[jira] [Commented] (IMPALA-10599) The coordinator should avoid distributing query fragments to busy nodes.

["Fifteen (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IMPALA-10599?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17512788#comment-17512788 ] 

Fifteen commented on IMPALA-10599:
----------------------------------

So, how to identify a node's status?

How about we add a new topic to the statestored? 
After that, the coordinator subscribes that topic and keeps a cache of node status.

The cache is refreshed periodically so that there won't be much RPC lag for small queries.

 

Moreover, if we have a global node status cache, maybe we can build a better scheduling policy than round-robin?    

> The coordinator should avoid distributing query fragments to busy nodes.
> ------------------------------------------------------------------------
>
>                 Key: IMPALA-10599
>                 URL: https://issues.apache.org/jira/browse/IMPALA-10599
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Distributed Exec
>    Affects Versions: Impala 3.2.0, Impala 3.4.0
>         Environment: LSB Version: core-4.1-amd64
> Distributor ID: CentOS
> Description: CentOS release 7.4.1708.7
> Release: 7.4.1708.7
> impala version: 3.2.0
>            Reporter: zhi tang
>            Priority: Major
>              Labels: features
>
> When I join[shuffle] with unbalanced fields, there will be data-skewed nodes during execution.This will result in the skewed node being under high load for a long time, because the query is likely to be endless.If other queries are committed or running during this time, they may be rejected or an exception thrown.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org