You are viewing a plain text version of this content. The canonical link for it is here.
Posted to mapreduce-issues@hadoop.apache.org by "Vinod K V (JIRA)" <ji...@apache.org> on 2009/10/23 11:40:59 UTC

[jira] Created: (MAPREDUCE-1141) Localization of a task's distributed-cache files gets blocked by deletion of old unrelated files when mapreduce.tasktracker.cache.local.size is hit.

Localization of a task's distributed-cache files gets blocked by deletion of old unrelated files when mapreduce.tasktracker.cache.local.size is hit.
----------------------------------------------------------------------------------------------------------------------------------------------------

                 Key: MAPREDUCE-1141
                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-1141
             Project: Hadoop Map/Reduce
          Issue Type: Bug
          Components: tasktracker
    Affects Versions: 0.20.2, 0.21.0, 0.22.0
            Reporter: Vinod K V




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


[jira] Resolved: (MAPREDUCE-1141) Localization of a task's distributed-cache files gets blocked by deletion of old unrelated files when mapreduce.tasktracker.cache.local.size is hit.

Posted by "Vinod K V (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/MAPREDUCE-1141?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Vinod K V resolved MAPREDUCE-1141.
----------------------------------

    Resolution: Duplicate

MAPREDUCE-1302 already started implementation for this issue. Closing this as duplicate.

> Localization of a task's distributed-cache files gets blocked by deletion of old unrelated files when mapreduce.tasktracker.cache.local.size is hit.
> ----------------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-1141
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-1141
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: tasktracker
>    Affects Versions: 0.20.2, 0.21.0, 0.22.0
>            Reporter: Vinod K V
>


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


[jira] Commented: (MAPREDUCE-1141) Localization of a task's distributed-cache files gets blocked by deletion of old unrelated files when mapreduce.tasktracker.cache.local.size is hit.

Posted by "Vinod K V (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/MAPREDUCE-1141?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12769151#action_12769151 ] 

Vinod K V commented on MAPREDUCE-1141:
--------------------------------------


Here's roughly the offending code:

{code}
getLocalizeCache(URI uri) {
   localizeCache();

   if(currentDiskUsage > conf.get(mapreduce.tasktracker.cache.local.size)) {
      deleteOldFilesInCache();
   }
}
{code}

So, when mapreduce.tasktracker.cache.local.size is hit, localization of a task will not complete _till_ the deleltion of (possibly a large number of) old DistributedCache files is completed (which can possibly take long time).

Localization of a task should not ideally be blocked on cleanup of old files. Clean-up can best be done in a separate thread.

> Localization of a task's distributed-cache files gets blocked by deletion of old unrelated files when mapreduce.tasktracker.cache.local.size is hit.
> ----------------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-1141
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-1141
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: tasktracker
>    Affects Versions: 0.20.2, 0.21.0, 0.22.0
>            Reporter: Vinod K V
>


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