wenjin272 commented on issue #1302: URL: https://github.com/apache/flink-agents/issues/1302#issuecomment-6491010794
Thanks for the clarification @da-daken. The C example was meant to explain why retaining C’s saved result would be unsafe once its preceding history changes. I agree that the current suffix invalidation protects against this, at the cost of re-execution. Regarding the proposed direction, I have a couple of concerns: - How would `originId` and `originSeq` be defined and remain stable across recovery and rescaling? If they are tied to upstream subtask identities or arrival order, the same logical input could receive different values after recovery. - For an unbounded stream, how would the merge decide when it is safe to process the next input? Ordering metadata alone does not tell us whether an earlier input is still on its way from another source. Waiting could block progress indefinitely, while skipping an idle source based on a timeout could make the order differ across replays. The main purpose of this issue is to record the known limitation, which is why I marked it as minor. I’d prefer to revisit the solution after ActionState has been further hardened through production use. Its recovery model may evolve during that process, potentially changing the assumptions behind a design proposed today. For now, I think we can keep this issue open to track the limitation without committing to a particular approach. -- 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]
