Closed Bug 620600 Opened 14 years ago Closed 14 years ago

Leaking virtual address space/threads during media mochitest run

Categories

(Core :: Audio/Video, defect)

defect
Not set
normal

Tracking

()

VERIFIED FIXED

People

(Reporter: cpearce, Unassigned)

References

Details

When I run media mochitests on Linux x64 and Win7 x64, I end up with lots of stray threads lying around, and ever increasing virtual address space usage. If I run test_playback in a loop in multiple threads, I soon run out of address space, and can't allocate any new thread stacks, and (on Linux) fork() starts to fail everywhere.

I've tried test_playback limited to only WAV files and limited to only Ogg/WebM files, and this still happens, so either the leak exists in both the nsWavDecoder and nsBuiltinDecoder, or it's in some other service used by both of them.

The leaked threads have pretty standard nsIThread event queue stacks.
So are we leaking refcounted nsThread objects in this case?  If so, is it worth trying to see who's holding those references?
XPCOM_MEM_LEAK_LOG is not picking up leaked nsThread objects at shutdown, some idle and now unused threads aren't being shutdown during the mochitest run.

I'd already started sprinkling assertions to check that the references are cleared, but no hits yet...
Fixed by bug 621430.
Status: NEW → RESOLVED
Closed: 14 years ago
Depends on: 621430
Resolution: --- → FIXED
Is there a way I can verify this bug?
It's the same bug as bug 621430, which is already marked verified.
Status: RESOLVED → VERIFIED
You need to log in before you can comment on or make changes to this bug.