Bump BWC Version to 2.18 and Fix Bugs (#1311) (#1313)

This PR includes the following updates and bug fixes:

* Bump BWC Version to 2.18: Updated BWC version to 2.18 since the 2.17 branch has been cut.
* Fix Confidence Value Exceeding 1 in RCF: Addressed a bug in RCF where the confidence value could exceed 1. Implemented a check to cap the confidence value at 1, preventing invalid confidence scores.
* Correct Parameter Assignment in GetAnomalyDetectorTransportAction: Fixed an issue where parameter assignments within a method did not affect external variables due to Java's pass-by-value nature.
* Fixed a bug in ResultProcessor where we were supposed to check whether the number of sent messages equals the number of received messages before starting imputation. However, the sent message count was mistakenly based on the number of pages rather than the actual number of messages.
* Fixed a bug where we mistakenly used the total reserved memory bytes as the memory size per entity in PriorityCache.

Testing done:
* added test cases for the buggy scenarios
* manual e2e testing

Signed-off-by: Kaituo Li <kaituo@amazon.com>

build.gradle

@@ -696,9 +696,8 @@ List<String> jacocoExclusions = [
 
         // TODO: add test coverage (kaituo)
         'org.opensearch.forecast.*',
+        'org.opensearch.ad.transport.ADHCImputeNodeResponse',
         'org.opensearch.ad.transport.GetAnomalyDetectorTransportAction',
-        'org.opensearch.ad.ml.ADColdStart',
-        'org.opensearch.ad.transport.ADHCImputeNodesResponse',
         'org.opensearch.timeseries.transport.BooleanNodeResponse',
         'org.opensearch.timeseries.ml.TimeSeriesSingleStreamCheckpointDao',
         'org.opensearch.timeseries.transport.JobRequest',
@@ -713,7 +712,6 @@ List<String> jacocoExclusions = [
         'org.opensearch.timeseries.transport.ResultBulkTransportAction',
         'org.opensearch.timeseries.transport.handler.IndexMemoryPressureAwareResultHandler',
         'org.opensearch.timeseries.transport.handler.ResultIndexingHandler',
-        'org.opensearch.ad.transport.ADHCImputeNodeResponse',
         'org.opensearch.timeseries.ml.Sample',
         'org.opensearch.timeseries.ratelimit.FeatureRequest',
         'org.opensearch.ad.transport.ADHCImputeNodeRequest',

src/main/java/org/opensearch/ad/model/AnomalyResult.java

@@ -446,7 +446,7 @@ public static AnomalyResult fromRawTRCFResult(
             taskId,
             rcfScore,
             Math.max(0, grade),
-            confidence,
+            Math.min(1, confidence),
             featureData,
             dataStartTime,
             dataEndTime,

src/main/java/org/opensearch/ad/transport/GetAnomalyDetectorTransportAction.java

@@ -86,10 +86,11 @@ public GetAnomalyDetectorTransportAction(
     }
 
     @Override
-    protected void fillInHistoricalTaskforBwc(Map<String, ADTask> tasks, Optional<ADTask> historicalAdTask) {
+    protected Optional<ADTask> fillInHistoricalTaskforBwc(Map<String, ADTask> tasks) {
         if (tasks.containsKey(ADTaskType.HISTORICAL.name())) {
-            historicalAdTask = Optional.ofNullable(tasks.get(ADTaskType.HISTORICAL.name()));
+            return Optional.ofNullable(tasks.get(ADTaskType.HISTORICAL.name()));
         }
+        return Optional.empty();
     }
 
     @Override

src/main/java/org/opensearch/forecast/model/ForecastResult.java

@@ -188,7 +188,7 @@ public static List<ForecastResult> fromRawRCFCasterResult(
                 new ForecastResult(
                     forecasterId,
                     taskId,
-                    dataQuality,
+                    Math.min(1, dataQuality),
                     featureData,
                     dataStartTime,
                     dataEndTime,
@@ -218,7 +218,7 @@ public static List<ForecastResult> fromRawRCFCasterResult(
                         new ForecastResult(
                             forecasterId,
                             taskId,
-                            dataQuality,
+                            Math.min(1, dataQuality),
                             null,
                             dataStartTime,
                             dataEndTime,

src/main/java/org/opensearch/timeseries/caching/PriorityCache.java

@@ -463,15 +463,16 @@ public Pair<List<Entity>, List<Entity>> selectUpdateCandidate(Collection<Entity>
         return Pair.of(hotEntities, coldEntities);
     }
 
