[jira] [Commented] (SPARK-28917) Jobs can hang because of race of RDD.dependencies

["Imran Rashid (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-28917?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16918764#comment-16918764 ] 

Imran Rashid commented on SPARK-28917:
--------------------------------------

[~markhamstra] [~jiangxb1987] [~tgraves] [~Ngone51] would appreciate your thoughts on this.  I think the bug I've described above is pretty clear.  However, the part which I'm wondering about a bit more is whether there is more mutability in RDD that could cause problems.

For the case I have of this, I only know for sure that the user is calling {{rdd.cache()}} in another thread.  But I can't see how that would leave to the symptoms I describe above.  I don't know that they are doing anything in ther user thread which would touch {{rdd.dependencies}}, but I also don't have full visibility into everything they are doing, so this still seems like the best explanation to me.

> Jobs can hang because of race of RDD.dependencies
> -------------------------------------------------
>
>                 Key: SPARK-28917
>                 URL: https://issues.apache.org/jira/browse/SPARK-28917
>             Project: Spark
>          Issue Type: Bug
>          Components: Scheduler
>    Affects Versions: 2.3.3, 2.4.3
>            Reporter: Imran Rashid
>            Priority: Major
>
> {{RDD.dependencies}} stores the precomputed cache value, but it is not thread-safe.  This can lead to a race where the value gets overwritten, but the DAGScheduler gets stuck in an inconsistent state.  In particular, this can happen when there is a race between the DAGScheduler event loop, and another thread (eg. a user thread, if there is multi-threaded job submission).
> First, a job is submitted by the user, which then computes the result Stage and its parents:
> https://github.com/apache/spark/blob/24655583f1cb5dae2e80bb572604fb4a9761ec07/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala#L983
> Which eventually makes a call to {{rdd.dependencies}}:
> https://github.com/apache/spark/blob/24655583f1cb5dae2e80bb572604fb4a9761ec07/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala#L519
> At the same time, the user could also touch {{rdd.dependencies}} in another thread, which could overwrite the stored value because of the race.
> Then the DAGScheduler checks the dependencies *again* later on in the job submission, via {{getMissingParentStages}}
> https://github.com/apache/spark/blob/24655583f1cb5dae2e80bb572604fb4a9761ec07/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala#L1025
> Because it will find new dependencies, it will create entirely different stages.  Now the job has some orphaned stages which will never run.
> The symptoms of this are seeing disjoint sets of stages in the "Parents of final stage" and the "Missing parents" messages on job submission, as well as seeing repeated messages "Registered RDD X" for the same RDD id.  eg:
> {noformat}
> [INFO] 2019-08-15 23:22:31,570 org.apache.spark.SparkContext logInfo - Starting job: count at XXX.scala:462
> ...
> [INFO] 2019-08-15 23:22:31,573 org.apache.spark.scheduler.DAGScheduler logInfo - Registering RDD 14 (repartition at XXX.scala:421)
> ...
> ...
> [INFO] 2019-08-15 23:22:31,582 org.apache.spark.scheduler.DAGScheduler logInfo - Got job 1 (count at XXX.scala:462) with 40 output partitions
> [INFO] 2019-08-15 23:22:31,582 org.apache.spark.scheduler.DAGScheduler logInfo - Final stage: ResultStage 5 (count at XXX.scala:462)
> [INFO] 2019-08-15 23:22:31,582 org.apache.spark.scheduler.DAGScheduler logInfo - Parents of final stage: List(ShuffleMapStage 4)
> [INFO] 2019-08-15 23:22:31,599 org.apache.spark.scheduler.DAGScheduler logInfo - Registering RDD 14 (repartition at XXX.scala:421)
> [INFO] 2019-08-15 23:22:31,599 org.apache.spark.scheduler.DAGScheduler logInfo - Missing parents: List(ShuffleMapStage 6)
> {noformat}
> Note that there is a similar issue w/ {{rdd.partitions}}. I don't see a way it could mess up the scheduler (seems its only used for {{rdd.partitions.length}}).  There is also an issue that {{rdd.storageLevel}} is read and cached in the scheduler, but it could be modified simultaneously by the user in another thread.   Similarly, I can't see a way it could effect the scheduler.
> *WORKAROUND*:
> (a) call {{rdd.dependencies}} while you know that RDD is only getting touched by one thread (eg. in the thread that created it, or before you submit multiple jobs touching that RDD from other threads). Then that value will get cached.
> (b) don't submit jobs from multiple threads.



--
This message was sent by Atlassian Jira
(v8.3.2#803003)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org