hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jason Lowe (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-5211) Reducer intermediate files can collide during merge
Date Thu, 30 May 2013 18:32:23 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-5211?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13670574#comment-13670574
] 

Jason Lowe commented on MAPREDUCE-5211:
---------------------------------------

Is this actually a problem in 2.0.4?  I believe after MAPREDUCE-2264, which is in 2.0.3, MapOutput
objects are not used to generate the pathnames and the problem does not occur.  The pathnames
can get very long, since they concatenate two absolute paths to form the new path, but I didn't
think it was possible for two paths to collide for the same reducer.  Also the reduce attempt
ID should appear at least once in the pathname, so it should also be impossible for paths
to collide between reducers.
                
> Reducer intermediate files can collide during merge
> ---------------------------------------------------
>
>                 Key: MAPREDUCE-5211
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5211
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mrv2
>    Affects Versions: 0.23.7
>            Reporter: Jason Lowe
>            Assignee: Jason Lowe
>            Priority: Blocker
>              Labels: 2.0.4.2
>             Fix For: 0.23.8
>
>         Attachments: MAPREDUCE-5211.branch-0.23.patch
>
>
> The OnDiskMerger.merge method constructs an output path that is not unique to a reduce
attempt, and as a result can result in a file collision with other reducers from the same
app that are running on the same node.  In addition the name of the output file is based on
MapOutput.toString which may not be unique in light of multi-pass merges on disk since the
mapId will be null and the basename ends up as "MapOutput(null, DISK)"

--
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

Mime
View raw message