Switch from `webrtc-rs` to `str0m`

[thomaseizinger]

Description

The webrtc-rs implementation has a few fundamental issues due to its design. async callbacks require locks and prevent the idiomatic use of &mut. This results in problems such as webrtc-rs/webrtc#413 which currently prevent us from upgrading to the latest version: #3552

The new kid on the block is str0m, a sans IO implementation of the webrtc stack: https://github.com/algesten/str0m

In the tests, it looks like they already support data channels, meaning we might be able to replace webrtc with it.

Motivation

• webrtc is an extremely heavy dependency and it bloats our dependency tree
• It doesn't follow idiomatic Rust and was hard to integrate into our codebase
• From the above follows: Easier maintenance
• A production-grade webrtc implementation

Related issues:

• webrtc: allow dialing without listening first #4071
• transport/webrtc: Short buffer to be filled #4049

Are you planning to do it yourself in a pull request?

No.

[thomaseizinger]

cc @melekes

Depending on how many problems you run into once you deploy libp2p-webrtc, we might want to consider devoting some time to try this out instead of patching webrtc-rs.

[thomaseizinger]

@mxinden Any chance we can get this on the roadmap?

[thomaseizinger]

@mxinden Any chance we can get this on the roadmap?

If we want to do libp2p/specs#544 (which I think is a really good idea), I believe this issue is a blocker.

[drHuangMHT]

Related: #3138 with upstream issue on webrtc #104.

[mxinden]

@mxinden Any chance we can get this on the roadmap?

Can you create a pull request @thomaseizinger? Allows us to discuss priorities with the other items.

[thomaseizinger]

I've opened an issue for the failed SDP parsing which is where we hit a roadblock when @mxinden and I last explored what it would take to switch: algesten/str0m#164

[thomaseizinger]

The playground @mxinden and I have been using is here: https://github.com/thomaseizinger/str0m/tree/libp2p-webrtc-playground

Will keep exploring on the side.

[thomaseizinger]

The playground @mxinden and I have been using is here: thomaseizinger/str0m@libp2p-webrtc-playground

Will keep exploring on the side.

I've tried #164 out and with an additional small change to the SDP, this now parses correctly. I've pushed it to the branch above. The new output is now:

2023-05-12T07:47:36.455616Z  INFO str0m::ice::agent: Set remote credentials: IceCreds { ufrag: "libp2p+webrtc+v1/b01d95e45dee6a3f4e77bb87081eb97b:libp2p+webrtc+v1/b01d95e45dee6a3f4e77bb87081eb97b", pass: "libp2p+webrtc+v1/b01d95e45dee6a3f4e77bb87081eb97b:libp2p+webrtc+v1/b01d95e45dee6a3f4e77bb87081eb97b" }
2023-05-12T07:47:36.455660Z  INFO str0m: DTLS setup is: false
2023-05-12T07:47:36.455720Z  INFO str0m::sctp: Uninited => AwaitRemoteAssociation
2023-05-12T07:47:36.455760Z DEBUG str0m::channel: Allocate channel id: ChannelId(0)
2023-05-12T07:47:36.455801Z TRACE str0m::ice::agent: Enqueueing event: IceRestart(IceCreds { ufrag: "77hg", pass: "ADebVBR6nBhkiq5R8a1iSa" })
2023-05-12T07:47:36.455829Z  INFO str0m::ice::agent: State change (new connection): New -> Checking
2023-05-12T07:47:36.455845Z TRACE str0m::ice::agent: Enqueueing event: IceConnectionStateChange(Checking)
2023-05-12T07:47:36.455864Z TRACE str0m::ice::agent: Stop timeout since ice-lite do no checks
2023-05-12T07:47:36.455871Z TRACE str0m::sctp: Handle timeout: Instant { tv_sec: 4949, tv_nsec: 489601509 }
2023-05-12T07:47:36.455882Z DEBUG str0m::channel: Associate stream id ChannelId(0) => 1
2023-05-12T07:47:36.455902Z TRACE str0m::ice::agent: Poll event: Some(IceRestart(IceCreds { ufrag: "77hg", pass: "ADebVBR6nBhkiq5R8a1iSa" }))
2023-05-12T07:47:36.455911Z TRACE str0m::ice::agent: Poll event: Some(IceConnectionStateChange(Checking))
2023-05-12T07:47:36.455918Z DEBUG str0m: IceConnectionStateChange(Checking)

This hang there which might be due to what is mentioned in algesten/str0m#164 (comment).

[thomaseizinger]

New tracking issue on the str0m side: algesten/str0m#166