During partition reassignment, Kafka will create more replicas in the target broker, and wait for the new replica catching up with the leader, and then remove the old replicas. So during this re-assignment, there will be more replicas than we expected showing up. Our alert rule should be tolerant to this situation.