deepakpanda93 commented on issue #15677: URL: https://github.com/apache/hudi/issues/15677#issuecomment-4892131420
This issue was reviewed as part of the JIRA-migrated backlog triage. This is fixed as of 0.12.3 (and in all later releases, including 1.x), via HUDI-5553 / #7669 "Prevent partition(s) from being dropped if there are pending table service actions." Root cause: dropping a partition whose file groups are inputs to a pending clustering/compaction could interleave the drop-partition replacecommit with the table-service commit, leaving inconsistent state - which for the CONSISTENT_HASHING bucket index (whose bucket resizing runs as clustering) showed up as read inconsistency after ALTER TABLE DROP PARTITION. The fix adds DeletePartitionUtils.checkForPendingTableServiceActions (called from SparkDeletePartitionCommitActionExecutor), which now blocks the drop with HoodieDeletePartitionPendingTableServiceException if the target partition has pending clustering/compaction, so the unsafe interleaving can't happen. The original report (#7634) was already closed as fixed on this basis. (Note: this issue's status:pr-available label points at #7645, an alternative "purge hashing metadata" approach that was not merged - the shipped fix is #7669.) Closing as fixed in 0.12.3+/1.x. Please reopen if you still see consistent-hashing read inconsistency after drop-partition on a recent 1.x release. Thanks! -- 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]
