You may be running into the issue described here:
https://issues.apache.org/jira/browse/KAFKA-16364
If that's the case, I believe you have two options:
1. Reduce emit.checkpoints.interval.seconds and
sync.group.offsets.interval.seconds to reduce the offset lag between source and
target cluster
Hi Massimiliano,
A small amount of re-delivery is expected when using MirrorMaker2 3.5+
offsets replication. This is because MM2 does not offer exactly once offset
replication.
I gave a conference talk recently about this which you may find
informative:
https://current.confluent.io/2024-sessions/
