[
https://issues.apache.org/jira/browse/MAPREDUCE-5417?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13719124#comment-13719124
]
Jason Lowe commented on MAPREDUCE-5417:
---------------------------------------
bq. When the settings for map/reduce tasks conflict, it should be okay to set
to the maximum value.
For heap/container settings I agree this will work for most cases, but what if
the environment variable settings are not identical? e.g.: passing environment
variables to maps but not reduces or worse, passing different values for the
same variable to maps vs. reduces.
I suspect when there are conflicts in map vs. reduce task settings we have to
not run the job as uber or have some hybrid where we run the maps within the AM
and the reduces as separate tasks.
> Propagate task settings to AM when job runs in uber mode
> --------------------------------------------------------
>
> Key: MAPREDUCE-5417
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5417
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: mrv2
> Affects Versions: 2.0.4-alpha
> Reporter: Jason Lowe
>
> When a job runs in Uber-AM mode the JVM settings (e.g.: heap sizing),
> container sizing, and environment variables are not propagated to the AM.
> This means running in uber-AM mode is not as simple as enabling
> {{mapreduce.job.ubertask.enable}} since lack of proper heap/container
> settings can doom the AM to crash or lack of proper environment variables can
> prevent the task code from properly running.
--
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
