You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Shane (Jira)" <ji...@apache.org> on 2020/07/24 22:08:00 UTC

[jira] [Comment Edited] (KAFKA-9731) Increased fetch request rate with leader selector due to HW propagation

    [ https://issues.apache.org/jira/browse/KAFKA-9731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17163061#comment-17163061 ] 

Shane edited comment on KAFKA-9731 at 7/24/20, 10:07 PM:
---------------------------------------------------------

Hey Kafka team, 
 What are the odds of back-porting this to 2.5.  I'm seeing the same issues in 2.5.
 I can add more details if you want to see, but we're seeing the same increase in fetch requests per second. 

Thanks!


was (Author: shanesaww):
Hey Apache team, 
What are the odds of back-porting this to 2.5.  I'm seeing the same issues in 2.5.
I can add more details if you want to see, but we're seeing the same increase in fetch requests per second. 

Thanks!

> Increased fetch request rate with leader selector due to HW propagation
> -----------------------------------------------------------------------
>
>                 Key: KAFKA-9731
>                 URL: https://issues.apache.org/jira/browse/KAFKA-9731
>             Project: Kafka
>          Issue Type: Improvement
>          Components: core
>    Affects Versions: 2.4.0, 2.4.1
>            Reporter: Vahid Hashemian
>            Assignee: Ismael Juma
>            Priority: Major
>             Fix For: 2.6.0
>
>         Attachments: image-2020-03-17-10-19-08-987.png
>
>
> KIP-392 adds high watermark propagation to followers as a means to better sync up followers HW with leader. The issue we have noticed after trying out 2.4.0 and 2.4.1 is a spike in fetch request rate in the default selector case (leader), that does not really require this high watermark propagation:
> !image-2020-03-17-10-19-08-987.png|width=811,height=354!
> This spike causes an increase in resource allocation (CPU) on the brokers.
> An easy solution would be to disable this propagation (at least) for the default leader selector case to improve the backward compatibility.



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