You are viewing a plain text version of this content. The canonical link for it is here.

Posted to jira@kafka.apache.org by "Matthias J. Sax (Jira)" <ji...@apache.org> on 2020/05/29 18:02:00 UTC

[jira] [Commented] (KAFKA-10068) Verify HighAvailabilityTaskAssignor performance with large clusters and topologies

    [ https://issues.apache.org/jira/browse/KAFKA-10068?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17119817#comment-17119817 ] 

Matthias J. Sax commented on KAFKA-10068:
-----------------------------------------

> 30 seconds is a good upper bound.

Why that? The broker side group coordinator applies `max.poll.interval.ms` as timeout on the consumer group leader to compute the assignment (otherwise, a new rebalance with new leader is done and the old leader is removed from the group) – hence, it might make sense to apply like 50% of `max.poll.interval.ms`.

This is also discussion via KIP-572.

> Verify HighAvailabilityTaskAssignor performance with large clusters and topologies
> ----------------------------------------------------------------------------------
>
>                 Key: KAFKA-10068
>                 URL: https://issues.apache.org/jira/browse/KAFKA-10068
>             Project: Kafka
>          Issue Type: Task
>          Components: streams
>    Affects Versions: 2.6.0
>            Reporter: John Roesler
>            Priority: Blocker
>
> While reviewing [https://github.com/apache/kafka/pull/8668/files,] I realized that we should have a similar test to make sure that the new task assignor completes well within the default assignment deadline. 30 seconds is a good upper bound.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)