[jira] [Resolved] (KUDU-3099) KuduBackup/KuduRestore System.exit(0) results in Spark on YARN failure with exitCode: 16

["Grant Henke (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/KUDU-3099?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Grant Henke resolved KUDU-3099.
-------------------------------
    Fix Version/s: 1.12.0
       Resolution: Fixed

> KuduBackup/KuduRestore System.exit(0) results in Spark on YARN failure with exitCode: 16
> ----------------------------------------------------------------------------------------
>
>                 Key: KUDU-3099
>                 URL: https://issues.apache.org/jira/browse/KUDU-3099
>             Project: Kudu
>          Issue Type: Bug
>          Components: backup, spark
>    Affects Versions: 1.10.0, 1.11.0
>            Reporter: Waleed Fateem
>            Assignee: Waleed Fateem
>            Priority: Major
>             Fix For: 1.12.0
>
>
> When running KuduBackup/KuduRestore the underlying Spark application can fail when running on YARN even when the backup/restore tasks complete successfully. The following was from the Spark driver log:
> {code:java}
> INFO spark.SparkContext: Submitted application: Kudu Table Backup
> ..
> INFO spark.SparkContext: Starting job: save at KuduBackup.scala:90
> INFO scheduler.DAGScheduler: Got job 0 (save at KuduBackup.scala:90) with 200 output partitions
> scheduler.DAGScheduler: Final stage: ResultStage 0 (save at KuduBackup.scala:90)
> ..
> INFO scheduler.DAGScheduler: Submitting 200 missing tasks from ResultStage 0 (MapPartitionsRDD[2] at save at KuduBackup.scala:90) (first 15 tasks are for partitions Vector(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14))
> INFO cluster.YarnClusterScheduler: Adding task set 0.0 with 200 tasks
> ..
> INFO cluster.YarnClusterScheduler: Removed TaskSet 0.0, whose tasks have all completed, from pool 
> INFO scheduler.DAGScheduler: Job 0 finished: save at KuduBackup.scala:90, took 20.007488 s
> ..
> INFO spark.SparkContext: Invoking stop() from shutdown hook
> ..
> INFO cluster.YarnClusterSchedulerBackend: Shutting down all executors
> ..
> INFO spark.SparkContext: Successfully stopped SparkContext
> INFO yarn.ApplicationMaster: Final app status: FAILED, exitCode: 16, (reason: Shutdown hook called before final status was reported.)
> INFO util.ShutdownHookManager: Shutdown hook called{code}
> Spark explicitly added this shutdown hook to catch System.exit() calls and in case this occurs before the SparkContext stops then the application status is considered a failure:
> [https://github.com/apache/spark/blob/branch-2.3/resource-managers/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala#L299]
> The System.exit() call added as part of KUDU-2787 can cause this race condition and that was merged in the 1.10.x and 1.11.x branches. 
>  



--
This message was sent by Atlassian Jira
(v8.3.4#803005)