deepakpanda93 commented on issue #15010: URL: https://github.com/apache/hudi/issues/15010#issuecomment-4891377748
This is resolved by the MOR log-reader rewrite in 1.x. The bug depended on readBlocksLazily=false causing eager block merging that defeated ROLLBACK_PREVIOUS_BLOCK command blocks — but that lazy/eager toggle no longer exists (the old AbstractHoodieLogRecordReader was replaced by BaseHoodieLogRecordReader). Rollback application is now independent of read mode: during log scanning the reader collects each rollback command block's target instant time (targetRollbackInstants) and invalidates the corresponding data/compacted blocks deterministically, before any merge. It explicitly handles non-adjacent rollbacks (e.g. multi-writer B1,B2,B3,B4,R1(B3),B5) and stacked/compacted blocks — including the restore-driven sequence of rollbacks described here. So rollback blocks can no longer be rendered useless by eager merging. Closing as fixed on current 1.x. Please reopen with a reproducing case on a recent 1.x release if you still observe rolled-back MOR data resurfacing. -- 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]
