1660144 - Remote video freezes instead of going to black., causing test failures

Status: RESOLVED FIXED

(Core :: WebRTC: Audio/Video, defect, P2)

Description

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

Attachment: FakeDeviceNotAdvancing.png

With patch-set in bug 1652884, I'm observing that the "fake" MediaEngineDefault device we use in tests is intermittently not advancing frames. This is causing test failures that seem related to test infrastructure only and unnecessarily blocking landing of bug 1652884.

STRs:

1. Open https://jsfiddle.net/jib1/ux5cw31e/
2. Wait 5 seconds

Expected results: Both squares show the same color.

Actual results: Only the square on the left changes color, the one on the right stays green.

Workarounds:
1 . Hitting the jsfiddle "Run" button seems to make it snap out of it.
2. Removing fake: true also works: actual camera video does not exhibit this problem.

Comment 1

[Andreas Pehrson [:pehrsons]]

I see the same thing without bug 1652884 FWIW. It has some trouble turning black on mute at times too, when the video is flowing as it should. That I only saw with bug 1652884, so maybe that's the root cause of the intermittent here.

I'll try to fix both.

Comment 2

[Andreas Pehrson [:pehrsons]]

From the STR in the description:
https://pernos.co/debug/YY8TaitQFl9N0DVeGbk4wQ/index.html

Comment 3

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

Comment 0 turned out to be a red herring caused by autoplay, which I had on in Nightly and not in local build. Sorry 😕

Comment 1 seems like a real problem though, which I can repro. With https://jsfiddle.net/jib1/ux5cw31e/2 if I disable the track (check mute on the page) after ~3 seconds then the remote video freezes instead of going to black.

Comment 4

[Andreas Pehrson [:pehrsons]]

I seem to be able to repro the test failure in headless mode but not in non-headless mode. Here's a recording of a timed out headless-mode run: https://pernos.co/debug/4V0lv1t5l7CzUwhYsAparA/index.html

Comment 5

[Andreas Pehrson [:pehrsons]]

Attachment: Bug 1660144 - Test video frame blackness in VideoFrameConverter gtests. r?dminor

This would have caught bug 1660144 before we landed bug 1652884.

Comment 6

[Andreas Pehrson [:pehrsons]]

Attachment: Bug 1660144 - On disabling VideoFrameConverter, take a path that actually turns the frame black. r?jib

Depends on D87790

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/4c02a0f5d8a5
Test video frame blackness in VideoFrameConverter gtests. r=dminor
https://hg.mozilla.org/integration/autoland/rev/0cb072c2593a
On disabling VideoFrameConverter, take a path that actually turns the frame black. r=jib

Comment 8

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/4c02a0f5d8a5
https://hg.mozilla.org/mozilla-central/rev/0cb072c2593a

Comment 9

[Karl Tomlinson (:karlt)]

Would you expect the changes here to mean that https://hg.mozilla.org/integration/autoland/rev/17e11043a3c8 can now be reverted?

Comment 10

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

Yes looks good. I'll re-enable it.