[SPARK-4609] Job can not finish if there is one bad slave in clusters

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Component/s: Spark Core
• Assignee: Unassigned
• Reporter: Davies Liu
• Created: 26/Nov/14 00:45
• Updated: 27/Jun/15 05:42
• Resolved: 27/Jun/15 05:42

Description

If there is one bad machine in the cluster, the executor will keep die (such as out of space in the disk), some task may be scheduled to this machines multiple times, then the job will failed after several failures of one task.

14/11/26 00:34:57 INFO TaskSetManager: Starting task 39.0 in stage 3.0 (TID 1255, spark-worker-028.c.lofty-inn-754.internal, PROCESS_LOCAL, 5119 bytes)
14/11/26 00:34:57 WARN TaskSetManager: Lost task 39.0 in stage 3.0 (TID 1255, spark-worker-028.c.lofty-inn-754.internal): ExecutorLostFailure (executor 60 lost)
14/11/26 00:35:02 INFO TaskSetManager: Starting task 39.1 in stage 3.0 (TID 1256, spark-worker-028.c.lofty-inn-754.internal, PROCESS_LOCAL, 5119 bytes)
14/11/26 00:35:03 WARN TaskSetManager: Lost task 39.1 in stage 3.0 (TID 1256, spark-worker-028.c.lofty-inn-754.internal): ExecutorLostFailure (executor 61 lost)
14/11/26 00:35:08 INFO TaskSetManager: Starting task 39.2 in stage 3.0 (TID 1257, spark-worker-028.c.lofty-inn-754.internal, PROCESS_LOCAL, 5119 bytes)
14/11/26 00:35:08 WARN TaskSetManager: Lost task 39.2 in stage 3.0 (TID 1257, spark-worker-028.c.lofty-inn-754.internal): ExecutorLostFailure (executor 62 lost)
14/11/26 00:35:13 INFO TaskSetManager: Starting task 39.3 in stage 3.0 (TID 1258, spark-worker-028.c.lofty-inn-754.internal, PROCESS_LOCAL, 5119 bytes)
14/11/26 00:35:14 WARN TaskSetManager: Lost task 39.3 in stage 3.0 (TID 1258, spark-worker-028.c.lofty-inn-754.internal): ExecutorLostFailure (executor 63 lost)
org.apache.spark.SparkException: Job aborted due to stage failure: Task 39 in stage 3.0 failed 4 times, most recent failure: Lost task 39.3 in stage 3.0 (TID 1258, spark-worker-028.c.lofty-inn-754.internal): ExecutorLostFailure (executor 63 lost)
Driver stacktrace:
	at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1207)
	at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1196)
	at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1195)
	at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
	at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
	at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1195)
	at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:697)
	at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:697)
	at scala.Option.foreach(Option.scala:236)
	at org.apache.spark.scheduler.DAGScheduler.handleTaskSetFailed(DAGScheduler.scala:697)
	at org.apache.spark.scheduler.DAGSchedulerEventProcessActor$$anonfun$receive$2.applyOrElse(DAGScheduler.scala:1413)
	at akka.actor.Actor$class.aroundReceive(Actor.scala:465)
	at org.apache.spark.scheduler.DAGSchedulerEventProcessActor.aroundReceive(DAGScheduler.scala:1368)
	at akka.actor.ActorCell.receiveMessage(ActorCell.scala:516)
	at akka.actor.ActorCell.invoke(ActorCell.scala:487)
	at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:238)
	at akka.dispatch.Mailbox.run(Mailbox.scala:220)
	at akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:393)
	at scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)
	at scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)
	at scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)
	at scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)

The task should not be scheduled to a machines for more than one times. Also, if one machine failed with executor lost, it should be put in black list for some time, then try again.

cc kayousterhout matei

Issue Links

• duplicates: SPARK-6183 Skip bad workers when re-launching executors
• is related to: SPARK-6183 Skip bad workers when re-launching executors
• links to: [Github] Pull Request #3541 (davies)