[
https://issues.apache.org/jira/browse/SPARK-9111?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15080452#comment-15080452
]
Steve Loughran commented on SPARK-9111:
---------------------------------------
the heap dump option could itself be useful; within a YARN container the
launch command could be set to something like {{-XX:HeapDumpPath=<<LOGDIR>>}}
then the heap dump would be automatically grabbed by the YARN Nodemanager and
copied to HDFS, where it would then be cleaned up by the normal YARN history
cleanup routines.
> Dumping the memory info when an executor dies abnormally
> --------------------------------------------------------
>
> Key: SPARK-9111
> URL: https://issues.apache.org/jira/browse/SPARK-9111
> Project: Spark
> Issue Type: Sub-task
> Components: Spark Core
> Reporter: Zhang, Liye
> Priority: Minor
>
> When an executor is not normally finished, we shall give out it's memory dump
> info right before the JVM shutting down. So that if the executor is killed
> because of OOM, we can easily checkout how is the memory used and which part
> cause the OOM.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
