adriangb commented on PR #25713: URL: https://github.com/apache/datafusion/pull/25713#issuecomment-5821816529
## Benchmark summary (83d19600b4) Base is merge-base 95bb0a0, the same base as the [#25677 results](https://github.com/apache/datafusion/pull/25677#issuecomment-5809027081). The measured change is therefore #25677 + split. Times are the minimum of 5 iterations. Change = PR / base - 1. ### Totals | Suite | Config | Result | Base | PR | Change | #25677 change | |---|---|---|---|---|---|---| | TPC-H SF1 | default | [link](https://github.com/apache/datafusion/pull/25713#issuecomment-5821758448) | 734.37 ms | 727.27 ms | -1.0% | -0.4% | | TPC-DS SF1 | default | [link](https://github.com/apache/datafusion/pull/25713#issuecomment-5821787004) | 9138.20 ms | 9022.41 ms | -1.3% | +1.7% | | TPC-H SF1 | `pushdown_filters=true` | [link](https://github.com/apache/datafusion/pull/25713#issuecomment-5821761074) | 847.72 ms | 815.06 ms | -3.9% | -4.1% | | TPC-H SF10 | `pushdown_filters=true` | [link](https://github.com/apache/datafusion/pull/25713#issuecomment-5821783995) | 8012.40 ms | 7903.48 ms | -1.4% | -3.2% | | TPC-DS SF1 | `pushdown_filters=true` | [link](https://github.com/apache/datafusion/pull/25713#issuecomment-5821775779) | 7826.07 ms | 7319.96 ms | **-6.5%** | -7.4% | ### Queries beyond ~7% | Suite | Config | Query | Base | PR | Change | #25677 change | Cause | |---|---|---|---|---|---|---|---| | TPC-DS SF1 | default | Q6 | 35.59 ms | 38.42 ms | +8.0% | +5.4% | noise (mean +6%) | | TPC-DS SF1 | default | Q34 | 10.75 ms | 9.57 ms | -11.0% | +1.5% | split or noise (mean -7%, pushdown run flat) | | TPC-DS SF1 | default | Q65 | 133.17 ms | 67.72 ms | **-49.1%** | -45.9% | #25677 | | TPC-DS SF1 | pushdown | Q17 | 114.94 ms | 68.54 ms | **-40.4%** | -36.0% | #25677 | | TPC-DS SF1 | pushdown | Q18 | 161.18 ms | 106.79 ms | **-33.7%** | -34.2% | #25677 | | TPC-DS SF1 | pushdown | Q25 | 115.21 ms | 66.47 ms | **-42.3%** | -37.7% | #25677 | | TPC-DS SF1 | pushdown | Q29 | 145.75 ms | 92.97 ms | **-36.2%** | -39.6% | #25677 | | TPC-DS SF1 | pushdown | Q50 | 124.75 ms | 106.40 ms | **-14.7%** | +1.6% | split | | TPC-DS SF1 | pushdown | Q64 | 510.41 ms | 337.43 ms | **-33.9%** | -33.1% | #25677 | | TPC-DS SF1 | pushdown | Q65 | 159.03 ms | 88.47 ms | **-44.4%** | -45.4% | #25677 | | TPC-DS SF1 | pushdown | Q85 | 164.45 ms | 128.29 ms | **-22.0%** | -3.1% | split | | TPC-DS SF1 | pushdown | Q95 | 92.42 ms | 55.75 ms | **-39.7%** | -41.4% | #25677 | | TPC-H SF1 | pushdown | Q17 | 71.08 ms | 44.62 ms | **-37.2%** | -39.5% | #25677 | | TPC-H SF1 | pushdown | Q18 | 74.25 ms | 66.47 ms | -10.5% | -10.5% | #25677 | | TPC-H SF10 | pushdown | Q4 | 296.72 ms | 321.48 ms | **+8.3%** | -1.3% | split (spread ±3-4 ms, not noise) | | TPC-H SF10 | pushdown | Q18 | 666.71 ms | 431.64 ms | **-35.3%** | -35.9% | #25677 | The bot cannot run `hj_ordered_subset`: it runs `bench.sh` from `main`, and the merge base has no suite files. **Conclusion:** the split adds TPC-DS Q50 (-15%) and Q85 (-22%) with pushdown on and keeps all #25677 gains, but TPC-H SF10 Q4 with pushdown on is 8% slower; suite totals are within noise of #25677. 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- 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]
