1586754 - Investigate possible ~4% regression on startup_about_home_paint_realworld_webextensions Talos test with Fission enabled on macOS

Status: RESOLVED WORKSFORME

(WebExtensions :: General, task)

Description

[Mike Conley (:mconley) (:⚙️)]

Here's the comparison link:

https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=d560df856665a4d4f332c235eea5ce8c6733be0a&newProject=try&newRevision=b3612e3ef1bb587afa79b5635b296b498debfa9a&framework=1

The probability of there being a real regression is apparently "low", but I think it's probably worth double-checking what's going on here.

Comment 1

[Dave Townsend [:mossop]]

Sounds likely to be webextensions?

Comment 2

[Alex Thayer [:alexical] (she/her)]

Attachment: Screen Shot 2019-10-28 at 9.39.05 AM.png

While the tool suggests the probability of a difference is "low", putting the runs side by side and eyeballing them it does look like there's something there.

Comment 3

[Alex Thayer [:alexical] (she/her)]

While the effect seems to be "real" to some extent, I have been digging at this and both through testing and reasoning through the code I can't identify any code paths which differ during the test window. So it could very well be something silly and random in a systemic way? We are adding an extra pref into the list, so maybe it's a memory layout quirk.

Comment 4

[Alex Thayer [:alexical] (she/her)]

Reran the test after rebasing to central, and the effect disappeared. Closing this out.