lhotari commented on PR #25044: URL: https://github.com/apache/pulsar/pull/25044#issuecomment-3625312028
> It seems that there might be a bug in replicated subscriptions related to the snapshot cache. The position used for comparison in the snapshot cache is the position of the marker message, not the snapshot request position. I think it should be the snapshot request position, since that position is the last position guaranteed to be in sync, as long as new messages are produced in only one cluster at a time. In the current solution, since the position used is the marker message position from the last snapshot response, the acknowledged position will be incorrect if new messages continue to be produced while snapshotting is happening but not all messages are acknowledged. It should be possible to reproduce this bug in a test. Actually it should be fine, since the remote cluster will only acknowledge up to the message id of the snapshot. However, the assumption of replicated subscription is that one cluster is active and producing messages at a time. If there are multiple active clusters, it seems that there could be corner cases where messages would get skipped. -- 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]
