Hi Chia-Ping,

Thanks for bringing some momentum back to this discussion. The original
implementation of this API had some flaws, which I attempted to address in
KAFKA-19122 <https://issues.apache.org/jira/browse/KAFKA-19122>.
However, I later realized that properly fixing the issue would require
substantial changes to the API itself.
And as you pointed out, if this API isn't being used, it might not be worth
the effort to overhaul it.

I agree with your suggestion — we can start by deprecating the
updateClusterMetadata API and revisit the current KIP if there's a need in
the future.
I'll start drafting another KIP to cover the deprecation. Thanks again!

Best,
Kuan-Po

On Tue, Jul 22, 2025 at 4:16 PM Chia-Ping Tsai <chia7...@apache.org> wrote:

> hi Kuan-Po
>
> The method `updateClusterMetadata` hasn't been implemented for over three
> years after kraft was introduced. It seems we still function well without
> it.
>
> If the method is rarely used, introducing so many new public APIs for it
> seems a bit overkill. We could file a KIP to just deprecate it and then
> remove it from 5.0.
>
> Of course, we could revisit this KIP if anyone has a use case for the
> method.
>
> Best,
> Chia-Ping
>
> On 2025/04/22 16:23:31 Kuan Po Tseng wrote:
> > Hello everyone,
> >
> > I’d like to kick off a discussion about redesigning
> > ClientQuotaCallback#updateClusterMetadata.
> > As it stands, the current implementation is quite inefficient, since it
> > requires creating
> > a full Cluster object each time any change is made. On top of that, some
> of
> > the information
> > within the Cluster object is outdated.
> >
> > With that in mind, I’d love to propose a lighter, more efficient
> solution,
> > along with a redefined
> > interface for the object passed into updateClusterMetadata. If you have a
> > moment, I’d really appreciate your thoughts and feedback on this.
> >
> > Thank you!
> >
> > KIP link: https://cwiki.apache.org/confluence/x/zInoF
> >
> > Best,
> > Kuan-Po Tseng
> >
>

Reply via email to