Rich-T-kid commented on PR #26348: URL: https://github.com/apache/datafusion/pull/26348#issuecomment-6194679560
Uncapped results for Arrow `c7f60c61 -> 55d9957a`. Both sides enable RLE dictionary reads; neither side sets an NDV cap. Lower is better, and the totals below sum the per-query means. | Suite | Base | PR | Delta | Faster / slower / unchanged | | --- | ---: | ---: | ---: | ---: | | [TPCH](https://github.com/apache/datafusion/pull/26348#issuecomment-6194334365) | 829.46 ms | 831.24 ms | 0.21% slower | 1 / 2 / 19 | | [TPC-DS](https://github.com/apache/datafusion/pull/26348#issuecomment-6194353344) | 9,170.83 ms | 9,099.71 ms | 0.78% faster | 1 / 1 / 97 | | [ClickBench partitioned](https://github.com/apache/datafusion/pull/26348#issuecomment-6194644208) | 76,787.19 ms | 76,149.35 ms | 0.83% faster | 1 / 1 / 41 | | [Wide schema](https://github.com/apache/datafusion/pull/26348#issuecomment-6194655342) | 4,176.8 ms | 4,160.1 ms | 0.40% faster | - | All queries completed with no failures. This is the relevant macro comparison for the reader change: the standard benchmark Parquet files lack NDV statistics, so the earlier cap of 2000 prevented dictionary promotion instead of selecting low-NDV dictionaries. -- 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]
