Warm Calendar Tab switch regresses against 4eb baseline #1576

Open
opened 2026-10-11 08:57:58 +00:00 by kayg · 0 comments
Owner

This issue tracks the warm Tab switch regression measured while completing #1557. The #1557 job recorded the result and stopped tuning after the acceptance rule failed.

Measurement

Five interleaved warm pairs on the perf VM, Chromium with macOS platform emulation, 1440 px, light theme, --smoke Home seed with the Calendar fixture retained. Baseline: 4ebeba0b8e13b06223bc2b24ee7a398a0933bda0. Candidate: 226b40228d982449292b0957b13e1d2c921c825a. Both used the shared 4eb server binary. Under-lock starting load average: 0.35 0.12 0.80.

All values are milliseconds; each value is p50 / p95.

Cell Baseline input→2 frames Candidate input→2 frames Baseline input→fully painted Candidate input→fully painted
Mail→Calendar 546.1 / 563.7 589.5 / 1593.5 842.5 / 862.6 1013.4 / 1943.5
Calendar→Files 408.4 / 533.7 399.0 / 499.3 808.4 / 967.0 751.6 / 987.4
Files→Calendar 567.3 / 649.0 730.4 / 826.2 932.3 / 1087.4 1098.1 / 1359.6

The #1557 acceptance rule required at least a 10% improvement in Mail→Calendar fully-painted p50 and no cell p95 regression over 10%. It failed: Mail→Calendar fully-painted p50 regressed 20.3%, fully-painted p95 regressed 125.3%, and two-frame p95 regressed 182.6%. Files→Calendar regressed in p50 and p95.

Trace evidence

Sampled self-time top entries (source frames are minified, so this does not establish cause):

  • Baseline: Pe 49.40 ms, scrollTo 44.00 ms, anonymous frame 10.56 ms.
  • Candidate: mt 98.90 ms, garbage collection 38.16 ms, w 22.58 ms.
  • Longest sampled RunTask: 138.86 ms baseline; 163.75 ms candidate.

Raw traces from the measured run:

Next work should identify the dominant client work from source-mapped traces and isolate the regression before proposing another change.

This issue tracks the warm Tab switch regression measured while completing #1557. The #1557 job recorded the result and stopped tuning after the acceptance rule failed. ## Measurement Five interleaved warm pairs on the perf VM, Chromium with macOS platform emulation, 1440 px, light theme, `--smoke` Home seed with the Calendar fixture retained. Baseline: `4ebeba0b8e13b06223bc2b24ee7a398a0933bda0`. Candidate: `226b40228d982449292b0957b13e1d2c921c825a`. Both used the shared 4eb server binary. Under-lock starting load average: `0.35 0.12 0.80`. All values are milliseconds; each value is p50 / p95. | Cell | Baseline input→2 frames | Candidate input→2 frames | Baseline input→fully painted | Candidate input→fully painted | | --- | ---: | ---: | ---: | ---: | | Mail→Calendar | 546.1 / 563.7 | 589.5 / 1593.5 | 842.5 / 862.6 | 1013.4 / 1943.5 | | Calendar→Files | 408.4 / 533.7 | 399.0 / 499.3 | 808.4 / 967.0 | 751.6 / 987.4 | | Files→Calendar | 567.3 / 649.0 | 730.4 / 826.2 | 932.3 / 1087.4 | 1098.1 / 1359.6 | The #1557 acceptance rule required at least a 10% improvement in Mail→Calendar fully-painted p50 and no cell p95 regression over 10%. It failed: Mail→Calendar fully-painted p50 regressed 20.3%, fully-painted p95 regressed 125.3%, and two-frame p95 regressed 182.6%. Files→Calendar regressed in p50 and p95. ## Trace evidence Sampled self-time top entries (source frames are minified, so this does not establish cause): - Baseline: `Pe` 49.40 ms, `scrollTo` 44.00 ms, anonymous frame 10.56 ms. - Candidate: `mt` 98.90 ms, garbage collection 38.16 ms, `w` 22.58 ms. - Longest sampled RunTask: 138.86 ms baseline; 163.75 ms candidate. Raw traces from the measured run: - [Baseline Mail→Calendar trace](https://git.kayg.org/attachments/6d98f45c-bddf-4f00-a640-e40a2d445698) - [Candidate Mail→Calendar trace](https://git.kayg.org/attachments/860695a8-7789-4521-a9a4-6d948b83a1c6) Next work should identify the dominant client work from source-mapped traces and isolate the regression before proposing another change.
Commenting is not possible because the repository is archived.
No labels
No milestone
No project
No assignees
1 participant
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/calternal-v0#1576
No description provided.