1215148 - Object-count based leak checking for Mochitest

Status: RESOLVED FIXED

(Testing :: Mochitest, defect)

Description

[Andrew McCreight [:mccr8]]

We leak a bunch of objects in content processes. Right now, we whitelist this by having a leak threshold. Unfortunately, this is not a very tight bound, because object sizes vary, etc., so you can end up leaking an entire extra IPC Channel without triggering the detector.

This could be improved by having a specific list of objects and the number of them that we allow to be leaked. It would also be a good excuse to account for these specific objects to make sure they are all coming from ImageBridgeChild and CompositorChild, and that there aren't some other leaks lurking like bug 1213407.

The current list is 1 AsyncTransactionTrackersHolder, 1 CompositorChild, 3 CondVar, 2 IPC::Channel, 1 MessagePump, 3 Mutex, 1 PCompositorChild, 1 PImageBridgeChild, 2 RefCountedMonitor, 4 RefCountedTask, 2 StoreRef, 1 WaitableEventKernel, 2 WeakReference<MessageListener>, 2 ipc::MessageChannel, 9 nsTArray_base, 1 nsThread.

Comment 1

[Andrew McCreight [:mccr8]]

At least for a simple Mochitest run, I confirmed that hackily shutting down ImageBridgeChild and CompositorChild results in zero leaks in the child process.

Comment 2

[Andrew McCreight [:mccr8]]

Attachment: Hacky ImageBridgeChild and CompositorChild shutdown. DO NOT LAND

Comment 3

[Andrew McCreight [:mccr8]]

Attachment: Add finer-grained leak checking for content processes in Mochitest. WIP

Comment 4

[Andrew McCreight [:mccr8]]

try run looks good: https://treeherder.mozilla.org/#/jobs?repo=try&revision=11728e8cdcb8

Comment 5

[Andrew McCreight [:mccr8]]

Attachment: Object-count based leak checking for Mochitest.

I have patches for the last two things leaks in here, so I may remove those before landing depending on how long it takes to get those reviewed and landed.

Comment 6

[Jonathan Griffin (:jgriffin)]

Comment on attachment 8676973 [details] [diff] [review]
Object-count based leak checking for Mochitest.

Review of attachment 8676973 [details] [diff] [review]:
-----------------------------------------------------------------

lgtm

Comment 7

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f0134eb0b78f

Comment 8

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/f0134eb0b78f