[
https://issues.apache.org/jira/browse/IGNITE-23597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17896578#comment-17896578
]
Roman Puchkovskiy commented on IGNITE-23597:
--------------------------------------------
The core problem seems to be that we have too many replicators sending their
heartbeats. There is a prototype in IGNITE-16183 where an attempt to solve this
(and probably other problems related to having many Raft groups) was made
> Idle Raft Replicator eats too much CPU
> --------------------------------------
>
> Key: IGNITE-23597
> URL: https://issues.apache.org/jira/browse/IGNITE-23597
> Project: Ignite
> Issue Type: Bug
> Reporter: Roman Puchkovskiy
> Assignee: Roman Puchkovskiy
> Priority: Major
> Labels: ignite-3
>
> I noted that nodes of an idle cluster (taking no transactional load) which
> has 3 nodes and 1200 partitions (replication factor is 2 so there is a
> replicator per each partition, so 400 replicators per node) eat a
> considerable amount of CPU (approximately 12% of each of 8 CPU cores).
> According to the JFR, all work is done by replicators (sending side) and
> AppendEntries controllers (receiving side). Both spend most of the time in
> LogStorage#getTerm().
> We should try to optimize this method (probably caching terms for last N
> indices).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
