[
https://issues.apache.org/jira/browse/KAFKA-15010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17728773#comment-17728773
]
Mickael Maison commented on KAFKA-15010:
----------------------------------------
Since the PR is merged I'm marking this as resolved.
If there is follow up work left to do, feel free to reopen (and adjust Fix
Version/s) or create a new ticket.
> KRaft Controller doesn't reconcile with Zookeeper metadata upon becoming new
> controller while in dual write mode.
> -----------------------------------------------------------------------------------------------------------------
>
> Key: KAFKA-15010
> URL: https://issues.apache.org/jira/browse/KAFKA-15010
> Project: Kafka
> Issue Type: Bug
> Components: kraft
> Affects Versions: 3.5.0
> Reporter: Akhilesh Chaganti
> Assignee: David Arthur
> Priority: Blocker
> Fix For: 3.5.0
>
>
> When a KRaft controller fails over, the existing migration driver (in dual
> write mode) can fail in between Zookeeper writes and may leave Zookeeper with
> incomplete and inconsistent data. So when a new controller becomes active
> (and by extension new migration driver becomes active), this first thing we
> should do is load the in-memory snapshot and use it to write metadata to
> Zookeeper to have a steady state. We currently do not do this and it may
> leave Zookeeper in inconsistent state.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
