Hi, Kafka. We found that the race in topic-deletion procedure could cause the preferred replica imbalance metric to be negative. The phenomenon can easily happen when many topics are deleted at once, and since we use the metric for monitoring, we have to restart the controller to fix the metric every time it happens.
I submitted a patch to fix it: https://github.com/apache/kafka/pull/12405 It'd be appreciated if anyone could review the PR. Thanks, -- ======================== Okada Haruki ocadar...@gmail.com ========================