Kalvin2077 commented on PR #3858: URL: https://github.com/apache/celeborn/pull/3858#issuecomment-6676759465
> @Kalvin2077, thanks for updates. I left one comment for stale observations when a former leader becomes leader again while the worker stays registered. PTAL. @SteNicholas I also found that overlapping heartbeats within one term could publish observations in completion order. Each admission now replaces the update reference while retaining the published value. An earlier request can therefore neither overwrite a newer observation nor clear it with missing metrics. A regression pauses the earlier heartbeat, publishes a later observation, and then releases the earlier request. Additional tests use a three-master Ratis cluster to cover Raft submission failure, collector read failure, worker re-registration, leader shutdown, and restart from the original Raft storage. They verify that failed writes do not publish incoming metrics, read failures preserve liveness, re-registration does not reinitialize the collector, and restarted masters restore membership with empty observations. The focused suites, Spotless, and Apache Rat passed locally. -- 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]
