[
https://issues.apache.org/jira/browse/MAPREDUCE-7070?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17272540#comment-17272540
]
Zhankun Tang commented on MAPREDUCE-7070:
-----------------------------------------
[~gphillips], Thanks for the contribution!
Is it better to develop a new "MRRackResolver" which decouple this from YARN?
In that case, the MR application can run on a YARN version without YARN-9332.
> Batch calls to RackResolver from AM
> -----------------------------------
>
> Key: MAPREDUCE-7070
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-7070
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: applicationmaster, performance
> Reporter: Greg Phillips
> Priority: Minor
> Attachments: MAPREDUCE-7070.patch
>
>
> TaskAttemptImpl calls RackResolver once for every split location. This
> results in many shells being spawned by the Application Master when the
> topology script is enabled. The worst case scenario would entail one call to
> the topology script for each Datanode. The NameNode is also topology aware,
> and uses batch calls to the topology script to improve performance.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
