hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Vinod K V (JIRA)" <j...@apache.org>
Subject [jira] Commented: (MAPREDUCE-1116) Task continues to get localized even though it is marked for killing already
Date Thu, 15 Oct 2009 09:46:31 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-1116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12765988#action_12765988
] 

Vinod K V commented on MAPREDUCE-1116:
--------------------------------------

To solve this, we will need checks for task-status at various stages like job localization,
task localization, and localization code in TaskRunner.

We may also want to wish to interrupt the download of files themselves from DFS.

Note that the jvm itself will not be launched anymore because of already existing checks in
JvmManager.

> Task continues to get localized even though it is marked for killing already
> ----------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-1116
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-1116
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: tasktracker
>            Reporter: Vinod K V
>
> Even after the task is marked for killing, because TT doesn't explicitly check for the
task state at various stages like downloading job jar, job conf and distributed cache files,
it continues to localize the task. This is waste of resources and time and in some cases affects
other tasks. Particularly it continues to download distributed cache files and aggravates
issues like MAPREDUCE-1098.

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


Mime
View raw message