONDARY_JARS)) + } - addFileToClasspath(mainJar, APP_JAR, env) - if (secondaryJars != null) { - secondaryJars.split(",").filter(_.nonEmpty).foreach { jar => - addFileToClasspath(jar, null, env) - } - } + private def getUserClasspath( + mainJar: Option[String], + secondaryJars: Option[String]): Array[URI] = { + val mainUri = mainJar.orElse(Some(APP_JAR)).map(new URI(_)) + val secondaryUris = secondaryJars.map(_.split(",")).toSeq.flatten.map(new URI(_)) + (mainUri ++ secondaryUris).toArray } /** @@ -846,27 +839,19 @@ object Client extends Logging { * * If not a "local:" file and no alternate name, the environment is not modified. * - * @param path Path to add to classpath (optional). + * @param uri URI to add to classpath (optional). * @param fileName Alternate name for the file (optional). * @param env Map holding the environment variables. */ private def addFileToClasspath( - path: String, + uri: URI, fileName: String, env: HashMap[String, String]): Unit = { - if (path != null) { - scala.util.control.Exception.ignoring(classOf[URISyntaxException]) { - val uri = new URI(path) - if (uri.getScheme == LOCAL_SCHEME) { - addClasspathEntry(uri.getPath, env) - return - } - } - } - if (fileName != null) { - addClasspathEntry( - YarnSparkHadoopUtil.expandEnvironment(Environment.PWD) + Path.SEPARATOR + fileName, env - ) + if (uri != null && uri.getScheme == LOCAL_SCHEME) { + addClasspathEntry(uri.getPath, env) + } else if (fileName != null) { + addClasspathEntry(buildPath( + YarnSparkHadoopUtil.expandEnvironment(Environment.PWD), fileName), env) } } @@ -962,4 +947,23 @@ object Client extends Logging { new Path(qualifiedURI) } + /** + * Whether to consider jars provided by the user to have precedence over the Spark jars when + * loading user classes. + */ + def isUserClassPathFirst(conf: SparkConf, isDriver: Boolean): Boolean = { + if (isDriver) { + conf.getBoolean("spark.driver.userClassPathFirst", false) + } else { + conf.getBoolean("spark.executor.userClassPathFirst", false) + } + } + + /** + * Joins all the path components using Path.SEPARATOR. + */ + def buildPath(components: String*): String = { + components.mkString(Path.SEPARATOR) + } + } diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala index 408cf09b9bdfa..6d5b8fda76ab8 100644 --- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala +++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala @@ -17,6 +17,7 @@ package org.apache.spark.deploy.yarn +import java.io.File import java.net.URI import java.nio.ByteBuffer @@ -57,7 +58,7 @@ class ExecutorRunnable( var nmClient: NMClient = _ val yarnConf: YarnConfiguration = new YarnConfiguration(conf) lazy val env = prepareEnvironment(container) - + def run = { logInfo("Starting Executor Container") nmClient = NMClient.createNMClient() @@ -128,14 +129,15 @@ class ExecutorRunnable( // Set the JVM memory val executorMemoryString = executorMemory + "m" - javaOpts += "-Xms" + executorMemoryString + " -Xmx" + executorMemoryString + " " + javaOpts += "-Xms" + executorMemoryString + javaOpts += "-Xmx" + executorMemoryString // Set extra Java options for the executor, if defined sys.props.get("spark.executor.extraJavaOptions").foreach { opts => - javaOpts += opts + javaOpts ++= Utils.splitCommandString(opts).map(YarnSparkHadoopUtil.escapeForShell) } sys.env.get("SPARK_JAVA_OPTS").foreach { opts => - javaOpts += opts + javaOpts ++= Utils.splitCommandString(opts).map(YarnSparkHadoopUtil.escapeForShell) } sys.props.get("spark.executor.extraLibraryPath").foreach { p => prefixEnv = Some(Utils.libraryPathEnvPrefix(Seq(p))) @@ -173,17 +175,27 @@ class ExecutorRunnable( // The options are based on // http://www.oracle.com/technetwork/java/gc-tuning-5-138395.html#0.0.0.