Rich-T-kid commented on PR #27312: URL: https://github.com/apache/datafusion/pull/27312#issuecomment-6778627031
Same-head RLE off vs. on results Both sides use `c45551c04a444cfb65e59c74cf5cf409431a0e9d`; only `datafusion.execution.parquet.enable_rle_to_dictionary` differs. The lists below use the benchmark tool's primary per-query classification, not unclassified raw timing differences. | Suite | RLE off | RLE on | Total change | Regressing queries | | --- | ---: | ---: | ---: | --- | | [TPC-H](https://github.com/apache/datafusion/pull/27312#issuecomment-6778449539) | 700.34 ms | 728.87 ms | **4.07% slower** | Q10, Q12, Q13, Q18, Q20, Q22 | | [TPC-DS](https://github.com/apache/datafusion/pull/27312#issuecomment-6778482101) | 8,474.54 ms | 8,441.51 ms | **0.39% faster** | Q4, Q11, Q27, Q30, Q37, Q41, Q68, Q74, Q81, Q82 | | [Partitioned ClickBench](https://github.com/apache/datafusion/pull/27312#issuecomment-6778572935) | 18,631.31 ms | 28,395.63 ms | **52.41% slower** | Q5, Q12-Q18, Q20-Q27, Q30-Q34, Q36-Q39 | | **Combined** | **27,806.19 ms** | **37,566.01 ms** | **35.10% slower** | **41 query regressions** | All three suites completed without query failures. This does not measure the Arrow fix versus its parent and does not attribute the remaining regressions to individual operators; it isolates the end-to-end effect of enabling RLE dictionary reads after the fix. -- 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]
