[
https://issues.apache.org/jira/browse/HBASE-26981?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17558356#comment-17558356
]
Duo Zhang commented on HBASE-26981:
-----------------------------------
What about the qps? Do we have higher qps then before?
Checked the flame graph, the result is good. There is no CellUtil.cloneRow any
more, and it seems that the CPU time spend in MetaCellComparator.compare is
also reduced?
So then you may need to use [~huaxiangsun]'s solution, use the meta read
replica feature to distribute the load to multiple region servers.
And out of interest, what is the typical work load on your cluister? Usually,
hbase clients should cache the location locally, so they do not need to request
meta too much.
Thanks.
> The CPU usage of the regionserver node where the meta table is located is too
> high
> ----------------------------------------------------------------------------------
>
> Key: HBASE-26981
> URL: https://issues.apache.org/jira/browse/HBASE-26981
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 2.3.4
> Reporter: zhengsicheng
> Priority: Major
> Attachments: 20220624-144119.html, image-2022-04-27-20-24-33-252.png,
> image-2022-04-27-20-45-09-227.png, image-2022-04-28-09-59-40-567.png,
> image-2022-06-24-14-48-26-451.png, image-2022-06-24-14-57-10-424.png, jstack
>
>
> When the read and write pressure is high, the CPU usage of the meta table
> node is too high
> !image-2022-04-28-09-59-40-567.png!
>
> !image-2022-04-27-20-24-33-252.png!
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
