You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Pierre-Yves Bigourdan (Jira)" <ji...@apache.org> on 2021/12/15 16:59:00 UTC

[jira] [Comment Edited] (KAFKA-8495) Make Round-robin / RangeAssignor to be "sticky" (part 5)

    [ https://issues.apache.org/jira/browse/KAFKA-8495?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17460099#comment-17460099 ] 

Pierre-Yves Bigourdan edited comment on KAFKA-8495 at 12/15/21, 4:58 PM:
-------------------------------------------------------------------------

[~ableegoldman] could you please clarify "round-robin/range assignors will deliberately not support cooperative rebalancing", or point to some relevant discussions providing more context?

I'm currently working on a project that would benefit from moving to a cooperative protocol to reduce the performance hit incurred by rebalances, but unfortunately the way the sticky algorithm assigns partitions to consumers isn't suitable for my use case.


was (Author: pyves):
[~ableegoldman] could you please clarify "round-robin/range assignors will deliberately not support cooperative rebalancing", or point to some relevant discussions providing more context?

I'm currently working on a project that would benefit from moving to a cooperative protocol to reduce the performance hit incurred from rebalances, but unfortunately the way the sticky algorithm assigns partitions to consumers isn't suitable for my use case.

> Make Round-robin / RangeAssignor to be "sticky" (part 5)
> --------------------------------------------------------
>
>                 Key: KAFKA-8495
>                 URL: https://issues.apache.org/jira/browse/KAFKA-8495
>             Project: Kafka
>          Issue Type: Sub-task
>          Components: consumer
>            Reporter: Guozhang Wang
>            Priority: Major
>
> For this new algorithm to be effective in reducing rebalance costs, it is really expecting the plug-in assignor to be "sticky" in some way, such that the diff of the newly-assigned-partitions and the existing-assigned-partitions can be small, and hence only a few subset of the total number of partitions need to be revoked / migrated at each rebalance in practice – otherwise, we are just paying more rebalance for little benefits.



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