Analysis of why this happens:

1.) PepperToVideoTrackAdapterTest creates an instance of ChildProcess, which, in turn, creates and sets the global |g_task_scheduler|.
2.) When ChildProcess is destroyed, its destructor does not clean up the global |g_task_scheduler|.
3.) WebContentsVideoCaptureDeviceTest.SafelyStartsUpAfterWebContentsHasGone creates a ScopedTaskScheduler. The constructor hits a DCHECK, because it expects the global |g_task_scheduler| to be nullptr.

Possible solutions:
1.) Do not use globals. Probably would require too much redesign/refactoring work to be practical.

2.) Have ChildProcess destructor clean up the |g_task_scheduler| if it set one. This may be dangerous, because (based on what the destructor currently does) it appears like certain things are allowed to keep running on |g_task_scheduler| even after it has shut down. Need to get input from someone intimately familiar with this to check whether or not this could be a solution.

3.) Have PepperToVideoTrackAdapterTest manually clean up the |g_task_scheduler| after its ChildProcess instance has been destroyed. Unfortunately, this is not a very scalable solution, since ChildProcess is created in many tests. The same issue may exist with other tests.

Another possible solution:
4.) Remove the DCHECK() that requires |g_task_scheduler| to be nullptr when creating a ScopedTaskScheduler. However, it would not be clear whether or not ScopedTaskScheduler should then clean up |g_task_scheduler| in its destructor.

The CL that lead to this issue is probably https://codereview.chromium.org/2627953002.

+ folks from that review

jam@, gab@: Could you take a look and let me know what would be a good way to handle this?

+ fdoray

"2.) When ChildProcess is destroyed, its destructor does not clean up the global |g_task_scheduler|."
This is intended. CONTINUE_ON_SHUTDOWN tasks can continue to run after ChildProcess has been destroyed.

This CL fixes this issue: https://codereview.chromium.org/2749543003/

#5: Thanks for preparing a CL so quickly. I believe the proposed CL will fix the issue for PepperToVideoTrackAdapterTest. However, this would only fix one symptom of the underlying problem. There may be other tests that create a ChildProcess instance and other tests that create ScopedTaskScheduler instances. With the current design, we may get the same kind of crash randomly depending on the order in which we execute the tests. Is there something we can do to fix the problem as a whole?

fdoray@: Any updates on this? I am mostly interested in learning what your plans are.

Bumping Priority up to 2, since I keep running into this and I think having unit tests fail depending on the order in which they are run is a serious issue.

Running out/Default/content_unittests --gtest_filter="*Video*" on the latest trunk currently produces 4 crashing tests:

MediaStreamVideoSourceTest.AddTrackAndStartSource
MediaStreamRemoteVideoSourceTest.StartTrack
HTMLVideoElementCapturerSourceTest.ConstructAndDestruct
PepperToVideoTrackAdapterTest.Open

[78102:78102:0613/094948.711195:2504928999760:FATAL:scoped_task_environment.cc(93)] Check failed: !TaskScheduler::GetInstance(). 

I'll create a ChildProcessForTesting class that does the required cleanup.

What's the status on the CL [1] ?
I ran into this issue again today. Got 3 random tests failing while trying to verify some unrelated change I was working on.

After [1] has landed, we still need to migrate all tests that use ChildProcess to ChildProcessForTesting.

[1] https://chromium-review.googlesource.com/c/541595

 Issue 751093  has been merged into this issue.

New solution under review: https://chromium-review.googlesource.com/c/chromium/src/+/716817

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/71d8c0a2541c85c818dc38faf7c13811267a284d

commit 71d8c0a2541c85c818dc38faf7c13811267a284d
Author: Francois Doray <fdoray@chromium.org>
Date: Wed Oct 18 13:30:37 2017

Add ScopedTaskEnvironment to media unit tests that use ChildProcess.

When ChildProcess is instantiated in a scope where there is no
existing TaskScheduler instance, it creates one that will be leaked.
Leaking allows CONTINUE_ON_SHUTDOWN tasks to continue running until
the very end of the process' life. Unfortunately, leaking a
TaskScheduler in a unit test because it prevents other unit tests
to instantiate their own TaskScheduler.

With this CL, a ScopedTaskEnvironment is instantiated before
ChildProcess in media unit tests. That prevents ChildProcess
from creating a TaskScheduler that will be leaked (instead,
ChildProcess uses the TaskScheduler whose lifetime is managed
by ScopedTaskEnvironment).

Bug:  682370 
Change-Id: I36d57a54e03c103871a596ca3f16eb4fc4307679
Reviewed-on: https://chromium-review.googlesource.com/716817
Commit-Queue: François Doray <fdoray@chromium.org>
Reviewed-by: Dale Curtis <dalecurtis@chromium.org>
Reviewed-by: Emircan Uysaler <emircan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#509758}
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/media_stream_video_capturer_source_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/media_stream_video_renderer_sink_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/media_stream_video_track_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/pepper_to_video_track_adapter_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/rtc_peer_connection_handler_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/user_media_client_impl_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/video_track_to_pepper_adapter_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/webrtc/media_stream_video_webrtc_sink_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/webrtc/webrtc_media_stream_adapter_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media/webrtc/webrtc_video_capturer_adapter_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media_capture_from_element/canvas_capture_handler_unittest.cc
[modify] https://crrev.com/71d8c0a2541c85c818dc38faf7c13811267a284d/content/renderer/media_recorder/video_track_recorder_unittest.cc