[jira] Updated: (HADOOP-2051) JobTracker's TaskCommitQueue is vulnerable to non-IOExceptions

["Arun C Murthy (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HADOOP-2051?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Arun C Murthy updated HADOOP-2051:
----------------------------------

    Attachment: HADOOP-2051_1_20071013.patch

Here is a patch which ignores all exceptions (after logging) in {{JobTracker#TaskCommitQueue#run}}, similar to other long-running threads in the {{JobTracker}}.

> JobTracker's TaskCommitQueue is vulnerable to non-IOExceptions
> --------------------------------------------------------------
>
>                 Key: HADOOP-2051
>                 URL: https://issues.apache.org/jira/browse/HADOOP-2051
>             Project: Hadoop
>          Issue Type: Bug
>          Components: mapred
>    Affects Versions: 0.15.0
>            Reporter: Arun C Murthy
>            Assignee: Arun C Murthy
>            Priority: Blocker
>             Fix For: 0.15.0
>
>         Attachments: HADOOP-2051_1_20071013.patch
>
>
> The {{JobTracker#TaskCommitQueue#run}} method only handles {{IOException}}s. Christian Kunz ran into a scenario where a job was stuck with all tasks in {{COMMIT_PENDING}} state and the stack traces showed that the "Task Commit Thread" wasn't even around.
> The work-around is to model {{TaskCommitQueue#run}} along the lines of other long-running threads in the {{JobTracer}} ({{ExpireLaunchingTasks}}, {{ExpireTrackers}} etc.) to catch, log and ignore any {{Exception}} in a loop.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.