Hi Jun, On Fri, May 2, 2025 at 1:47 PM Jun Rao <j...@confluent.io.invalid> wrote: > Since we are only doing the fast HWM propagation for the KRaft partition, > should we remove the changes in Replica manager where the leader will > unpark the follower fetch request if its HWM is larger than the follower? > Also, followers for non-KRaft partitions probably don't need to set the HWM > field.
I updated the KIP to remove the changes to replica manager and that both consumer and replica fetcher should not set this field. Note that the issue is not specific to KRaft. It is more of a question of how to efficiently replicate the HWM when there are few low-throughput partitions vs a lot of high-throughput partitions. Thanks, -- -José
