[jira] [Commented] (MAPREDUCE-4901) JobHistoryEventHandler errors should be fatal

["Robert Joseph Evans (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAPREDUCE-4901?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13539581#comment-13539581 ] 

Robert Joseph Evans commented on MAPREDUCE-4901:
------------------------------------------------

Oh so looking at the code a bit more, it also appears that if an error does happen the thread that writes to the log dies and the event queue then grows without bound, so technically there is a memory leak involved here too.
                
> JobHistoryEventHandler errors should be fatal
> ---------------------------------------------
>
>                 Key: MAPREDUCE-4901
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-4901
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mrv2
>    Affects Versions: 0.23.0, 2.0.0-alpha
>            Reporter: Robert Joseph Evans
>            Assignee: Robert Joseph Evans
>
> To be able to truly fix issues like MAPREDUCE-4819 and MAPREDUCE-4832, we need a 2 phase commit where a subsequent AM can be sure that at a specific point in time it knows exactly if any tasks/jobs are committing.  The job history log is already used for similar functionality so we would like to reuse this, but we need to be sure that errors while writing out to the job history log are now fatal.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira