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

Posted to jira@kafka.apache.org by "Divij Vaidya (Jira)" <ji...@apache.org> on 2022/11/28 14:27:00 UTC

[jira] [Updated] (KAFKA-14423) Optimize CPU bottleneck caused by metrics capture

     [ https://issues.apache.org/jira/browse/KAFKA-14423?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Divij Vaidya updated KAFKA-14423:
---------------------------------
    Attachment: flamegraph1.html

> Optimize CPU bottleneck caused by metrics capture
> -------------------------------------------------
>
>                 Key: KAFKA-14423
>                 URL: https://issues.apache.org/jira/browse/KAFKA-14423
>             Project: Kafka
>          Issue Type: Improvement
>          Components: network
>            Reporter: Divij Vaidya
>            Priority: Major
>         Attachments: Screenshot 2022-11-28 at 12.11.23.png, Screenshot 2022-11-28 at 15.26.30.png, flamegraph1.html
>
>
> I ran a produce-only workload on a plaintext single broker. One of the main findings was that kafka.network.Processor.updateRequestMetrics(kafka.network.RequestChannel$Response method is expensive to the extent that 90% of the time taken by Processor.processCompletedSends() is taken by this method.
> The flamegraph is attached.
> The objective of this ticket is to optimize the performance of metrics collection here.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)