[
https://issues.apache.org/jira/browse/MAPREDUCE-6763?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jason Lowe updated MAPREDUCE-6763:
----------------------------------
Attachment: MAPREDUCE-6763.001.patch
Attaching a patch that makes the shuffle server listen queue length
configurable via mapreduce.shuffle.listen.queue.size and defaults to 128. I
couldn't find a good way to unit test it, so I manually tested by starting up
the nodemanager with varying listen queue length settings and verifying via
{{ss -ltn}}.
> Shuffle server listen queue is too small
> ----------------------------------------
>
> Key: MAPREDUCE-6763
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-6763
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mrv2
> Reporter: Jason Lowe
> Assignee: Jason Lowe
> Attachments: MAPREDUCE-6763.001.patch
>
>
> ShuffleHandler doesn't specify a listen queue length for the server port, so
> it ends up getting the default listen queue length of 50. This is too small
> to handle bursts of shuffle traffic on large clusters. It's also
> inconsistent with the default Hadoop uses for RPC servers (default=128).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
