1588162 - Investigate possible ~13% regression on sessionrestore Talos test on win32 shippable with Fission enabled

Status: RESOLVED WORKSFORME

(Firefox :: Session Restore, task)

Description

[Mike Conley (:mconley) (:⚙️)]

https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=d892263b6c1036f9801560772722bc502d610e24&newProject=try&newRevision=c5acc86520570100fad8c86be18aedfe7003878d&framework=1

Comment 1

[Mike Conley (:mconley) (:⚙️)]

Hey dthayer,

I was looking at bug 1588165 the other day, and I think what I found might be related to this. It's not the same issue (the remoteness flipping not being reported for SSTabRestored), but I suspect it's that we're doing remoteness flipping where we weren't before, which is causing the parent process to do more work, which is eating up the 13%.

Anyhow, that was mainly speculation, but it might be worth looking at.

Comment 2

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

(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #1)

Hey dthayer,

I was looking at bug 1588165 the other day, and I think what I found might be related to this. It's not the same issue (the remoteness flipping not being reported for SSTabRestored), but I suspect it's that we're doing remoteness flipping where we weren't before, which is causing the parent process to do more work, which is eating up the 13%.

Anyhow, that was mainly speculation, but it might be worth looking at.

I don't see any indication of a remoteness flip? No calls to updateBrowserRemoteness or anything - for the sessionrestore test these are all lazy browsers anyway, so they shouldn't be backed by an actual process to begin with.

Comment 3

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

The delta here seems to have gone away. This is a comparison from a more recent build:

https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=69bf5e03a79885320fdf9d07b1abf0b82adfb0fd&newProject=try&newRevision=8de664c7782463b6f289054fb8069170472538a9&framework=1

Going to close it out for now.