xylaaaaa commented on issue #69884: URL: https://github.com/apache/doris/issues/69884#issuecomment-6658155938
The three deletion-vector pruning cases from the [coverage review](https://github.com/apache/doris/issues/69884#issuecomment-6654170183) are now permanent regressions in [PR #69904](https://github.com/apache/doris/pull/69904), [commit `a8954bdc70`](https://github.com/apache/doris/pull/69904/commits/a8954bdc70dd0d450e39dabe6201a6042544dfb5). The Spark fixture writes a unique 64-character payload containing the row's id. The regression compares the complete `(id, payload)` pairs on both sides of the pruning controls and checks for mismatches over the table. | Case | Result on both filesystem and HMS | | --- | --- | | Disjoint pages in one row group | For `id IN (5,200,201,204)`, only 201 and 204 returned with their own payloads. The profile showed two selected ranges in one row group. | | All matching rows deleted | The interval `[199,201)` returned zero rows. Its `COUNT(*), MIN(id), MAX(id)` were `0, NULL, NULL`. | | Page pruning across file splits | With a 32 KiB split size both plans produced five splits. Only 102 and 9006 survived out of `(101,102,9005,9006)`. Both profiles read two row groups and skipped 9,800 rows using the page index. | The basic and Parquet baselines passed on the regenerated data. The Doris runner generated the new golden, and the ordinary rerun passed all 32 result blocks. All 12 range profiles in each run met their assertions. Independent row-by-row checks matched all 9,990 live pairs in each catalog. The Parquet file still has 10,000 physical rows and the deletion vector has cardinality 10. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
