Hi, Jose, The follower of regular topic could have fetched to the end of the log from the leader, but need to wait 500ms for the true HWM to be propagated. This will delay the consumption of the consumer on the follower by 500ms since it's gated by HWM, right?
Also, in FetchResponse, we use -1 as the default for various offsets. Should we follow that convention for the new field? Thanks, Jun On Wed, Apr 23, 2025 at 2:16 PM José Armando García Sancio <jsan...@confluent.io.invalid> wrote: > Hi Jun, > > > The issue described in the KIP applies to regular topics too. If the HWM > is > > delayed in the follower, it could delay the consumption for a client > > fetching from a follower. It would be useful to document that too. > > Let me look at the code again, but I think that regular topics are not > affected by this issue because Kafka consumers tend to consume with a > FetchIsolation.HIGH_WATERMARK. In that case the FetchOffset in the > FETCH request plays a similar role as HighWatermark for > FetchIsolation.LOG_END. > > Or when you say follower, do you mean broker replicas of regular > topics? To my knowledge we don't have any read-only coordinators for > regular topics like we do the cluster metadata partition. We can still > fix this for regular topics but I think that the impact for regular > topics is less given the current usage. > > What do you think? > > Thanks, > -- > -José >
