Closed Bug 1083952 Opened 5 years ago Closed 5 years ago

Intermittent leakcheck | default process: 340 bytes leaked (BackstagePass, ProtoAndIfaceCache, XPCNativeScriptableInfo, XPCNativeScriptableShared, XPCWrappedNative, ...)

Categories

(Core :: XPConnect, defect)

x86
Windows XP
defect
Not set

Tracking

()

RESOLVED WORKSFORME
Tracking Status
firefox35 --- affected
firefox36 --- unaffected

People

(Reporter: RyanVM, Unassigned)

References

Details

(Keywords: intermittent-failure, memory-leak)

https://treeherder.mozilla.org/ui/logviewer.html#?job_id=921747&repo=fx-team

builder 	Windows XP 32-bit fx-team debug test mochitest-3
buildid 	20141016073742
builduid 	9b4893cb7c0f4296821b527aa2034805
results 	warnings (1)
revision 	b0843f9cb54107dfe1fdb631ee6f926308c224ba
slave 		t-xp32-ix-110
starttime 	Thu Oct 16 2014 13:14:38 GMT-0400 (Eastern Standard Time)

10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 BackstagePass (40 bytes)
10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 ProtoAndIfaceCache (8 bytes)
10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 XPCNativeScriptableInfo (8 bytes)
10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 XPCNativeScriptableShared (196 bytes)
10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 XPCWrappedNative (48 bytes)
10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 XPCWrappedNativeProto (24 bytes)
10:37:44 INFO - TEST-INFO | leakcheck | default process: leaked 1 nsSystemPrincipal (16 bytes)
10:37:44 WARNING - TEST-UNEXPECTED-FAIL | leakcheck | default process: 340 bytes leaked (BackstagePass, ProtoAndIfaceCache, XPCNativeScriptableInfo, XPCNativeScriptableShared, XPCWrappedNative, ...)
Duplicate of this bug: 1085261
This is a pretty widespread bug. Can we get some eyes on it soonish please? :)
Flags: needinfo?(bobbyholley)
And suddenly these leaks appear on Aurora with today's round of uplifts starting with rev 8949cb909baf. Including the prior push, that gives us the following regression range:
https://hg.mozilla.org/releases/mozilla-aurora/pushloghtml?fromchange=d0081f5a668f&tochange=8949cb909baf
Given comment 15, sounds like we can pretty easily bisect down to a regressing changeset with try pushes. Can you do that and NI the author Ryan? I'm not going to have time to look at this for a while.
Flags: needinfo?(bobbyholley) → needinfo?(ryanvm)
Tried that yesterday and didn't get anywhere
Flags: needinfo?(ryanvm)
Andrew, any thoughts on this?
Flags: needinfo?(continuation)
See Also: → 1059771
(In reply to Bobby Holley (:bholley) from comment #44)
> Andrew, any thoughts on this?

There's a similar leak showing up in LSan, which includes the JS objects that are leaking.  I'm not really sure if this is XPConnect or JS.  But other than that, not really.
Flags: needinfo?(continuation)
Andrew - Do you have a chance to look at this at some point? Johnny suggested that you have a lot more tools and expertise than the rest of us to quickly identify these things.
Flags: needinfo?(continuation)
Somebody just needs to repro with MOZ_CC_LOG_SHUTDOWN=1 and look at the logs.  I can help interpret the logs as needed.

Or bisect against Aurora so we can at least see what caused it.
Flags: needinfo?(continuation)
Aurora retriggers are pointing at bug 1054959 being the culprit if that makes any sense.
https://hg.mozilla.org/releases/mozilla-aurora/rev/ed68a14922fe

Retriggering a bunch more runs on the push prior to confirm.
(In reply to Ryan VanderMeulen [:RyanVM UTC-4] from comment #117)
> Aurora retriggers are pointing at bug 1054959 being the culprit if that
> makes any sense.
> https://hg.mozilla.org/releases/mozilla-aurora/rev/ed68a14922fe
> 
> Retriggering a bunch more runs on the push prior to confirm.

The retriggers are strongly pointing to bug 1054959 being the culprit (50+ run w/o leaking on the push prior). Try runs of it backed out:
Trunk:  https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=4a346111abe2
Aurora: https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=ba3f5426333f
Blocks: 1054959
(In reply to Ryan VanderMeulen [:RyanVM UTC-4] from comment #125)
> Trunk: 
> https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=4a346111abe2
> Aurora:
> https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=ba3f5426333f

50 dt3 runs each. Solid green.
Brad/Gavin, this leak is too frequent to leave things as-is. We need a fix or backout ASAP. I'll do the backout myself if needed.
Flags: needinfo?(gavin.sharp)
Flags: needinfo?(blassey.bugs)
Thanks for tracking this down, Ryan!
See Also: → 1072931
(In reply to Ryan VanderMeulen [:RyanVM UTC-4] from comment #127)
> Brad/Gavin, this leak is too frequent to leave things as-is. We need a fix
> or backout ASAP. I'll do the backout myself if needed.

If you need to back this out, don't back out the strings (i.e. the change to browser/locales/en-US/chrome/browser/browser.dtd).
Flags: needinfo?(gavin.sharp)
I wonder if quickstubs removal (bug 1019191) somehow fixed this.  The timing seems to sort of match up.  That landed 10/22 at 8:41am, and the last star here is at 4pm that day, which given test and starring delay could line up.  Sadly that's not really backportable.
Bug 1054959 backed out from Aurora (with the dtd change left intact per Gavin's request). I'm going to leave this open for any future investigation that needs doing.

https://hg.mozilla.org/releases/mozilla-aurora/rev/ed94bc5a9ace
Flags: needinfo?(blassey.bugs)
If we wontfix bug 1054959 for Fx35, we can just resolve this as WONTFIX since trunk fixed itself.
Last star looks unrelated.
Status: NEW → RESOLVED
Closed: 5 years ago
Resolution: --- → WORKSFORME
Looks like whatever this is, is back.
Status: RESOLVED → REOPENED
Resolution: WORKSFORME → ---
(In reply to Andrew McCreight [:mccr8] from comment #148)
> Looks like whatever this is, is back.

I've got retriggers running to hopefully find where this started.
This was most likely bug 1071880, which is now backed out.
This started happening shortly after bug 1071880 relanded, so I assume that is at fault.
Inactive; closing (see bug 1180138).
Status: REOPENED → RESOLVED
Closed: 5 years ago5 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.