-    private CacheBufferType computeBufferIfAbsent(Config config, String configId) {
+    public CacheBufferType computeBufferIfAbsent(Config config, String configId) {
         CacheBufferType buffer = activeEnities.get(configId);
         if (buffer == null) {
-            long requiredBytes = getRequiredMemory(config, config.isHighCardinality() ? hcDedicatedCacheSize : 1);
+            long bytesPerEntityModel = getRequiredMemoryPerEntity(config, memoryTracker, numberOfTrees);
+            long requiredBytes = bytesPerEntityModel * (config.isHighCardinality() ? hcDedicatedCacheSize : 1);
             if (memoryTracker.canAllocateReserved(requiredBytes)) {
                 memoryTracker.consumeMemory(requiredBytes, true, origin);
                 buffer = createEmptyCacheBuffer(
                     config,
-                    requiredBytes,
+                    bytesPerEntityModel,
                     priorityTrackerMap
                         .getOrDefault(
                             configId,
@@ -496,16 +497,6 @@ private CacheBufferType computeBufferIfAbsent(Config config, String configId) {
         return buffer;
     }
 
-    /**
-     *
-     * @param config Detector config accessor
-     * @param numberOfEntity number of entities
-     * @return Memory in bytes required for hosting numberOfEntity entities
-     */
-    private long getRequiredMemory(Config config, int numberOfEntity) {
-        return numberOfEntity * getRequiredMemoryPerEntity(config, memoryTracker, numberOfTrees);
-    }
-
     /**
      * Whether the candidate entity can replace any entity in the shared cache.
      * We can have race conditions when multiple threads try to evaluate this

src/main/java/org/opensearch/timeseries/transport/BaseGetConfigTransportAction.java

@@ -180,7 +180,7 @@ public void doExecute(Task task, ActionRequest request, ActionListener<GetConfig
         }
     }
 
-    protected void getConfigAndJob(
+    public void getConfigAndJob(
         String configID,
         boolean returnJob,
         boolean returnTask,
@@ -251,7 +251,7 @@ public void getExecute(GetConfigRequest request, ActionListener<GetConfigRespons
                             } else {
                                 // AD needs to provides custom behavior for bwc, while forecasting can inherit
                                 // the empty implementation
-                                fillInHistoricalTaskforBwc(tasks, historicalTask);
+                                historicalTask = fillInHistoricalTaskforBwc(tasks);
                             }
                         }
                         getConfigAndJob(configID, returnJob, returnTask, realtimeTask, historicalTask, listener);
@@ -357,7 +357,9 @@ public void onFailure(Exception e) {
         };
     }
 
-    protected void fillInHistoricalTaskforBwc(Map<String, TaskClass> tasks, Optional<TaskClass> historicalAdTask) {}
+    protected Optional<TaskClass> fillInHistoricalTaskforBwc(Map<String, TaskClass> tasks) {
+        return Optional.empty();
+    }
 
     protected void getExecuteProfile(
         GetConfigRequest request,

src/main/java/org/opensearch/timeseries/transport/ResultProcessor.java

@@ -229,8 +229,6 @@ public void onResponse(CompositeRetriever.Page entityFeatures) {
                 pageIterator.next(this);
             }
             if (entityFeatures != null && false == entityFeatures.isEmpty()) {
-                sentOutPages.incrementAndGet();
-
                 LOG
                     .info(
                         "Sending an HC request to process data from timestamp {} to {} for config {}",
@@ -285,6 +283,7 @@ public void onResponse(CompositeRetriever.Page entityFeatures) {
                         final AtomicReference<Exception> failure = new AtomicReference<>();
 
