Fix latency stream

[lukh]

providing a fix for big latency when reading streams from web.
the fix comes from here:
https://discourse.mopidy.com/t/streaming-radio-station-pausing-buffering/1251/19

[codecov]

Codecov Report

Merging #1845 into develop will increase coverage by 0.01%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           develop    #1845      +/-   ##
===========================================
+ Coverage     82.2%   82.22%   +0.01%     
===========================================
  Files           77       77              
  Lines         6334     6339       +5     
===========================================
+ Hits          5207     5212       +5     
  Misses        1127     1127

Impacted Files	Coverage Δ
mopidy/audio/actor.py	59.72% <100%> (+0.46%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 400e0ea...884b489. Read the comment docs.

[kingosticks]

This change affects all audio we get over http, including stuff we want buffered properly.

[lukh]

hello,
Can you please help me understand the situation here ?
Is there any better solution ?
For me, having a 5sec latency when streaming is a problem,
and I don't see the side effects of this PR ?
Is there any solution ? as a fix/add to mopidy or as a local fix for specific projects ?
Thank for your help,
Bests,

[adamcik]

This fix would work great for things that are actually streaming, e.g. things you can't seek in. But say you have backend that is http based but lets you stream with HTTP ranges than you're breaking them with this.

As such we need a way for backends to give a hint to audio system about if this is live or not. Currently we have two main ideas for this:

1. Creating a new Ref type stream - which backends would use for live content like radio streams
2. Changing

   mopidy/mopidy/backend.py

   Line 216 in 5720197

   def change_track(self, track):

   (or something thereabouts) and the audio API to let Backend Playback providers indicate that live should be used, or to trigger the download flag in GStreamer.

Basically this needs to be something the backends opt in to, not something on by default. Does this make more sense?

[adamcik]

#1608 is essentially the "same" problem, just a different GStreamer flag.

[lukh]

Ok thank you for your answer ! I get it now (at least, I think so...)
I added a new flag in Track, live_stream, that is set or not by the backend.

So Lib providers are free to set "live_stream", and the PlaybackProvider.change_track has the info (via track object) and passes it to the audio actor.

The audio actor has a flag as well, and I kept the implementation but with the flag check. (maybe I should use gstreamer flags but I don"t see how download flag work with streams ??) (from #1608)

I didn't get how to use the Ref objects, because there is no way to recover the Ref object from change track ?

[lukh]

Hello, @adamcik, What do you think of this implementation ? Is that what you meant ?

Would you consider merging ? And would you consider merging for the 3.0.0 release if we reach the quality required ?

Thank you !

[adamcik]

Indeed this is along the lines of what I was thinking, main thing perhaps @jodal or @kingosticks could also chime in on is naming and having the state in the model vs asking the playback provider? Even though these are probably bikesheds.

I can't think of any strong reasons for one over the other, so if nobody else has objections I would have no problem with going with this as is.

[kingosticks]

We don't want to test with real streams. The tests should run quickly. Is there not a way to mock what you want?

[lukh]

Yeah it looks a bad idea. I wanted to test the new section in _on_source_setup, inside
if self._live_stream and hasattr(source.props, "is_live"):
It is why the codecov decreased I think: if i use the uris given in base test, the is_live property is false, therefor, I can't test this part.
I am afraid I don't know how to mock up this part, because the _on_source_setup is called by Gst 's playbin, on a signal after a set_uri.
I will try to make a test for _on_source_setup only, but i think it should start with set_uri.

[lukh]

This is all I can think of for now, what do you think ?

[jodal]

Looks good to me. Feel free to add a changelog entry in the audio section of v3.0.0 in docs/changelog.rst.

[jodal]

Thanks! I've squashed and merged this in commit d3c1824.