DanielLeens commented on issue #10203: URL: https://github.com/apache/seatunnel/issues/10203#issuecomment-6367122733
Thank you for the transition table. It makes the recovery boundary much clearer: the enumerator is the only registry writer, a table joins the incremental filter only with its `W_high` in a completed checkpoint, and the cut-2 replay limitation is now stated honestly. Classification: D / Connector-V2 MySQL-CDC live table discovery and exactly-once recovery design. There is one sequencing update: #12567 was closed without merging, so its per-table restore-watermark change is not a dependency we can rely on. #11206 is the active overlapping live-created-table path and still has review items outstanding. Please keep this work as design only until that boundary is resolved; do not open a parallel implementation that changes the same admission/recovery paths. For the next design revision, please make the acceptance contract explicit for (1) atomic checkpoint persistence of `pendingSnapshotTables`, `admittedTables`, and `tableWatermarks`; (2) dispatching an admission snapshot only to the reader that owns the live stream, including reassignment; and (3) config-time rejection when the feature is enabled without exactly-once. The test plan should cover discovery-off compatibility, every crash cut with deterministic checkpoint boundaries, and the one-client `STREAMING -> CLOSING -> SNAPSHOT -> BACKFILL -> STREAMING` state machine. Keep MySQL and OceanBase support claims separate. Once #11206 is settled and those invariants are captured in an agreed design, a focused implementation PR with the corresponding deterministic tests will be welcome. -- 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]