%20When%20to%20Use // %20the%20Concurrent%20Low%20Pause%20Collector|outline - javaOpts += " -XX:+UseConcMarkSweepGC " - javaOpts += " -XX:+CMSIncrementalMode " - javaOpts += " -XX:+CMSIncrementalPacing " - javaOpts += " -XX:CMSIncrementalDutyCycleMin=0 " - javaOpts += " -XX:CMSIncrementalDutyCycle=10 " + javaOpts += "-XX:+UseConcMarkSweepGC" + javaOpts += "-XX:+CMSIncrementalMode" + javaOpts += "-XX:+CMSIncrementalPacing" + javaOpts += "-XX:CMSIncrementalDutyCycleMin=0" + javaOpts += "-XX:CMSIncrementalDutyCycle=10" } */ // For log4j configuration to reference javaOpts += ("-Dspark.yarn.app.container.log.dir=" + ApplicationConstants.LOG_DIR_EXPANSION_VAR) + val userClassPath = Client.getUserClasspath(sparkConf).flatMap { uri => + val absPath = + if (new File(uri.getPath()).isAbsolute()) { + uri.getPath() + } else { + Client.buildPath(Environment.PWD.$(), uri.getPath()) + } + Seq("--user-class-path", "file:" + absPath) + }.toSeq + val commands = prefixEnv ++ Seq( YarnSparkHadoopUtil.expandEnvironment(Environment.JAVA_HOME) + "/bin/java", "-server", @@ -195,11 +207,13 @@ class ExecutorRunnable( "-XX:OnOutOfMemoryError='kill %p'") ++ javaOpts ++ Seq("org.apache.spark.executor.CoarseGrainedExecutorBackend", - masterAddress.toString, - slaveId.toString, - hostname.toString, - executorCores.toString, - appId, + "--driver-url", masterAddress.toString, + "--executor-id", slaveId.toString, + "--hostname", hostname.toString, + "--cores", executorCores.toString, + "--app-id", appId) ++ + userClassPath ++ + Seq( "1>", ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stdout", "2>", ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stderr") diff --git a/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala b/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala index 690f927e938c3..f1b5aafac4066 100644 --- a/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala +++ b/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala @@ -20,6 +20,7 @@ package org.apache.spark.scheduler.cluster import scala.collection.mutable.ArrayBuffer import org.apache.hadoop.yarn.api.records.{ApplicationId, YarnApplicationState} +import org.apache.hadoop.yarn.exceptions.ApplicationNotFoundException import org.apache.spark.{SparkException, Logging, SparkContext} import org.apache.spark.deploy.yarn.{Client, ClientArguments} @@ -133,8 +134,14 @@ private[spark] class YarnClientSchedulerBackend( val t = new Thread { override def run() { while (!stopping) { - val report = client.getApplicationReport(appId) - val state = report.getYarnApplicationState() + var state: YarnApplicationState = null + try { + val report = client.getApplicationReport(appId) + state = report.getYarnApplicationState() + } catch { + case e: ApplicationNotFoundException => + state = YarnApplicationState.KILLED + } if (state == YarnApplicationState.FINISHED || state == YarnApplicationState.KILLED || state == YarnApplicationState.FAILED) { diff --git a/yarn/src/test/resources/log4j.properties b/yarn/src/test/resources/log4j.properties index 287c8e3563503..aab41fa49430f 100644 --- a/yarn/src/test/resources/log4j.properties +++ b/yarn/src/test/resources/log4j.properties @@ -16,7 +16,7 @@ # # Set everything to be logged to the file target/unit-tests.log -log4j.rootCategory=INFO, file +log4j.rootCategory=DEBUG, file log4j.appender.file=org.apache.log4j.FileAppender log4j.appender.file.append=true log4j.appender.file.file=target/unit-tests.log @@ -25,4 +25,4 @@ log4j.appender.file.layout.ConversionPattern=%d{yy/MM/dd HH:mm:ss.SSS} %t %p %c{ # Ignore messages below warning level from Jetty, because it's a bit verbose log4j.logger.org.eclipse.jetty=WARN -org.eclipse.jetty.LEVEL=WARN +log4j.logger.org.apache.hadoop=WARN diff --git a/yarn/src/test/scala/org/apache/spark/deploy/yarn/ClientSuite.scala