1360214 - SanityTest.js is expensive during startup

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Florian Quèze [:florian]]

See this profile https://perfht.ml/2qj0y5z where SanityTest.js costs us 144ms (out of 2.2s) during startup before the first browser window is shown.

Can SanityTest.js avoid doing main thread I/O during startup? (the mozilla::Preferences::SavePrefFile call is expensive)

Could some of this run on a different process?

Comment 1

[Florian Quèze [:florian]]

The second savePrefFile call at http://searchfox.org/mozilla-central/rev/3dc6ceb42746ab40f1441e1e659ffb8f62ae78e3/toolkit/components/gfx/SanityTest.js#65 during reportResult costs us 132ms of UI jank in this profile: https://perfht.ml/2p1pMFw (and the first savePrefFile before the first window is shown costs us 128ms of startup time... but I believe that first flush is to ensure the information has reached the disk if we crash while testing gfx stuff. I can't see a justification for the second flush though).

Comment 2

[(no longer active)]

Bug 1361262 has some discussion of SavePrefFile() as well.

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

Florian, do you have time to build with patches attached to bug 789945 and re-run this test?

Comment 4

[Florian Quèze [:florian]]

(In reply to Milan Sreckovic [:milan] from comment #3)
> Florian, do you have time to build with patches attached to bug 789945 and
> re-run this test?

If you give me a try server build, I can profile it next week. I'm not going to try building on the reference hardware.

Comment 5

[Milan Sreckovic [:milan] (needinfo for best results)]

That would be amazing:  https://treeherder.mozilla.org/#/jobs?repo=try&revision=1b3f9b49921dff1a8eebfdb27e7e63bff28ccbc2

Comment 6

[Mike Conley (:mconley) (:⚙️)]

ni? to florian now that try build has been provided (comment 5).

Comment 7

[(no longer active)]

Florian: ping?

Comment 8

[Milan Sreckovic [:milan] (needinfo for best results)]

Probably not worth profiling on that try build at this point, the code has changed enough.  Let me do another build with up to date patches.

Comment 9

[Milan Sreckovic [:milan] (needinfo for best results)]

This build should be fine for testing: https://treeherder.mozilla.org/#/jobs?repo=try&revision=59aa8545e836cbd171fe830825a7d9b199d6c6cb

A preference "preferences.allow.omt-write", true by default, can be set to false to force back the original behaviour.  With the default true, we should have async/off main thread preference saves in majority of the cases.

Comment 10

[Florian Quèze [:florian]]

Here is a profile of the first startup (so running SanityTest.js) of the try build from comment 9: https://perfht.ml/2qI7Osr

Unfortunately without symbols the profile isn't very useful to look at non-JS code.

Comment 11

[Milan Sreckovic [:milan] (needinfo for best results)]

Patches that landed in bug 789945 should take care of this.