FEAT: How to do load balancing and service discovery for WEBRTC deployment in K8S

[jinleileiking]

When deploying WebRTC streaming using k8s, you may encounter an issue:

graph TD;
    pod1-->lb-->player;
    pod2-->lb;
    pod3-->lb;
    pod4-->lb;

Loading

Assuming the streaming is happening in pod1, due to load balancing, UDP packets may be routed to pod3. In this case, there are two possible solutions:

1. Introduce a proxy to handle WebRTC signaling messages, which is a complex solution. Refer to the solution provided in Cluster: K8s: Proxy to extend origin servers. #3138.

graph TD;
    pod1-->proxy-->lb-->player;
    pod2-->proxy;
    pod3-->proxy;
    pod4-->proxy;

Loading

2. pod3, find pod1 and start pulling the stream.

graph TD;
    pod1-->lb-->player;
    pod2-->lb;
    pod3-->lb;
    pod4-->lb;
    pod3-->pod1;

Loading

TRANS_BY_GPT3

[winlinvip]

WebRTC Cascading: #2091

WebRTC Proxy: #3138

For tens of millions or hundreds of millions of concurrent connections, cascading is a necessary solution, which is quite complex but also highly scalable.

For general situations, Proxy is simple and reliable, with decent scalability support. We will prioritize supporting the Proxy solution.

TRANS_BY_GPT3