[
https://issues.apache.org/jira/browse/OOZIE-1335?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13637244#comment-13637244
]
Rohini Palaniswamy commented on OOZIE-1335:
-------------------------------------------
You need to ensure that the uber mode setting applies only to the launcher job.
It should not be propagated to the jobs launched by the launcher job. The input
split size is not always an indicator of the size of the job. For eg: distcp
the input file of the job is small but the data it works on it is huge and it
will be a disaster to run it in uber mode.
> The launcher job should use uber mode in Hadoop 2 by default
> ------------------------------------------------------------
>
> Key: OOZIE-1335
> URL: https://issues.apache.org/jira/browse/OOZIE-1335
> Project: Oozie
> Issue Type: Bug
> Affects Versions: trunk
> Reporter: Robert Kanter
> Assignee: Robert Kanter
> Fix For: trunk
>
>
> In theory, running the launcher job in Hadoop 2's uber mode should make the
> overall action time faster because we don't need to launch as many JVMs.
> I tested this out on a few of the examples in a VM, and it seems to save
> about 10 - 50 seconds per action. While my testing wasn't scientific and was
> only on a single VM, setting the launcher job to run in uber mode by default
> is such a trivial change and should at least not make it slower that I think
> we should do it.
--
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
