DanielLeens commented on issue #13253: URL: https://github.com/apache/seatunnel/issues/13253#issuecomment-6823246413
Thanks for the precise reproduction and for tracing it back to the shared enumerator. I confirmed the current `SingleSplitEnumerator` path has both conditions described here: `addSplitsBack` unconditionally reads `splits.get(0)`, and a returned non-empty split leaves `assigned` true, so the later assignment path is skipped. This is a connector-common recovery defect; please keep the fix scoped there rather than coupling it to the GraphQL behavior in #13222. A focused PR is welcome. Please include a deterministic regression in the connector-common test area that proves all of the following: 1. a returned single split clears the old assignment state and is assigned again when the restarted reader registers; 2. an empty `addSplitsBack` call is harmless and does not create a second assignment or overwrite a pending returned split; 3. the normal first assignment is unchanged. The test should exercise the enumerator through its context/reader-registration contract, rather than asserting only the flag value. Please also retain the existing GraphQL E2E restart suppression in #13222 as its own failure-reporting choice; this fix must stand on its own and cover the shared source path. Once the focused PR is open, link it here and we can review the current head and CI evidence. -- 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]
