hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arun C Murthy (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-2803) Race condition in DistributedCache
Date Wed, 13 Feb 2008 22:19:07 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-2803?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12568729#action_12568729
] 

Arun C Murthy commented on HADOOP-2803:
---------------------------------------

bq. We ran a job with the wrong file, then around 50 minutes later we put the fixed version
into DFS, and ran the same job again.

Just to clarify: so it was the 2nd job which had these exceptions? And the 3rd job succeeded?
(same code, different jobids)

> Race condition in DistributedCache
> ----------------------------------
>
>                 Key: HADOOP-2803
>                 URL: https://issues.apache.org/jira/browse/HADOOP-2803
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: mapred
>    Affects Versions: 0.16.0
>            Reporter: Christian Kunz
>            Assignee: Mahadev konar
>             Fix For: 0.17.0
>
>
> When an older version of a file in DistributedCache exists locally and multiple tasks
per node start, they can run into a race condition:
> dir/mapred/local/taskTracker/archive/subdir/filename is in use and cannot be refreshed
> 	at org.apache.hadoop.filecache.DistributedCache.localizeCache(DistributedCache.java:313)
> 	at org.apache.hadoop.filecache.DistributedCache.getLocalCache(DistributedCache.java:161)
> 	at org.apache.hadoop.mapred.TaskRunner.run(TaskRunner.java:134)
> We ran a job with the wrong file, then around 50 minutes later we put the fixed version
into DFS, and ran the same job again. The job had 11,000 maps ~ about 4-5 waves of map tasks
and produced 3,500 failed tasks with above error. We eventually killed it and restarted the
same job again, with no problems this time.

-- 
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