[
https://issues.apache.org/jira/browse/IMPALA-3393?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17140787#comment-17140787
]
Tim Armstrong commented on IMPALA-3393:
---------------------------------------
I think there might be value to adding more jitter to status reporting for
longer-running queries. I think, regardless, we can't avoid some
synchronisation of status reports since backends will all tend to finish at the
same time. . But IMPALA-7213 should fix many scalability problems.
> Report thread wake up interval is not well distributed
> ------------------------------------------------------
>
> Key: IMPALA-3393
> URL: https://issues.apache.org/jira/browse/IMPALA-3393
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Affects Versions: Impala 2.5.0
> Reporter: Huaisi Xu
> Priority: Minor
>
> https://github.com/cloudera/Impala/blob/cdh5-trunk/be/src/runtime/plan-fragment-executor.cc#L398
> We take an integer from 0 to 4. so if there are 200 nodes then ~40 of them
> will report at the same time every 5 seconds. This does not take into account
> fragment start time is basically random...
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
