SEZ9 commented on PR #13276: URL: https://github.com/apache/seatunnel/pull/13276#issuecomment-6965430461
@alexey-bass thanks for the follow-up and for posting the fork run for `6649ff5`. Two of the earlier points look resolved in the current code: - The decoder fallback catches `Exception` instead of `Throwable`, so fatal errors propagate and no longer trigger a second allocation-heavy attempt (Issue 1). - The fast path rejects a non-positive `createTime` and falls back to full deserialization. You also mention that the regression fixture was reduced to 17 MiB while still crossing the 16 MiB boundary; thanks for lowering the memory cost. One item remains, and it is blocking for test quality (Issue 2). It was not called out in the earlier review round, so apologies for the late notice. The new boundary tests in `JobInfoServiceNullSafetyTest` invoke private helpers through reflection. Please keep the public summary test, which is valuable, and rework the other cases in one of two ways: - drive them through the public summary path with a controlled full-decoder collaborator, or - make the decoder package-private and assert its actual contract without reflection. On CI, your request to approve the "PR signal" workflow run is noted, so the Build check can attach to the PR. Run 43632918057 is a useful reference in the meantime. For the three failing jobs (`jdbc-connectors-it-part-7` on 11 and 17, `all-connectors-it-5` on 17), your description points to container setup (`JdbcIrisIT` `Connection refused`, `TiDBCDCIT` `Communications link failure`) rather than this change. Could you link the `dev` build where the same two `jdbc-connectors-it-part-7` jobs fail, so that is easy to confirm? <!-- streview-comment:2425 --> -- 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]
