litiliu commented on PR #4608: URL: https://github.com/apache/flink-cdc/pull/4608#issuecomment-6789665118
Thanks for the review, @loserwang1024 and @fxbing. I've updated the PR and squashed the changes into `fdcdbac40`. [Incremental diff from the reviewed `7fb412e`](https://github.com/apache/flink-cdc/compare/7fb412e15c8079dfe99ea70bca1b38fe181f6253..fdcdbac408b8b883541974396bb0368908999c08) @loserwang1024, the [mixed initialization failure](https://github.com/apache/flink-cdc/pull/4608#pullrequestreview-5822732714) now retries only the surviving buckets when part of the batch has been removed. A fully retired batch is ignored; a failure with all buckets still active remains fatal. The new full-mode regression follows your delayed initialization sequence and verifies that only `retained` is assigned, without duplicate assignment. @fxbing, I've replied to all five inline comments with the changes and regression coverage. For the subscription race, I used fresh Admin metadata confirmation on the fetcher rather than triggering Enumerator discovery. The reasoning and reconciliation tradeoff are described [in this reply](https://github.com/apache/flink-cdc/pull/4608#discussion_r4535027200). Validation passed on both Flink 1.20.3 and 2.2.0: 38 Enumerator cases, 15 SourceReader cases, 7 SDK integration cases, and 4 checkpoint-restore cases. The restore cases cover earliest/full startup and partition deletion while running or after checkpoint-and-cancel; they verify restoration from the exact checkpoint, no replay of retained data, and continued checkpointing. After the final batch-local confirmation cache change, all 23 SplitReader unit cases were rerun and passed on each Flink version. Could you please take another look when you have a chance? In particular, I'd appreciate your feedback on the local confirmation 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]
