[
https://issues.apache.org/jira/browse/RATIS-1982?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wei-Chiu Chuang updated RATIS-1982:
-----------------------------------
Description:
[^hbase_rs-ccycloud-2.weichiu.root.comops.site.html]
Running HBase RegionServer as an Ozone/Ratis client, I noticed more than 20% of
heap allocation (chart uploaded in the attachment) is spent updating
LEADER_CACHE guava cache in RaftClientImpl. In this application, client
receives hundreds to thousands of relies per second. Requests are small.
Supposedly this cache doesn't change very often. Perhaps we can reduce heap
allocation by skipping the cache update when the existing object matches the
value to be inserted.
was:
[^hbase_rs-ccycloud-2.weichiu.root.comops.site.html]
Running HBase RegionServer as an Ozone/Ratis client, I noticed more than 20% of
heap allocation (chart uploaded in the attachment) is spent updating
LEADER_CACHE guava cache in RaftClientImpl. In this application, client
receives hundreds to thousands of relies per second.
Supposedly this cache doesn't change very often. Perhaps we can reduce heap
allocation by skipping the cache update when the existing object matches the
value to be inserted.
> Optimize Ratis Client LEADER_CACHE heap allocation
> --------------------------------------------------
>
> Key: RATIS-1982
> URL: https://issues.apache.org/jira/browse/RATIS-1982
> Project: Ratis
> Issue Type: Improvement
> Reporter: Wei-Chiu Chuang
> Assignee: Wei-Chiu Chuang
> Priority: Major
> Attachments: hbase_rs-ccycloud-2.weichiu.root.comops.site.html
>
>
> [^hbase_rs-ccycloud-2.weichiu.root.comops.site.html]
> Running HBase RegionServer as an Ozone/Ratis client, I noticed more than 20%
> of heap allocation (chart uploaded in the attachment) is spent updating
> LEADER_CACHE guava cache in RaftClientImpl. In this application, client
> receives hundreds to thousands of relies per second. Requests are small.
> Supposedly this cache doesn't change very often. Perhaps we can reduce heap
> allocation by skipping the cache update when the existing object matches the
> value to be inserted.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
