Hi Krik, On Fri, Jul 14, 2023 at 10:59 AM Kirk True <k...@kirktrue.pro> wrote: > Is the requested restructuring of the response “simply” to preserve bytes, or > is it possible that the fetch response could/should/would return leadership > changes for partitions that we’re specifically requested?
Both. My reasoning for the restructuring is that embedding the node endpoint in the partition response would lead to duplicate information being returned and as you point out the node endpoint information is orthogonal to the partition leader. > > 3. In the future, I may use this information in the KRaft/Metadata > > implementation of FETCH. In that implementation not all of the > > replicas are brokers. > > Side point: any references to the change you’re referring to? The idea of > non-brokers serving as replicas is blowing my mind a bit :) I am especially referring to the Draft KIP for KRaft Controller Membership Change (https://cwiki.apache.org/confluence/x/nyH1D). The Fetch RPC is used by KRaft's cluster metadata partition which implements a different consensus protocol that is used by both Controllers and Brokers. Thanks! -- -José