                         node2Entities.stream().forEach(nodeEntity -> {
+                            sentOutPages.incrementAndGet();
                             DiscoveryNode node = nodeEntity.getKey();
                             transportService
                                 .sendRequest(
@@ -370,7 +369,15 @@ public void run() {
                             cancellable.get().cancel();
                         }
                     } else if (Instant.now().toEpochMilli() >= timeoutMillis) {
-                        LOG.warn("Scheduled impute HC task is cancelled due to timeout");
+                        LOG
+                            .warn(
+                                "Scheduled impute HC task is cancelled due to timeout, current epoch {}, timeout epoch {}, dataEndTime {}, sent out {}, receive {}",
+                                Instant.now().toEpochMilli(),
+                                timeoutMillis,
+                                dataEndTime,
+                                sentOutPages.get(),
+                                receivedPages.get()
+                            );
                         if (cancellable != null) {
                             cancellable.get().cancel();
                         }

src/test/java/org/opensearch/ad/HistoricalAnalysisRestTestCase.java

@@ -266,6 +266,29 @@ protected List<Object> waitUntilTaskReachState(String detectorId, Set<String> ta
         return results;
     }
 
+    protected List<Object> waitUntilTaskReachNumberOfEntities(String detectorId, int categoricalValuesCount) throws InterruptedException {
+        List<Object> results = new ArrayList<>();
+        int i = 0;
+        ADTaskProfile adTaskProfile = null;
+        // Increase retryTimes if some task can't reach done state
+        while ((adTaskProfile == null
+            || adTaskProfile.getTotalEntitiesCount() == null
+            || adTaskProfile.getTotalEntitiesCount().intValue() != categoricalValuesCount) && i < MAX_RETRY_TIMES) {
+            try {
+                adTaskProfile = getADTaskProfile(detectorId);
+            } catch (Exception e) {
+                logger.error("failed to get ADTaskProfile", e);
+            } finally {
+                Thread.sleep(1000);
+            }
+            i++;
+        }
+        assertNotNull(adTaskProfile);
+        results.add(adTaskProfile);
+        results.add(i);
+        return results;
+    }
+
     protected List<Object> waitUntilEntityCountAvailable(String detectorId) throws InterruptedException {
         List<Object> results = new ArrayList<>();
         int i = 0;

src/test/java/org/opensearch/ad/caching/PriorityCacheTests.java

@@ -26,6 +26,7 @@
 import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;
 
+import java.io.IOException;
 import java.time.Duration;
 import java.time.Instant;
 import java.util.ArrayDeque;
@@ -62,6 +63,7 @@
 import org.opensearch.threadpool.Scheduler.ScheduledCancellable;
 import org.opensearch.threadpool.ThreadPool;
 import org.opensearch.timeseries.MemoryTracker;
+import org.opensearch.timeseries.TestHelpers;
 import org.opensearch.timeseries.breaker.CircuitBreakerService;
 import org.opensearch.timeseries.common.exception.LimitExceededException;
 import org.opensearch.timeseries.common.exception.TimeSeriesException;
@@ -788,4 +790,48 @@ public void testGetTotalUpdates_orElseGetBranchWithNullSamples() {
         // Assert that the result is 0L
         assertEquals(0L, result);
     }
+
+    public void testAllocation() throws IOException {
+        JvmService jvmService = mock(JvmService.class);
+        JvmInfo info = mock(JvmInfo.class);
+
+        when(jvmService.info()).thenReturn(info);
+
+        Mem mem = mock(Mem.class);
+        when(mem.getHeapMax()).thenReturn(new ByteSizeValue(800_000_000L));
+        when(info.getMem()).thenReturn(mem);
+
+        CircuitBreakerService circuitBreaker = mock(CircuitBreakerService.class);
+        when(circuitBreaker.isOpen()).thenReturn(false);
+        MemoryTracker tracker = new MemoryTracker(jvmService, 0.1, clusterService, circuitBreaker);
+
+        dedicatedCacheSize = 10;
+        ADPriorityCache cache = new ADPriorityCache(
+            checkpoint,
+            dedicatedCacheSize,
+            AnomalyDetectorSettings.AD_CHECKPOINT_TTL,
+            AnomalyDetectorSettings.MAX_INACTIVE_ENTITIES,
+            tracker,
+            TimeSeriesSettings.NUM_TREES,
+            clock,
+            clusterService,
+            TimeSeriesSettings.HOURLY_MAINTENANCE,
+            threadPool,
+            TimeSeriesSettings.MAINTENANCE_FREQ_CONSTANT,
+            Settings.EMPTY,
+            AnomalyDetectorSettings.AD_CHECKPOINT_SAVING_FREQ,
+            checkpointWriteQueue,
+            checkpointMaintainQueue
+        );
+
+        List<String> categoryFields = Arrays.asList("category_field_1", "category_field_2");
+        AnomalyDetector anomalyDetector = TestHelpers.AnomalyDetectorBuilder
+            .newInstance(5)
+            .setShingleSize(8)
+            .setCategoryFields(categoryFields)
+            .build();
+        ADCacheBuffer buffer = cache.computeBufferIfAbsent(anomalyDetector, anomalyDetector.getId());
+        assertEquals(698336, buffer.getMemoryConsumptionPerModel());
+        assertEquals(698336 * dedicatedCacheSize, tracker.getTotalMemoryBytes());
+    }
 }

