[SPARK-1749] DAGScheduler supervisor strategy broken with Mesos

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 1.0.0
• Fix Version/s: 1.0.1, 1.1.0
• Component/s: Mesos, Spark Core
• Labels: mesos scheduler scheduling
• Target Version/s: 1.0.1, 1.1.0
• Assignee: Mark Hamstra
• Reporter: Bouke van der Bijl
• Created: 07/May/14 19:19
• Updated: 26/Jun/14 03:58
• Resolved: 26/Jun/14 03:58

Description

Any bad Python code will trigger this bug, for example `sc.parallelize(range(100)).map(lambda n: undefined_variable * 2).collect()` will cause a `undefined_variable isn't defined`, which will cause spark to try to kill the task, resulting in the following stacktrace:

java.lang.UnsupportedOperationException
at org.apache.spark.scheduler.SchedulerBackend$class.killTask(SchedulerBackend.scala:32)
at org.apache.spark.scheduler.cluster.mesos.MesosSchedulerBackend.killTask(MesosSchedulerBackend.scala:41)
at org.apache.spark.scheduler.TaskSchedulerImpl$$anonfun$cancelTasks$3$$anonfun$apply$1.apply$mcVJ$sp(TaskSchedulerImpl.scala:184)
at org.apache.spark.scheduler.TaskSchedulerImpl$$anonfun$cancelTasks$3$$anonfun$apply$1.apply(TaskSchedulerImpl.scala:182)
at org.apache.spark.scheduler.TaskSchedulerImpl$$anonfun$cancelTasks$3$$anonfun$apply$1.apply(TaskSchedulerImpl.scala:182)
at scala.collection.mutable.HashSet.foreach(HashSet.scala:79)
at org.apache.spark.scheduler.TaskSchedulerImpl$$anonfun$cancelTasks$3.apply(TaskSchedulerImpl.scala:182)
at org.apache.spark.scheduler.TaskSchedulerImpl$$anonfun$cancelTasks$3.apply(TaskSchedulerImpl.scala:175)
at scala.Option.foreach(Option.scala:236)
at org.apache.spark.scheduler.TaskSchedulerImpl.cancelTasks(TaskSchedulerImpl.scala:175)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages$1.apply$mcVI$sp(DAGScheduler.scala:1058)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages$1.apply(DAGScheduler.scala:1045)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages$1.apply(DAGScheduler.scala:1045)
at scala.collection.mutable.HashSet.foreach(HashSet.scala:79)
at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1045)
at org.apache.spark.scheduler.DAGScheduler.handleJobCancellation(DAGScheduler.scala:998)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$doCancelAllJobs$1.apply$mcVI$sp(DAGScheduler.scala:499)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$doCancelAllJobs$1.apply(DAGScheduler.scala:499)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$doCancelAllJobs$1.apply(DAGScheduler.scala:499)
at scala.collection.mutable.HashSet.foreach(HashSet.scala:79)
at org.apache.spark.scheduler.DAGScheduler.doCancelAllJobs(DAGScheduler.scala:499)
at org.apache.spark.scheduler.DAGSchedulerActorSupervisor$$anonfun$2.applyOrElse(DAGScheduler.scala:1151)
at org.apache.spark.scheduler.DAGSchedulerActorSupervisor$$anonfun$2.applyOrElse(DAGScheduler.scala:1147)
at akka.actor.SupervisorStrategy.handleFailure(FaultHandling.scala:295)
at akka.actor.dungeon.FaultHandling$class.handleFailure(FaultHandling.scala:253)
at akka.actor.ActorCell.handleFailure(ActorCell.scala:338)
at akka.actor.ActorCell.invokeAll$1(ActorCell.scala:423)
at akka.actor.ActorCell.systemInvoke(ActorCell.scala:447)
at akka.dispatch.Mailbox.processAllSystemMessages(Mailbox.scala:262)
at akka.dispatch.Mailbox.run(Mailbox.scala:218)
at akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:386)
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)

This is because killTask isn't implemented for the MesosSchedulerBackend. I assume this isn't pyspark-specific, as there will be other instances where you might want to kill the task