Hi Andrew:

Thank you very much for joining the discussion and providing your comments.
I feel this KIP shouldn’t be very controversial since it follows a standard
practice and kafka already have internal topic practices to reference. So
I’m thinking of leaving the discussion open for another week, and then we
can probably start/continue the voting. This also exceeds the one-month
discussion period to ensure everyone has had a chance to see it.

Thanks again.

Regards
Jian

Andrew Schofield <[email protected]> 于2025年12月3日周三 01:46写道:

> Hi Jian,
> Thanks for the KIP. This looks like a sensible idea to improve the default
> configuration for this topic.
>
> Thanks,
> Andrew
>
> > On 14 Nov 2025, at 05:27, jian fu <[email protected]> wrote:
> >
> > Just bumping this thread for more discussion.
> >
> > Thanks!
> >
> > jian fu <[email protected]> 于2025年11月5日周三 14:54写道:
> >
> >> Hi everyone,
> >>
> >> I'd like to start a discussion on KIP-1235, a small change to set
> >> min.isr=2 for the __remote_log_metadata topic.
> >>
> >> KIP:
> >>
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-1235%3A+Correct+the+default+min.insync.replicas+to+2+for+the+__remote_log_metadata+topic
> >>
> >> Problem:
> >>
> >> Currently, __remote_log_metadata relies on the broker default for
> min.isr
> >> (typically 1).
> >> With min.isr=1 and factor=3, acknowledged writes can be lost if the
> leader
> >> crashes before replication
> >> completes, making remote data permanently unreachable.
> >>
> >> What's more, the similar topic: __transaction_state sets min.isr=2 via
> >> transaction.state.log.min.isr.
> >> We should provide the same durability for __remote_log_metadata.
> >>
> >>
> >> Proposed Change:
> >>
> >> Add one new broker config: remote.log.metadata.topic.min.isr (default:
> 2)
> >>
> >> When creating __remote_log_metadata, TopicBasedRemoteLogMetadataManager
> >> will set
> >> min.insync.replicas=2 at the topic level. The change only affects new
> >> clusters enabling Tiered Storage. Existing topics are unchanged.
> >>
> >>
> >> Looking forward to your feedback!
> >>
> >> Best regards,
> >> Jian
> >>
>
>

Reply via email to