DanielLeens commented on issue #12212: URL: https://github.com/apache/seatunnel/issues/12212#issuecomment-5602963790
Thank you for the native binlog and checkpoint evidence. It materially narrows the investigation: `mysqlbinlog -vv` can decode the stored transaction through COMMIT, so the report does not justify treating the persisted binlog bytes as deterministically malformed. The row-decoder EOF and the recovery-position question must remain separate hypotheses. The current reader code also leaves a real ordering question to test. `SourceReaderBase.snapshotState()` serializes the in-memory split states directly, while `SplitFetcherManager.checkErrors()` is called when `getNextFetch()` begins. That means we cannot dismiss the reported checkpoint window from source reading alone, but the logs do not yet prove that checkpoint 15279 contained an offset beyond undelivered records. Please do not submit a parser-only reconnect patch from the current evidence. The next reproducible slice should use a deterministic fetcher/reader harness that publishes a batch, advances its candidate offset, fails before the reader hands off the batch, and races `snapshotState()`. It must assert both the accepted or rejected checkpoint behavior and the restored offset. For the production incident, please also provide the redacted source split state from checkpoint 15279 and the first restored-offset log after recovery. That will distinguish a reader/checkpoint ordering defect from a sink or deployment-specific observation gap. -- 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]
