1171778 - High CPU usage when hitting end of MSE content where endOfStream wasn't called

Status: RESOLVED WORKSFORME

(Core :: Audio/Video, defect)

Description

[Jean-Yves Avenard [:jya]]

This can be seen in very simplistic MSE page like:
http://people.mozilla.org/~cpearce/simple-dash/

or:
http://people.mozilla.org/~jyavenard/tests/mse_mp4/paper.html

These pages load data, but don't call mediasource.endOfStream().

As such, we don't hit EOS/ended once reaching the end of the data, but "stalled" instead.

The MediaDecoderStateMachine will continuously attempt to decode the next audio/video frame (by calling Advance Frame) even though we have nothing to decode.

The MDSM has a mechanism to know when data has been added, it shouldn't unecessarily retries forever to see if new data is available.

On my machine it causes nightly to reach 115% CPU usage.

Comment 1

[Jean-Yves Avenard [:jya]]

In the log you can see hundreds of mediapromise being created / resolved and the MDSM scheduling a new run several times per second.

Comment 2

[Jean-Yves Avenard [:jya]]

On Aurora, the effect is most surprising.

the video never finish playing. 
http://people.mozilla.org/~jyavenard/tests/mse_mp4/bipbop.html

Comment 3

[Chris Pearce [:cpearce (Not reading bugmail)]]

I would have thought that the Request{Audio,Video}Data() promise would not be resolved until we have a sample? Or is it resolved with "waiting for data", and we try again?