[
https://issues.apache.org/jira/browse/RATIS-2313?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17987031#comment-17987031
]
Tsz-wo Sze commented on RATIS-2313:
-----------------------------------
LeaderReady means that the leader is able to commit a new transaction. Ratis
should not care about how the StateMachine applies the transactions.
Also, some StateMachines may not want to wait for its pending StateMachine
futures since it will slow down the changing leader process. If a StateMachine
want to wait, it could wait for its pending StateMachine futures when getting
the notification.
I understand that this was to fix a race condition bug from HDDS-12596. Before
HDDS-12596, Ozone does not need it. We should not change the behavior of all
the StateMachines for just one type of StateMachines.
Just an analogy, if there is a StateMachine needs to wait for a minute (for
fixing an internal race condition), we should not add a minute sleep to ratis.
> Wait for pending StateMachine futures to complete before marking the leader
> as ready
> ------------------------------------------------------------------------------------
>
> Key: RATIS-2313
> URL: https://issues.apache.org/jira/browse/RATIS-2313
> Project: Ratis
> Issue Type: Bug
> Components: server
> Reporter: Nandakumar
> Assignee: Nandakumar
> Priority: Major
>
> During leader election we don't wait for the pending StateMachine futures to
> complete before making the leader ready.
> This creates condition where the StateMachine might think that all the
> pending transactions from Raft Log are applied to the StateMachine as the
> leader is marked as ready, but the statemachine futures are not yet completed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
