1503973 - MemoryTelemetry.jsm janks the browser

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[(no longer active)]

See this profile: https://perfht.ml/2QaHfst

The culprit is GetProcSelfSmapsPrivate at https://searchfox.org/mozilla-central/rev/39cb1e96cf97713c444c5a0404d4f84627aee85d/xpcom/base/nsMemoryReporterManager.cpp#90.

Pretty sure I've filed this before too but I can't find the bug any more...

Comment 1

[Daniel Holbert [:dholbert]]

The profile shows a 76ms event processing delay in the chrome process's main thread.

Looks like kmag touched this function in bug 1475899, though I'm not sure if that caused this or not.  kmag: do you have any thoughts on how this could be made better, and do you know if it might've been introduced by that bug or another recent change? e.g. is it possible for us to collect this data off-main thread somehow?

Comment 2

[Kris Maglione [:kmag]]

The majority of the overhead here is just reading /proc/self/smaps, which there isn't really anything we can do about, aside from moving it to a background thread.

Comment 3

[(no longer active)]

(In reply to Daniel Holbert [:dholbert] from comment #1)
> The profile shows a 76ms event processing delay in the chrome process's main
> thread.

FWIW the reason why I filed the bug was repeated janks of this nature.  It can take various amounts of time of course.  Don't pay a lot of attention to the 76ms part specifically.

Comment 4

[Chris H-C :chutten]

:kmag, what would be the appropriate component for memory reporting work?

More broadly: we'd be happy to have the collection code move out of toolkit/components/telemetry (it's only there for historical reasons), and we can help consult on any data collection questions you might have.

Comment 5

[Kris Maglione [:kmag]]

(In reply to Chris H-C :chutten from comment #4)
> :kmag, what would be the appropriate component for memory reporting work?

Probably Core :: XPCOM or Toolkit :: about:memory

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

I've seen this now several times happening during page load.
Adding [qf] so that we can re-triage this and possibly move to [qf:p1:pageload] 

http://bit.ly/2RonqP9