twthorn opened a new pull request, #23680: URL: https://github.com/apache/kafka/pull/23680
Implements the aggressive and balanced unclean recovery strategies from KIP-966. Follows up on work that was implemented then [reverted](https://github.com/apache/kafka/pull/20371). This takes a push-based model to minimize the changes needed. Instead of pulling the replica info from brokers (new RPC) it adds in leaderless partition info (log end offset, last epoch) to the heartbeats already pushed to controllers regularly. The controller elects replica with the highest epoch, then the highest log end offset (minimize data loss). Balanced waits for last known ELR members to report. Both fall back to the best report after `unclean.recovery.timeout.ms`. Disabled by default and gated under metadata version. Testing: unit tests in ReplicationControlManagerTest, BrokerLifecycleManagerTest, ReplicaManagerTest, MetadataVersionTest. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
