haohuaijin commented on PR #10141:
URL: https://github.com/apache/arrow-rs/pull/10141#issuecomment-4791423784

   I tried one more change in #10209: inline the selector-backed bodies of 
`scan_ranges` and `expand_to_batch_boundaries` directly into the `RowSelection` 
methods, so the selector path keeps the same code shape as `main`.
   
   I ran `arrow_reader_clickbench` three times on that branch. The results are 
not stable across runs:
   
   | case | Run 1 | Run 2 | Run 3 |
   |---|---|---|---|
   | `async/Q20` | **+25.8%** | −12.5% | −14.5% |
   | `async/Q21` | **+27.2%** | −16.7% | −19.1% |
   | `async/Q22` | **+16.4%** | −16.3% | −7.8% |
   | `sync/Q13` | +0.8% | **+16.6%** | **+16.7%** |
   | `async_object_store/Q23` | −1.9% | −9.7% | −4.2% |
   
   Same commit, same merge-base, but results swing 30–46 percentage points 
between runs. The `sync/Q13` regression only shows up in 2 of 3 runs, and the 
big `async/Q20/Q21/Q22` improvements only show up in 2 of 3 runs.
   
   Filtering to cases consistent across all three runs (spread ≤ 3pp, same 
sign), the remaining regressions are all small (≤ 2.2%):
   
   ```
   async_object_store/Q36   +2.2%
   async_object_store/Q38   +1.5%
   sync/Q12                 +1.5%
   async_object_store/Q20   +1.2%
   async_object_store/Q27   +1.1%
   sync/Q36                 +0.9%
   async/Q14                +0.6%
   ```
   
   No stable improvement across all three runs either.


-- 
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]

Reply via email to