webrtcsink
All-batteries included GStreamer WebRTC producer, that tries its best to do The Right Thing™.
Use case
The webrtcbin element in GStreamer is extremely flexible and powerful, but using it can be a difficult exercise. When all you want to do is serve a fixed set of streams to any number of consumers, webrtcsink (which wraps webrtcbin internally) can be a useful alternative.
Features
webrtcsink implements the following features:
-
Built-in signaller: when using the default signalling server (provided as a python script here), this element will perform signalling without requiring application interaction. This makes it usable directly from
gst-launch. -
Application-provided signalling:
webrtcsinkcan be instantiated by an application with a custom signaller. That signaller must be a GObject, and must implement theSignallableinterface as defined here. The default signaller can be used as an example. -
Sandboxed consumers: when a consumer is added, its encoder / payloader / webrtcbin elements run in a separately managed pipeline. This provides a certain level of sandboxing, as opposed to having those elements running inside the element itself.
It is important to note that at this moment, encoding is not shared between consumers. While this is not on the roadmap at the moment, nothing in the design prevents implementing this optimization.
-
Congestion control: the element levarages transport-wide congestion control feedback messages in order to adapt the bitrate of individual consumers' video encoders to the available bandwidth.
-
Configuration: the level of user control over the element is at the moment quite narrow, as the only interface exposed is control over proposed codecs, as well as their order of priority, and disabling congestion control. Consult
gst-inspect=1.0for more information.
More features are on the roadmap, focusing on mechanisms for mitigating packet loss.
It is important to note that full control over the individual elements used by webrtcsink is not on the roadmap, as it will act as a black box in that respect, for example webrtcsink wants to reserve control over the bitrate for congestion control.
If more granular control is required, applications should use webrtcbin directly, webrtcsink will focus on trying to just do the right thing, although it might expose interfaces to guide and tune the heuristics it employs.
Building
Prerequisites
The element has only been tested for now against GStreamer master.
For testing, it is recommended to simply build GStreamer locally and run in the uninstalled devenv.
Make sure to install the development packages for some codec libraries beforehand, such as libx264, libvpx and libopusenc, exact names depend on your distribution.
git clone https://gitlab.freedesktop.org/meh/gstreamer/-/tree/webrtcsink
meson build
ninja -C build
ninja -C build devenv
Compiling
cargo build
Usage
Open three terminals. In the first, run:
cd signalling
python3 simple-server.py --addr=127.0.0.1 --disable-ssl
In the second, run:
cd www
python3 -m http.server
In the third, run:
export GST_PLUGIN_PATH=$PWD/target/debug:$GST_PLUGIN_PATH
gst-launch-1.0 webrtcsink name=ws videotestsrc ! ws. audiotestsrc ! ws.
When the pipeline above is running succesfully, open a browser and point it to the python server:
xdg-open http://127.0.0.1:8000
You should see an identifier listed in the left-hand panel, click on it. You should see a test video stream, and hear a test tone.
Configuration
The element itself can be configured through its properties, see gst-inspect-1.0 webrtcsink for more information about that, in addition the default signaller also exposes properties for configuring it, in particular setting the signalling server address, those properties can be accessed through the gst::ChildProxy interface, for example with gst-launch:
gst-launch-1.0 webrtcsink signaller::address="ws://127.0.0.1:8443" ..
The signaller object can not be inspected, refer to the source code for the list of properties.
Testing congestion control
For the purpose of testing congestion in a reproducible manner, a simple tool has been used, I only used it on Linux but it is documented as usable on MacOS too. I had to run the client browser on a separate machine on my local network for congestion to actually be applied, I didn't look into why that was necessary.
My testing procedure was:
-
identify the server machine network interface (eg with
ifconfigon Linux) -
identify the client machine IP address (eg with
ifconfigon Linux) -
start the various services as explained in the Usage section (use
GST_DEBUG=webrtcsink:7to get detailed logs about congestion control) -
start playback in the client browser
-
Run a
comcastcommand on the server machine, for instance:/home/meh/go/bin/comcast --device=$SERVER_INTERFACE --target-bw 3000 --target-addr=$CLIENT_IP --target-port=1:65535 --target-proto=udp
-
Observe the bitrate sharply decreasing, playback should slow down briefly then catch back up
-
Remove the bandwidth limitation, and observe the bitrate eventually increasing back to a maximum:
/home/meh/go/bin/comcast --device=$SERVER_INTERFACE --stop
For comparison, the congestion control property can be set to disabled on webrtcsink, then the above procedure applied again, the expected result is for playback to simply crawl down to a halt until the bandwidth limitation is lifted:
gst-launch-1.0 webrtcsink congestion-control=disabled
License
All code in this repository is licensed under the MIT license.
363 Jan 5, 2023
34 Nov 10, 2022
58 Dec 11, 2022
3 Dec 4, 2022
15 Jan 4, 2023
13 Oct 23, 2023
5 Jul 8, 2023
13 Oct 25, 2021
17 Jun 7, 2022
41 Nov 10, 2022
1.9k Jan 8, 2023
12 May 11, 2022
326 Dec 21, 2022
3 Sep 25, 2022
2 Sep 11, 2022
1 Apr 20, 2022
2 Dec 21, 2021
7 Nov 25, 2022