[
https://issues.apache.org/jira/browse/RATIS-2311?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Andika updated RATIS-2311:
-------------------------------
Description:
Currently TransferLeadership#isFollowerUpToDate only waits for the up-to-date
matchIndex before triggering election on the target Ratis server.
However, if the target Ratis server appliedIndex is way below the matchIndex,
once the Ratis server is promoted to a leader, it might take a while for it to
catch up to the current appliedIndex and acknowlegding the requests. This can
have a noticeable effect on the latency and throughput of the Raft group.
We might make the transfer leadership check to be more strict by ensuring that
leader and follower appliedIndex gap is within a certain acceptable threshold
(e.g 10,000).
was:
Currently TransferLeadership#isFollowerUpToDate only waits for the up-to-date
matchIndex before triggering election on the target Ratis server.
However, if the target Ratis server appliedIndex is way below the matchIndex,
once the Ratis server is promoted to a leader, it might take a while for it to
catch up to the current appliedIndex and acknowlegding the requests. This can
have a noticeable effect on the latency and throughput of the leader.
We might make the transfer leadership check to be more strict by ensuring that
leader and follower appliedIndex gap is within a certain acceptable threshold
(e.g 10,000).
> Allow TransferLeadership#isFollowerUpToDate to wait for appliedIndex
> --------------------------------------------------------------------
>
> Key: RATIS-2311
> URL: https://issues.apache.org/jira/browse/RATIS-2311
> Project: Ratis
> Issue Type: Improvement
> Reporter: Ivan Andika
> Priority: Major
>
> Currently TransferLeadership#isFollowerUpToDate only waits for the up-to-date
> matchIndex before triggering election on the target Ratis server.
> However, if the target Ratis server appliedIndex is way below the matchIndex,
> once the Ratis server is promoted to a leader, it might take a while for it
> to catch up to the current appliedIndex and acknowlegding the requests. This
> can have a noticeable effect on the latency and throughput of the Raft group.
> We might make the transfer leadership check to be more strict by ensuring
> that leader and follower appliedIndex gap is within a certain acceptable
> threshold (e.g 10,000).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
