SEZ9 commented on PR #11206: URL: https://github.com/apache/seatunnel/pull/11206#issuecomment-6870567365
Thanks for the update. Here is where things stand against the earlier review items: **Merge conflict.** GitHub now reports `mergeable_state=dirty` for head `93afd1a996b26e84b8ab843d31c736aa20ae3e9c` against `dev` (diverged, ahead 32 / behind 36). @davidzollo mentions a resolution is prepared locally that keeps the upstream parallel flush implementation and snapshot-only restore logic alongside this PR's runtime registration and restore gate — please push that so we can review the resolved head. Since the restore gate touches PR11206-F6 (old-checkpoint restore / rollback of `MultiTableState`), please also describe in the PR how a pre-upgrade checkpoint is handled after the rebase. **PR11206-F5 (retry/idempotency of runtime writer creation).** The blocker @davidzollo found during the conflict audit is a concrete instance of this finding: if registration succeeds for the first queue and then exhausts retries for a later queue under CONTINUE_OTHER_TABLES, failure handling calls `removeTableWriters` while the other queue workers still hold their monitors and are waiting on the schema-change barrier, so the dispatcher and those workers can end up waiting on each other. Please defer isolation until after the barrier releases and add a test for partial registration with more than one queue. The related case — timer registration when the first flush-capable child is created at runtime — should be fixed and covered in the same change. **PR11206-F2 (sinks without newly-created-table support).** The JdbcSinkWriterTest failures remain because the runtime-create fixtures use a dialect mock without schema-change support enabled. Please fix the fixtures, and while there, make the contract explicit in `SupportMultiTableSinkWriter` for what happens when a runtime `CreateTableEvent` reaches a writer whose `supportsNewlyCreatedTable()` is false (fail fast vs. skip), with a test for that path. **PR11206-F1, F3, F4, F7, F8.** No new information this round; they stay open as previously described. As @davidzollo noted, these come from source reading with no new runtime validation, so a full run on the resolved head is still needed. Concrete asks: push the conflict resolution, the deadlock fix plus multi-queue partial-registration test, the timer-registration fix, the fixture fix with the F2 contract, and the F6 restore note. <!-- streview-comment:2335 --> -- 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]
