Aias00 opened a new issue, #7022: URL: https://github.com/apache/rocketmq-dashboard/issues/7022
## Evidence On rocketmq-studio c34535db0, AiEventSink removes an event from its queue and swallows ordinary persistence exceptions. A controlled failure of the answer INSERT followed by a successful run_status INSERT saved only run_status, while highWaterSeq advanced to 2 and close did not retry. #5667 tests current failure semantics but does not add visibility. ## Design and implementation Keep provider completion and cancellation semantics truthful; a successful model response is not itself a provider failure. Track whether an event failed persistence, emit a fixed non-secret warning to live observers, and carry a stable ai.timeline.persistence_failed diagnostic in run metadata when there is no more informative existing run error. Best-effort persist a warning at finalization when the database is writable so replay also explains incomplete history. Do not retry ambiguous INSERT failures, allocate duplicate terminal events, change schema, or claim recovery during a database outage. Existing provider/Stop terminal status and error precedence remain intact. ## Validation Fault-injection regressions for answer/terminal writes and serialization failures; provider error precedence, completed early-result frame and one terminal frame; Java21 conversation suite and Checkstyle. Document the best-effort durability boundary. -- 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]