b/yarn/src/test/scala/org/apache/spark/deploy/yarn/ClientSuite.scala index 2bb3dcffd61d9..f8f8129d220e4 100644 --- a/yarn/src/test/scala/org/apache/spark/deploy/yarn/ClientSuite.scala +++ b/yarn/src/test/scala/org/apache/spark/deploy/yarn/ClientSuite.scala @@ -82,6 +82,7 @@ class ClientSuite extends FunSuite with Matchers { test("Local jar URIs") { val conf = new Configuration() val sparkConf = new SparkConf().set(Client.CONF_SPARK_JAR, SPARK) + .set("spark.yarn.user.classpath.first", "true") val env = new MutableHashMap[String, String]() val args = new ClientArguments(Array("--jar", USER, "--addJars", ADDED), sparkConf) @@ -98,13 +99,10 @@ class ClientSuite extends FunSuite with Matchers { }) if (classOf[Environment].getMethods().exists(_.getName == "$$")) { cp should contain("{{PWD}}") - cp should contain(s"{{PWD}}${Path.SEPARATOR}*") } else if (Utils.isWindows) { cp should contain("%PWD%") - cp should contain(s"%PWD%${Path.SEPARATOR}*") } else { cp should contain(Environment.PWD.$()) - cp should contain(s"${Environment.PWD.$()}${File.separator}*") } cp should not contain (Client.SPARK_JAR) cp should not contain (Client.APP_JAR) @@ -117,7 +115,7 @@ class ClientSuite extends FunSuite with Matchers { val client = spy(new Client(args, conf, sparkConf)) doReturn(new Path("/")).when(client).copyFileToRemote(any(classOf[Path]), - any(classOf[Path]), anyShort(), anyBoolean()) + any(classOf[Path]), anyShort()) val tempDir = Utils.createTempDir() try { diff --git a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala index eda40efc4c77f..0e37276ba724b 100644 --- a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala +++ b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala @@ -17,27 +17,34 @@ package org.apache.spark.deploy.yarn -import java.io.File +import java.io.{File, FileOutputStream, OutputStreamWriter} +import java.util.Properties import java.util.concurrent.TimeUnit import scala.collection.JavaConversions._ import scala.collection.mutable -import com.google.common.base.Charsets +import com.google.common.base.Charsets.UTF_8 +import com.google.common.io.ByteStreams import com.google.common.io.Files import org.apache.hadoop.yarn.conf.YarnConfiguration import org.apache.hadoop.yarn.server.MiniYARNCluster import org.scalatest.{BeforeAndAfterAll, FunSuite, Matchers} -import org.apache.spark.{Logging, SparkConf, SparkContext, SparkException} +import org.apache.spark.{Logging, SparkConf, SparkContext, SparkException, TestUtils} import org.apache.spark.scheduler.cluster.ExecutorInfo import org.apache.spark.scheduler.{SparkListener, SparkListenerExecutorAdded} import org.apache.spark.util.Utils +/** + * Integration tests for YARN; these tests use a mini Yarn cluster to run Spark-on-YARN + * applications, and require the Spark assembly to be built before they can be successfully + * run. + */ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers with Logging { - // log4j configuration for the Yarn containers, so that their output is collected - // by Yarn instead of trying to overwrite unit-tests.log. + // log4j configuration for the YARN containers, so that their output is collected + // by YARN instead of trying to overwrite unit-tests.log. private val LOG4J_CONF = """ |log4j.rootCategory=DEBUG, console |log4j.appender.console=org.apache.log4j.ConsoleAppender @@ -52,13 +59,11 @@ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers wit | |from pyspark import SparkConf , SparkContext |if __name__ == "__main__": - | if len(sys.argv) != 3: - | print >> sys.stderr, "Usage: test.py [master] [result file]" + | if len(sys.argv) != 2: + | print >> sys.stderr, "Usage: test.py [result file]" | exit(-1) - | conf = SparkConf() - | conf.setMaster(sys.argv[1]).setAppName("python test in yarn cluster mode") - | sc = SparkContext(conf=conf) - | status = open(sys.argv[2],'w') + | sc = SparkContext(conf=SparkConf()) + | status = open(sys.argv[1],'w') | result = "failure" | rdd = sc.parallelize(range(10)) | cnt = rdd.count() @@ -72,21 +77,17 @@ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers wit private var yarnCluster: MiniYARNCluster = _ private var tempDir: File = _ private var fakeSparkJar: File = _ - private var oldConf: Map[String, String] = _ + private var logConfDir: File = _ override def beforeAll() { - tempDir = Utils.createTempDir() + super.beforeAll() - val logConfDir = new File(tempDir, "log4j") + tempDir = Utils.createTempDir() + logConfDir = new File(tempDir, "log4j") logConfDir.mkdir() val logConfFile = new File(logConfDir, "log4j.properties") - Files.write(LOG4J_CONF, logConfFile, Charsets.UTF_8) - - val childClasspath = logConfDir.getAbsolutePath() + File.pathSeparator + - sys.props("java.class.path") - - oldConf = sys.props.filter { case (k, v) => k.startsWith("spark.") }.toMap + Files.write(LOG4J_CONF, logConfFile, UTF_8) yarnCluster = new MiniYARNCluster(getClass().getName(), 1, 1, 1) yarnCluster.init(new YarnConfiguration()) @@ -117,99 +118,165 @@ class YarnClusterSuite extends FunSuite with BeforeAndAfterAll with Matchers wit } logInfo(s"RM address in configuration is ${config.get(YarnConfiguration.RM_ADDRESS)}") - config.foreach { e => - sys.props += ("spark.hadoop." + e.getKey() -> e.getValue()) - } fakeSparkJar = File.createTempFile("sparkJar", null, tempDir) - val sparkHome = sys.props.getOrElse("spark.test.home", fail("spark.test.home is not set!")) - sys.props += ("spark.yarn.appMasterEnv.SPARK_HOME" -> sparkHome) - sys.props += ("spark.executorEnv.SPARK_HOME" -> sparkHome) - sys.props += ("spark.yarn.jar" -> ("local:" + fakeSparkJar.getAbsolutePath())) - sys.props += ("spark.executor.instances" -> "1") - sys.props += ("spark.driver.extraClassPath" -> childClasspath) - sys.props += ("spark.executor.extraClassPath" -> childClasspath) - - super.beforeAll() } override def afterAll() { yarnCluster.stop() - sys.props.retain { case (k, v) => !k.startsWith("spark.") } - sys.props ++= oldConf super.afterAll() } test("run Spark in yarn-client mode") { - var result = File.createTempFile("result", null, tempDir) - YarnClusterDriver.main(Array("yarn-client", result.getAbsolutePath())) - checkResult(result) - - // verify log urls are present - YarnClusterDriver.listener.addedExecutorInfos.values.foreach { info => - assert(info.logUrlMap.nonEmpty) - } + testBasicYarnApp(true) } test("run Spark in yarn-cluster mode") { - val main = YarnClusterDriver.getClass.getName().stripSuffix("$") - var result = File.createTempFile("result", null, tempDir) - - val args = Array("--class", main, - "--jar", "file:" + fakeSparkJar.getAbsolutePath(), - "--arg", "yarn-cluster", - "--arg", result.getAbsolutePath(), - "--num-executors", "1") - Client.main(args) - checkResult(result) - - // verify log urls are present. - YarnClusterDriver.listener.addedExecutorInfos.values.foreach { info => - assert(info.logUrlMap.nonEmpty) - } + testBasicYarnApp(false) } test("run Spark in yarn-cluster mode unsuccessfully") { - val main = YarnClusterDriver.getClass.getName().stripSuffix("$") - - // Use only one argument so the driver will fail - val args = Array("--class", main, - "--jar", "file:" + fakeSparkJar.getAbsolutePath(), - "--arg", "yarn-cluster", - "--num-executors", "1") + // Don't provide arguments so the driver will fail. val exception = intercept[SparkException] { - Client.main(args) + runSpark(false, mainClassName(YarnClusterDriver.getClass)) + fail("Spark application should have failed.") } - assert(Utils.exceptionString(exception).contains("Application finished with failed status")) } test("run Python application in yarn-cluster