search

brave / brave-browser

Brave browser for Android, iOS, Linux, macOS, Windows.
https://brave.com
Mozilla Public License 2.0
18k stars 2.36k forks source link

[Perf] regression in timeToFirstMeaningfulPaint loading.desktop.brave after cr128 #40420

Open atuchin-m opened 3 months ago

atuchin-m commented 3 months ago 
Alert information:
Median of segment before: 33.773
Median of segment after: 61.905
Relative change: 83.3%
Test: BravePerf/mac-mini-m1-typical/loading.desktop.brave/timeToFirstMeaningfulPaint/cold/example.com_cold
image
atuchin-m commented 3 months ago

compare build to confirm the regression: https://ci.brave.com/job/brave-browser-test-perf-macos/1920/

atuchin-m commented 3 months ago
atuchin-m commented 3 months ago

Here is who it looks in traces when the browser load example.com before (good): image

after (bad): the same events, but extra gap after:

Screenshot 2024-08-14 at 3 13 15 PM
atuchin-m commented 3 months ago

It sounds like some changes in Compositor. I've tried to cherry-pick https://chromium-review.googlesource.com/c/chromium/src/+/5734612, but it doesn't give any effect.

atuchin-m commented 3 months ago

checking the regression on example.com on system_health scenario doesn't confirm it:

image

Checking the real browser installation also doesn't show any significant differences for FirstPaint metrics.

looks like it's:

  1. only specific for some navigations..
  2. chromium also has it.
atuchin-m commented 3 months ago

The issue happens only for navigations from a special pages (like about:blank) to a regular one in a same renderer. Regular navigations look good if they target to another renderer. For the same render it doesn't look good:

kjozwiak commented 1 week ago

@atuchin-m can we close this one off? Believe the above was found during C128 and we're now on C131 and about to merge C132 into master once QA completed our verifications.