[
https://issues.apache.org/jira/browse/MAPREDUCE-5814?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13949774#comment-13949774
]
Gera Shegalov commented on MAPREDUCE-5814:
------------------------------------------
Sangjin, I was thinking of this... but *-site.xml unlike job.xml are not meant
to be shipped, they are meant to be installed. default.xml is packaged with
hadoop, and can therefore make it into a fat jar. It's a good idea as a
defensive measure. The only reason that prevented me from doing this is that
the list of such files can be quite long, and include all kinds of log4j and
other properties. But they can be legally shipped too. So far I have settled
for the least requirement. Looking forward to the pros and cons.
> fat jar with *-default.xml may fail when mapreduce.job.classloader=true.
> ------------------------------------------------------------------------
>
> Key: MAPREDUCE-5814
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5814
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mrv2
> Affects Versions: 2.3.0
> Reporter: Gera Shegalov
> Assignee: Gera Shegalov
> Attachments: MAPREDUCE-5814.v01.patch
>
>
> We faced a failure when a job.jar compiled against 0.20+ hadoop artifacts had
> to run with {{mapreduce.job.classloader=true}} because it needed a more
> recent guava as a dependency. The job failed because the cluster's
> {{*-default.xml}} files were overshadowed by the ones in the fat jar. We
> propose to treat these default config files like the system packages
> {{org.apache.hadoop.}} to avoid a counterintuitivie behavior as if we had
> {{mapreduce.job.user.classpath.first}} set.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
