JingsongLi commented on PR #10187: URL: https://github.com/apache/paimon/pull/10187#issuecomment-5951614767
The load/repair change has clear end-to-end value, and I did not find an introduced implementation regression in the reviewed diff. Validation at 459325c416: - All 52 `IcebergCompatibilityTest` cases pass with normal Maven checks, including the new legacy-load and retry-publication cases, historical type rejection, mirror enablement, and ordinary Iceberg reads. - An additional persisted Parquet TIMESTAMP(9) table preserves `2024-01-02 03:04:05.123456789` through legacy-schema loading and `copyWithLatestSchema`. A commit with the mirror still enabled is rejected before publishing a replacement snapshot or Iceberg metadata. After ALTER disables the mirror and the table/writer are reloaded, another actual commit and read succeed without losing the nanoseconds. - Independent checks of the publication paths confirm that normal commits and rollback-as-latest retain their pre-publication guards, the new retry guard runs before metadata/pointer repair, and tag callbacks retain existing metadata schemas rather than publishing newly loaded schema history. The Spark 4 CI failure still needs a green completed check. Its recorded failure is `ShuffleStatusNotFoundException` inside the test helper `RemoveOrphanBlobsProcedureTest.clearShuffleOutputs`, for a shuffle that no longer exists. I ran that test class locally on Spark 4.1.2 and it passed, including the failing CI case; this does not turn the remote failed run into a successful result. Please rerun/resolve it before merging. For rollout, verify the repair using a freshly loaded table and recreated writer/job after disabling the mirror. Existing handles retain their old options. As the PR description states, this restores Paimon read/repair access; it does not repair previously emitted Iceberg metadata or external catalog pointers. -- 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]
