[SPARK-1210] Prevent ContextClassLoader of Actor from becoming ClassLoader of Executor

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.9.0
• Fix Version/s: 1.0.0
• Component/s: Spark Core
• Assignee: Takuya Ueshin
• Reporter: Takuya Ueshin
• Created: 10/Mar/14 01:34
• Updated: 28/Mar/14 10:32
• Resolved: 27/Mar/14 22:18

Description

Constructor of org.apache.spark.executor.Executor should not set context class loader of current thread, which is backend Actor's thread.

Run the following code in local-mode REPL.

scala> case class Foo(i: Int)
scala> val ret = sc.parallelize((1 to 100).map(Foo), 10).collect

This causes errors as follows:

ERROR actor.OneForOneStrategy: [L$line5.$read$$iwC$$iwC$$iwC$$iwC$Foo;
java.lang.ArrayStoreException: [L$line5.$read$$iwC$$iwC$$iwC$$iwC$Foo;
at scala.runtime.ScalaRunTime$.array_update(ScalaRunTime.scala:88)
at org.apache.spark.SparkContext$$anonfun$runJob$3.apply(SparkContext.scala:870)
at org.apache.spark.SparkContext$$anonfun$runJob$3.apply(SparkContext.scala:870)
at org.apache.spark.scheduler.JobWaiter.taskSucceeded(JobWaiter.scala:56)
at org.apache.spark.scheduler.DAGScheduler.handleTaskCompletion(DAGScheduler.scala:859)
at org.apache.spark.scheduler.DAGScheduler.processEvent(DAGScheduler.scala:616)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$start$1$$anon$2$$anonfun$receive$1.applyOrElse(DAGScheduler.scala:207)
at akka.actor.ActorCell.receiveMessage(ActorCell.scala:498)
at akka.actor.ActorCell.invoke(ActorCell.scala:456)
at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:237)
at akka.dispatch.Mailbox.run(Mailbox.scala:219)
at akka.dispatch.ForkJoinExecutorConfigurator$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 the class loaders to deserialize result Foo instances might be different from backend Actor's, and the Actor's class loader should be the same as Driver's.

See PR: https://github.com/apache/spark/pull/15

Issue Links

• is duplicated by: SPARK-1152 ArrayStoreException on mapping RDD on cluster
• is related to: SPARK-1346 Backport SPARK-1210 into 0.9 branch