[SPARK-29244] ArrayIndexOutOfBoundsException on TaskCompletionListener during releasing of memory blocks

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4.0
• Fix Version/s: 2.4.5, 3.0.0
• Component/s: Spark Core

Environment

Release label:emr-5.20.0
Hadoop distribution:Amazon 2.8.5
Applications:Livy 0.5.0, Spark 2.4.0

• Assignee: L. C. Hsieh
• Reporter: Viacheslav Tradunsky
• Created: 25/Sep/19 10:34
• Updated: 01/Oct/19 20:27
• Resolved: 01/Oct/19 18:51

Description

At the end of task completion an exception happened:

19/09/25 09:03:58 ERROR TaskContextImpl: Error in TaskCompletionListener19/09/25 09:03:58 ERROR TaskContextImpl: Error in TaskCompletionListenerjava.lang.ArrayIndexOutOfBoundsException: -3 at org.apache.spark.memory.TaskMemoryManager.freePage(TaskMemoryManager.java:333) at org.apache.spark.memory.MemoryConsumer.freePage(MemoryConsumer.java:130) at org.apache.spark.memory.MemoryConsumer.freeArray(MemoryConsumer.java:108) at org.apache.spark.unsafe.map.BytesToBytesMap.free(BytesToBytesMap.java:803) at org.apache.spark.sql.execution.UnsafeFixedWidthAggregationMap.free(UnsafeFixedWidthAggregationMap.java:225) at org.apache.spark.sql.execution.UnsafeFixedWidthAggregationMap.lambda$new$0(UnsafeFixedWidthAggregationMap.java:111) at org.apache.spark.TaskContextImpl$$anonfun$markTaskCompleted$1.apply(TaskContextImpl.scala:117) at org.apache.spark.TaskContextImpl$$anonfun$markTaskCompleted$1.apply(TaskContextImpl.scala:117) at org.apache.spark.TaskContextImpl$$anonfun$invokeListeners$1.apply(TaskContextImpl.scala:130) at org.apache.spark.TaskContextImpl$$anonfun$invokeListeners$1.apply(TaskContextImpl.scala:128) at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59) at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48) at org.apache.spark.TaskContextImpl.invokeListeners(TaskContextImpl.scala:128) at org.apache.spark.TaskContextImpl.markTaskCompleted(TaskContextImpl.scala:116) at org.apache.spark.scheduler.Task.run(Task.scala:131) at org.apache.spark.executor.Executor$TaskRunner$$anonfun$10.apply(Executor.scala:402) at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1360) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:408) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748)

 

Important to note, that before this one, there was OOM of allocating some pages. It looks like everything related to each other, but on OOM the whole flow goes abnormally, so no resources are fried correctly.

java.lang.NullPointerExceptionjava.lang.NullPointerException at org.apache.spark.util.collection.unsafe.sort.UnsafeInMemorySorter.getMemoryUsage(UnsafeInMemorySorter.java:208) at org.apache.spark.util.collection.unsafe.sort.UnsafeExternalSorter.getMemoryUsage(UnsafeExternalSorter.java:249) at org.apache.spark.util.collection.unsafe.sort.UnsafeExternalSorter.updatePeakMemoryUsed(UnsafeExternalSorter.java:253) at org.apache.spark.util.collection.unsafe.sort.UnsafeExternalSorter.freeMemory(UnsafeExternalSorter.java:296) at org.apache.spark.util.collection.unsafe.sort.UnsafeExternalSorter.cleanupResources(UnsafeExternalSorter.java:328) at org.apache.spark.util.collection.unsafe.sort.UnsafeExternalSorter.lambda$new$0(UnsafeExternalSorter.java:178) at org.apache.spark.TaskContextImpl$$anonfun$markTaskCompleted$1.apply(TaskContextImpl.scala:117) at org.apache.spark.TaskContextImpl$$anonfun$markTaskCompleted$1.apply(TaskContextImpl.scala:117) at org.apache.spark.TaskContextImpl$$anonfun$invokeListeners$1.apply(TaskContextImpl.scala:130) at org.apache.spark.TaskContextImpl$$anonfun$invokeListeners$1.apply(TaskContextImpl.scala:128) at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59) at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48) at org.apache.spark.TaskContextImpl.invokeListeners(TaskContextImpl.scala:128) at org.apache.spark.TaskContextImpl.markTaskCompleted(TaskContextImpl.scala:116) at org.apache.spark.scheduler.Task.run(Task.scala:131) at org.apache.spark.executor.Executor$TaskRunner$$anonfun$10.apply(Executor.scala:402) at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1360) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:408) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748)

 

This is must be something with job planning, but taking so many exceptions into account doesn't make things easier. Would be happy to provide more details.

Issue Links

• links to: GitHub Pull Request #25953