src/test/java/org/opensearch/ad/e2e/AbstractRuleTestCase.java

@@ -47,7 +47,7 @@ protected TrainResult ingestTrainDataAndCreateDetector(
         int trainTestSplit,
         boolean useDateNanos
     ) throws Exception {
-        return ingestTrainDataAndCreateDetector(datasetName, intervalMinutes, numberOfEntities, trainTestSplit, useDateNanos, -1);
+        return ingestTrainDataAndCreateDetector(datasetName, intervalMinutes, numberOfEntities, trainTestSplit, useDateNanos, -1, true);
     }
 
     protected TrainResult ingestTrainDataAndCreateDetector(
@@ -56,7 +56,8 @@ protected TrainResult ingestTrainDataAndCreateDetector(
         int numberOfEntities,
         int trainTestSplit,
         boolean useDateNanos,
-        int ingestDataSize
+        int ingestDataSize,
+        boolean relative
     ) throws Exception {
         TrainResult trainResult = ingestTrainData(
             datasetName,
@@ -67,15 +68,30 @@ protected TrainResult ingestTrainDataAndCreateDetector(
             ingestDataSize
         );
 
-        String detector = genDetector(datasetName, intervalMinutes, trainTestSplit, trainResult);
+        String detector = genDetector(datasetName, intervalMinutes, trainTestSplit, trainResult, relative);
         String detectorId = createDetector(client(), detector);
         LOG.info("Created detector {}", detectorId);
         trainResult.detectorId = detectorId;
 
         return trainResult;
     }
 
-    protected String genDetector(String datasetName, int intervalMinutes, int trainTestSplit, TrainResult trainResult) {
+    protected String genDetector(String datasetName, int intervalMinutes, int trainTestSplit, TrainResult trainResult, boolean relative) {
+        // Determine threshold types and values based on the 'relative' parameter
+        String thresholdType1;
+        String thresholdType2;
+        double value;
+        if (relative) {
+            thresholdType1 = "actual_over_expected_ratio";
+            thresholdType2 = "expected_over_actual_ratio";
+            value = 0.3;
+        } else {
+            thresholdType1 = "actual_over_expected_margin";
+            thresholdType2 = "expected_over_actual_margin";
+            value = 3000.0;
+        }
+
+        // Generate the detector JSON string with the appropriate threshold types and values
         String detector = String
             .format(
                 Locale.ROOT,
@@ -87,15 +103,20 @@ protected String genDetector(String datasetName, int intervalMinutes, int trainT
                     + "\"window_delay\": { \"period\": {\"interval\": %d, \"unit\": \"MINUTES\"}},"
                     + "\"history\": %d,"
                     + "\"schema_version\": 0,"
-                    + "\"rules\": [{\"action\": \"ignore_anomaly\", \"conditions\": [{\"feature_name\": \"feature 1\", \"threshold_type\": \"actual_over_expected_ratio\", \"operator\": \"lte\", \"value\": 0.3}, "
-                    + "{\"feature_name\": \"feature 1\", \"threshold_type\": \"expected_over_actual_ratio\", \"operator\": \"lte\", \"value\": 0.3}"
+                    + "\"rules\": [{\"action\": \"ignore_anomaly\", \"conditions\": ["
+                    + "{ \"feature_name\": \"feature 1\", \"threshold_type\": \"%s\", \"operator\": \"lte\", \"value\": %f }, "
+                    + "{ \"feature_name\": \"feature 1\", \"threshold_type\": \"%s\", \"operator\": \"lte\", \"value\": %f }"
                     + "]}]"
                     + "}",
                 datasetName,
                 intervalMinutes,
                 categoricalField,
                 trainResult.windowDelay.toMinutes(),
-                trainTestSplit - 1
+                trainTestSplit - 1,
+                thresholdType1,
+                value,
+                thresholdType2,
+                value
             );
         return detector;
     }

