Closed Bug 1137404 Opened 5 years ago Closed 5 years ago

Intermittent talos xperf mainthreadio | File 'c:\windows\system32\en-us\dwrite.dll.mui' was accessed and we were not expecting it

Categories

(Firefox Health Report Graveyard :: Client: Desktop, defect)

x86
Windows 7
defect
Not set

Tracking

(firefox38 unaffected, firefox39 fixed, firefox40 fixed, firefox-esr31 unaffected)

RESOLVED FIXED
Firefox 40
Tracking Status
firefox38 --- unaffected
firefox39 --- fixed
firefox40 --- fixed
firefox-esr31 --- unaffected

People

(Reporter: aklotz, Assigned: Dexter)

References

(Depends on 1 open bug)

Details

(Keywords: intermittent-failure)

Attachments

(3 files, 1 obsolete file)

See https://bugzilla.mozilla.org/show_bug.cgi?id=1136351#c31. It sounds like some new code is accessing nsGfxInfo in a way that is triggering this I/O.

Probably added in bug 1122052.

We should look at nsGfxInfo to see what's going on there. Otherwise we'll have to adjust how we access this in FHR/telemetry.
:aklotz, we are almost to the limit where we need to hide xperf (i.e. it won't get sheriff coverage).  Can we push to get a fix for this, or should we temporarily adjust the mtio-whitelist.json until this can get resolved?

Historically we have been pretty diligent about not modifying the file unless needed, not sure we should be making exceptions here.
Flags: needinfo?(aklotz)
It's the DWriteVersion field that is causing the problem here. Is there anything that we can to do postpone gathering this field until a later time?
See comment 143.
Flags: needinfo?(alessio.placitelli)
(In reply to Aaron Klotz [:aklotz] (please use needinfo) from comment #143)
> It's the DWriteVersion field that is causing the problem here. Is there
> anything that we can to do postpone gathering this field until a later time?

Given [1], we shouldn't be accessing gfxData from TelemetryEnvironment (and so DWriteVersion) at startup, at least not before 60s (TELEMETRY_DELAY). Even after Telemetry starts up, we only gather gfx data if something triggers an important change in the environment (new theme, new plugin, new addon, etc.). Are these tests triggering any of those changes?

[1] https://hg.mozilla.org/mozilla-central/annotate/0b3c520002ad/toolkit/components/telemetry/TelemetryPing.jsm#l712
Flags: needinfo?(alessio.placitelli)
we already had this in the file, our access times have gone up, almost all of them are <60ms.
Assignee: nobody → jmaher
Status: NEW → ASSIGNED
Flags: needinfo?(aklotz)
Attachment #8572737 - Flags: review?(vdjeric)
Comment on attachment 8572737 [details] [diff] [review]
ignore duration of dwrite.dll.mui up to 60ms (1.0)

Review of attachment 8572737 [details] [diff] [review]:
-----------------------------------------------------------------

r+ for a temporary fix until i figure out why DWrite version is being fetched earlier in the unified Telemetry/FHR
Attachment #8572737 - Flags: review?(vdjeric) → review+