When you run this: ./mach mochitest-plain --e10s --auto-close dom/media/tests/mochitest/test_dataChannel_basicDataOnly.html The leak includes 1 DataChannelShutdown. (The rest of the stuff in there is normal for e10s).

This still happens, though you have to run ./mach mochitest -f plain --e10s dom/media/tests/mochitest/test_dataChannel_basicDataOnly.html and then manually close it

Another way to trigger this leak is with ./mach mochitest -f plain dom/media/tests/mochitest/ipc/test_ipc.html

It looks like this is supposed to be deleted in DataChannelShutdown::Observe(), but it is trying to observe profile-change-net-teardown, which, if I recall correctly, does not happen in the child process.

Does it needs to listen to "content-child-shutdown" instead? I wonder how much code still adds observers that never fire due to the arrival of e10s. It'd be interesting to instrument addObserver to warn/assert every time an observer is added in the wrong process.

As far as I can tell, there's no need for the global: the observer service holds a strong reference and nobody else looks at the global, except to clear it in shutdown. This fixes the leak, but of course does not fix the issue of this code never running.

(In reply to Jan-Ivar Bruaroey [:jib] from comment #4) > I wonder how much code still adds observers that never fire due to the > arrival of e10s. It'd be interesting to instrument addObserver to > warn/assert every time an observer is added in the wrong process. That's a good idea! I filed bug 1216252.

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/12a6564962d7 https://hg.mozilla.org/integration/mozilla-inbound/rev/ca1313dd7411

https://hg.mozilla.org/mozilla-central/rev/12a6564962d7 https://hg.mozilla.org/mozilla-central/rev/ca1313dd7411