src/test/java/org/opensearch/ad/e2e/PreviewRuleIT.java

@@ -32,7 +32,7 @@ public void testRule() throws Exception {
                 (trainTestSplit + 1) * numberOfEntities
             );
 
-            String detector = genDetector(datasetName, intervalMinutes, trainTestSplit, trainResult);
+            String detector = genDetector(datasetName, intervalMinutes, trainTestSplit, trainResult, true);
             Map<String, Object> result = preview(detector, trainResult.firstDataTime, trainResult.finalDataTime, client());
             List<Object> results = (List<Object>) XContentMapValues.extractValue(result, "anomaly_result");
             assertTrue(results.size() > 100);

src/test/java/org/opensearch/ad/e2e/RealTimeRuleIT.java

@@ -15,7 +15,7 @@
 import com.google.gson.JsonObject;
 
 public class RealTimeRuleIT extends AbstractRuleTestCase {
-    public void testRuleWithDateNanos() throws Exception {
+    private void template(boolean reltive) throws Exception {
         // TODO: this test case will run for a much longer time and timeout with security enabled
         if (!isHttps()) {
             disableResourceNotFoundFaultTolerence();
@@ -32,7 +32,8 @@ public void testRuleWithDateNanos() throws Exception {
                 trainTestSplit,
                 true,
                 // ingest just enough for finish the test
-                (trainTestSplit + 1) * numberOfEntities
+                (trainTestSplit + 1) * numberOfEntities,
+                reltive
             );
 
             startRealTimeDetector(trainResult, numberOfEntities, intervalMinutes, false);
@@ -90,4 +91,12 @@ public void testRuleWithDateNanos() throws Exception {
             }
         }
     }
+
+    public void testRelativeRule() throws Exception {
+        template(true);
+    }
+
+    public void testAbsoluateRule() throws Exception {
+        template(false);
+    }
 }

src/test/java/org/opensearch/ad/ml/EntityColdStarterTests.java

@@ -1275,4 +1275,25 @@ public void testNotEnoughTrainingData() throws IOException, InterruptedException
         checkSemaphoreRelease();
         assertTrue(modelState.getModel().isEmpty());
     }
+
+    public void testTrainModelFromInvalidSamplesNotEnoughSamples() {
+        Deque<Sample> samples = new ArrayDeque<>();
+        // we have at least numMinSamples samples before executing the null check of trainModelFromDataSegments
+        for (int i = 0; i < numMinSamples; i++) {
+            samples.add(new Sample());
+        }
+
+        modelState = new ModelState<ThresholdedRandomCutForest>(
+            null,
+            modelId,
+            detectorId,
+            ModelManager.ModelType.TRCF.getName(),
+            clock,
+            priority,
+            Optional.of(entity),
+            samples
+        );
+        entityColdStarter.trainModelFromExistingSamples(modelState, detector, "123");
+        assertTrue(modelState.getModel().isEmpty());
+    }
 }