@lindong28 Thanks for sharing the context. In my opinion, the current PR address 1) and all the other issues can be addressed by adding broker generation because: 1. For 2), what we need is essentially a mechanism to detect bounced brokers. If we have broker generation, controller can check it in BrokerChange event and implement the fix without the trick we used in the hotfix patch. Note that this is a bug that can potentially happen with and without this PR. 2. For 3) and 4), what we need is essentially a mechnism to let broker reject stall controller request. If we have broker generation, controller can check it when handling StopReplica/LeaderAndIsr/UpdateMetadata and reject requests sent to the previous generation.
What https://issues.apache.org/jira/browse/KAFKA-7235 suggests is adding broker generation support. This requires a KIP and I am currently working on it. Will submit a KIP soon. We can revisit this PR after we have broker generation support. [ Full content available at: https://github.com/apache/kafka/pull/5498 ] This message was relayed via gitbox.apache.org for [email protected]
