krishvishal opened a new issue, #4515: URL: https://github.com/apache/iggy/issues/4515
## Summary Group membership lives in the memory of one gateway, and Metadata reports a single broker (`gateways/kafka/docs/CONSUMER_GROUPS.md`, "One gateway per bootstrap endpoint"). Who hits it: every production deployment. Clients list several bootstrap servers and expect no single point of failure. - A gateway restart disconnects every client, and every group rebalances. - Two gateways behind one endpoint split each group in two, and every record is delivered twice. - One gateway is also a throughput ceiling. ## Open question Where does group state live: in Iggy, or with group ownership by hash and FindCoordinator pointing at the owner? ## Depends on - #4513: producer state moves with partition leadership ## Scope - [ ] Shared group coordination, as the open question decides - [ ] Each gateway as a broker with a stable node id. One gateway leads each partition, ideally the one next to its Iggy primary - [ ] Producer state moves with partition leadership - [ ] Failover tests: a gateway stops during a rebalance, and during a burst of Produce requests ## Acceptance criteria - [ ] Two gateways behind one bootstrap list serve one consumer group, and each partition has one owner - [ ] When one gateway stops, its clients move to the other, and each group keeps its committed offsets -- 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]
