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 45f71d9927673..2758c26fa7ec6 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
@@ -482,22 +482,26 @@ private[spark] class Master(
 
     // First schedule drivers, they take strict precedence over applications
     // Randomization helps balance drivers
-    val shuffledAliveWorkers = Random.shuffle(workers).toArray
-    val aliveWorkerNum = shuffledAliveWorkers.size
-    var curPos = aliveWorkerNum - 1
-    for (driver <- List(waitingDrivers: _*)) {
-      val startFlag = curPos
-      curPos = (curPos + 1) % aliveWorkerNum
+    val shuffledWorkers = Random.shuffle(workers).toArray
+    val workerNum = shuffledWorkers.size
+    var curPos = 0
+    for (driver <- List(waitingDrivers: _*)) { // iterate over a copy of waitingDrivers
+      // For each waiting driver we pick a worker that has enough resources to launch it.
+      // The picking does in a round-robin fashion, starting from position behind last
+      // worker on which driver was just launched and ending with driver being launched
+      // or we have iterated over all workers.
+      val startPos = curPos
+      curPos = (curPos + 1) % workerNum
       var launched = false
-      while (curPos != startFlag && !launched) {
-        val worker = shuffledAliveWorkers(curPos)
+      while (curPos - 1 != startPos && !launched) {
+        val worker = shuffledWorkers(curPos)
         if (worker.state == WorkerState.ALIVE
           && worker.memoryFree >= driver.desc.mem && worker.coresFree >= driver.desc.cores) {
           launchDriver(worker, driver)
           waitingDrivers -= driver
           launched = true
         }
-        curPos = (curPos + 1) % aliveWorkerNum
+        curPos = (curPos + 1) % workerNum
       }
     }