mode") { val primaryPyFile = new File(tempDir, "test.py") - Files.write(TEST_PYFILE, primaryPyFile, Charsets.UTF_8) + Files.write(TEST_PYFILE, primaryPyFile, UTF_8) val pyFile = new File(tempDir, "test2.py") - Files.write(TEST_PYFILE, pyFile, Charsets.UTF_8) + Files.write(TEST_PYFILE, pyFile, UTF_8) var result = File.createTempFile("result", null, tempDir) - val args = Array("--class", "org.apache.spark.deploy.PythonRunner", - "--primary-py-file", primaryPyFile.getAbsolutePath(), - "--py-files", pyFile.getAbsolutePath(), - "--arg", "yarn-cluster", - "--arg", result.getAbsolutePath(), - "--name", "python test in yarn-cluster mode", - "--num-executors", "1") - Client.main(args) + // The sbt assembly does not include pyspark / py4j python dependencies, so we need to + // propagate SPARK_HOME so that those are added to PYTHONPATH. See PythonUtils.scala. + val sparkHome = sys.props("spark.test.home") + val extraConf = Map( + "spark.executorEnv.SPARK_HOME" -> sparkHome, + "spark.yarn.appMasterEnv.SPARK_HOME" -> sparkHome) + + runSpark(false, primaryPyFile.getAbsolutePath(), + sparkArgs = Seq("--py-files", pyFile.getAbsolutePath()), + appArgs = Seq(result.getAbsolutePath()), + extraConf = extraConf) checkResult(result) } + test("user class path first in client mode") { + testUseClassPathFirst(true) + } + + test("user class path first in cluster mode") { + testUseClassPathFirst(false) + } + + private def testBasicYarnApp(clientMode: Boolean): Unit = { + var result = File.createTempFile("result", null, tempDir) + runSpark(clientMode, mainClassName(YarnClusterDriver.getClass), + appArgs = Seq(result.getAbsolutePath())) + checkResult(result) + } + + private def testUseClassPathFirst(clientMode: Boolean): Unit = { + // Create a jar file that contains a different version of "test.resource". + val originalJar = TestUtils.createJarWithFiles(Map("test.resource" -> "ORIGINAL"), tempDir) + val userJar = TestUtils.createJarWithFiles(Map("test.resource" -> "OVERRIDDEN"), tempDir) + val driverResult = File.createTempFile("driver", null, tempDir) + val executorResult = File.createTempFile("executor", null, tempDir) + runSpark(clientMode, mainClassName(YarnClasspathTest.getClass), + appArgs = Seq(driverResult.getAbsolutePath(), executorResult.getAbsolutePath()), + extraClassPath = Seq(originalJar.getPath()), + extraJars = Seq("local:" + userJar.getPath()), + extraConf = Map( + "spark.driver.userClassPathFirst" -> "true", + "spark.executor.userClassPathFirst" -> "true")) + checkResult(driverResult, "OVERRIDDEN") + checkResult(executorResult, "OVERRIDDEN") + } + + private def runSpark( + clientMode: Boolean, + klass: String, + appArgs: Seq[String] = Nil, + sparkArgs: Seq[String] = Nil, + extraClassPath: Seq[String] = Nil, + extraJars: Seq[String] = Nil, + extraConf: Map[String, String] = Map()): Unit = { + val master = if (clientMode) "yarn-client" else "yarn-cluster" + val props = new Properties() + + props.setProperty("spark.yarn.jar", "local:" + fakeSparkJar.getAbsolutePath()) + + val childClasspath = logConfDir.getAbsolutePath() + + File.pathSeparator + + sys.props("java.class.path") + + File.pathSeparator + + extraClassPath.mkString(File.pathSeparator) + props.setProperty("spark.driver.extraClassPath", childClasspath) + props.setProperty("spark.executor.extraClassPath", childClasspath) + + // SPARK-4267: make sure java options are propagated correctly. + props.setProperty("spark.driver.extraJavaOptions", "-Dfoo=\"one two three\"") + props.setProperty("spark.executor.extraJavaOptions", "-Dfoo=\"one two three\"") + + yarnCluster.getConfig().foreach { e => + props.setProperty("spark.hadoop." + e.getKey(), e.getValue()) + } + + sys.props.foreach { case (k, v) => + if (k.startsWith("spark.")) { + props.setProperty(k, v) + } + } + + extraConf.foreach { case (k, v) => props.setProperty(k, v) } + + val propsFile = File.createTempFile("spark", ".properties", tempDir) + val writer = new OutputStreamWriter(new FileOutputStream(propsFile), UTF_8) + props.store(writer, "Spark properties.") + writer.close() + + val extraJarArgs = if (!extraJars.isEmpty()) Seq("--jars", extraJars.mkString(",")) else Nil + val mainArgs = + if (klass.endsWith(".py")) { + Seq(klass) + } else { + Seq("--class", klass, fakeSparkJar.getAbsolutePath()) + } + val argv = + Seq( + new File(sys.props("spark.test.home"), "bin/spark-submit").getAbsolutePath(), + "--master", master, + "--num-executors", "1", + "--properties-file", propsFile.getAbsolutePath()) ++ + extraJarArgs ++ + sparkArgs ++ + mainArgs ++ + appArgs + + Utils.executeAndGetOutput(argv, + extraEnvironment = Map("YARN_CONF_DIR" -> tempDir.getAbsolutePath())) + } + /** * This is a workaround for an issue with yarn-cluster mode: the Client class will not provide * any sort of error when the job process finishes successfully, but the job itself fails. So * the tests enforce that something is written to a file after everything is ok to indicate * that the job succeeded. */ - private def checkResult(result: File) = { - var resultString = Files.toString(result, Charsets.UTF_8) - resultString should be ("success") + private def checkResult(result: File): Unit = { + checkResult(result, "success") + } + + private def checkResult(result: File, expected: String): Unit = { + var resultString = Files.toString(result, UTF_8) + resultString should be (expected) + } + + private def mainClassName(klass: Class[_]): String = { + klass.getName().stripSuffix("$") } } @@ -227,22 +294,22 @@ private object YarnClusterDriver extends Logging with Matchers { val WAIT_TIMEOUT_MILLIS = 10000 var listener: SaveExecutorInfo = null - def main(args: Array[String]) = { - if (args.length != 2) { + def main(args: Array[String]): Unit = { + if (args.length != 1) { System.err.println( s""" |Invalid command line: ${args.mkString(" ")} | - |Usage: YarnClusterDriver [master] [result file] + |Usage: YarnClusterDriver [result file] """.stripMargin) System.exit(1) } listener = new SaveExecutorInfo - val sc = new SparkContext(new SparkConf().setMaster(args(0)) + val sc = new SparkContext(new SparkConf() .setAppName("yarn \"test app\" 'with quotes' and \\back\\slashes and $dollarSigns")) sc.addSparkListener(listener) - val status = new File(args(1)) + val status = new File(args(0)) var result = "failure" try { val data = sc.parallelize(1 to 4, 4).collect().toSet @@ -251,7 +318,48 @@ private object YarnClusterDriver extends Logging with Matchers { result = "success" } finally { sc.stop() - Files.write(result, status, Charsets.UTF_8) + Files.write(result, status, UTF_8) + } + + // verify log urls are present + listener.addedExecutorInfos.values.foreach { info => + assert(info.logUrlMap.nonEmpty) + } + } + +} + +private object YarnClasspathTest { + + def main(args: Array[String]): Unit = { + if (args.length != 2) { + System.err.println( + s""" + |Invalid command line: ${args.mkString(" ")} + | + |Usage: YarnClasspathTest [driver result file] [executor result file] + """.stripMargin) + System.exit(1) + } + + readResource(args(0)) + val sc = new SparkContext(new SparkConf()) + try { + sc.parallelize(Seq(1)).foreach { x => readResource(args(1)) } + } finally { + sc.stop() + } + } + + private def readResource(resultPath: String): Unit = { + var result = "failure" + try { + val ccl = Thread.currentThread().getContextClassLoader() + val resource = ccl.getResourceAsStream("test.resource") + val bytes = ByteStreams.toByteArray(resource) + result = new String(bytes, 0, bytes.length, UTF_8) + } finally { + Files.write(result, new File(resultPath), UTF_8) } }