1397538 - Video render start slows down over time

Status: RESOLVED INVALID

(Core :: WebRTC, defect, P3)

Description

[Nils Ohlmeier [:drno]]

- Open http://mozilla.github.io/webrtc-landing/pc_test.html
- Click "Start" and give access to cam and mic
- Notice how quickly the video elements for the remote video start to render the video

Now repeat about 20-30 times:
- Click "Stop"
- Click "Start" again
- Allow access to cam and mic (I haven't tried to give permanent permissions)


After the 30 times compare how long it starts for the big video elements to start rendering (ignore the smaller preview windows).

Something appears to slow down the initial rendering significantly. The page create new PeerConnections every time "Start" get clicked. So it's probably not a PeerConnection related problem, but more something around how the video element interacts with the web renderer or so. Just filling for here now as it's a WebRTC call which triggers the problem.

Comment 1

[Nils Ohlmeier [:drno]]

I was able to repro this all the way back to ESR52.

Comment 2

[Nils Ohlmeier [:drno]]

Attachment: Screen Shot 2017-09-06 at 16.26.22.png

As I'm experiencing this on a Mac I tried to look at it with Instruments. This is a screen shot of the call stack which appears to take a long time roughly from the time after accepting the cam/mix access prompt to when video rendering actually starts.
Is it normal that so much rendering is actually happening inside the callback function from CoreFoundation?

Comment 3

[Randell Jesup [:jesup] (needinfo me)]

Milan, who would be the right person on gfx side to help investigating this?  See the attachment - looks like VSync/RefreshDriver...

Comment 4

[Nils Ohlmeier [:drno]]

I actually wrote a test case for this here https://nils-ohlmeier.github.io/webrtc-landing/pc_test_loop.html

But I'm no longer able to repro. Maybe some state accumulation yesterday slowed down things...

Comment 5

[Milan Sreckovic [:milan] (needinfo for best results)]

Mason, can you take a look?

Comment 6

[Mason Chang [Inactive] [:mchang]]

I'm unable to reproduce this. 

(In reply to Randell Jesup [:jesup] from comment #3)
> Milan, who would be the right person on gfx side to help investigating this?
> See the attachment - looks like VSync/RefreshDriver...

This is just at the top of every call stack because that's where we begin to render. We need a call stack at the bottom.

(In reply to Nils Ohlmeier [:drno] from comment #4)
> I actually wrote a test case for this here
> https://nils-ohlmeier.github.io/webrtc-landing/pc_test_loop.html
> 
> But I'm no longer able to repro. Maybe some state accumulation yesterday
> slowed down things...

I can't reproduce this at all. If you can again, can you get a call stack that's all the way at the bottom?

Comment 7

[Bulk Bug Changes for mreavy's org]

Mass change P2->P3 to align with new Mozilla triage process.