[
https://issues.apache.org/jira/browse/KAFKA-18225?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17905568#comment-17905568
]
Colin McCabe commented on KAFKA-18225:
--------------------------------------
The way that the `updateClusterMetadata` API works is going to be incredibly
inefficient since we have to create a full `Cluster` object any time any change
is made. We could support this but we shouldn't be creating the objects unless
the callback is configured. I think we should also have a follow-on KIP to
deprecate this API and introduce one that doesn't require so much copying.
> ClientQuotaCallback#updateClusterMetadata is unsupported by kraft
> -----------------------------------------------------------------
>
> Key: KAFKA-18225
> URL: https://issues.apache.org/jira/browse/KAFKA-18225
> Project: Kafka
> Issue Type: Task
> Reporter: Chia-Ping Tsai
> Assignee: Chia-Ping Tsai
> Priority: Blocker
>
> `ClientQuotaCallback#updateClusterMetadata` is invoked by
> `KafkaAPIs#handleUpdateMetadataRequest` only. However, that is a zk-only API
> and hence, `updateClusterMetadata` is never executed in kraft mode. That
> brings two issues:
> 1. CustomQuotaCallbackTest needs to be rewritten hugely in order to migrate
> to kraft
> 2. Should we deprecate `updateClusterMetadata` directly in 4.0? or make it
> work on kraft mode or introduce new API to replace it
> Regardless to '2', we can enhance `DynamicClientQuotaPublisher` to support
> `ClientQuotaCallback#updateClusterMetadata` when handling `newImage`.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
