LiveKit Community

GH integration enabled in <#C01LPRVTQMU|>

hard to tell but most likely yeah. we're going to keep just filing issues + working through them

:+1::skin-tone-4: cheers, we’ll chip away at them on our end

Was just thinking about this, this morning! Integrating Krisp or how hard it would be to roll our own.

krisp is pretty good on mac / windows. On linux I use noisetorch. With a GPU its super easy / lightweight. Haven’t messed with a CPU based model to see how much resource it would take

I wonder what they charge B2B but might be interesting to make it “1 click active krisp”

The good thing is the time to build would be worth it: you can charge for it

Noticed yesterday that Agora also has a noise cancelling option via DNN

just released 0.3.3, fixed a bunch of connectivity stability issues.

<https://twitter.com/_pion/status/1359263749853437952?s=21|https://twitter.com/_pion/status/1359263749853437952?s=21>

I’ll grab lots shortly I think it might be a react hooks type issue but not sure 

So what is the difference with two vs one?

it uses a dedicated peerconnection for receiving client tracks, and another for pushing down tracks

for some reasons with a single PC it tends to get confused sometimes.. especially with simulcast turned on

ah thats really interesting. So when a client receives a track, they just render it on screen and the peer connection will facilitate streaming in the packets to the browser?

So the uploading of the packets is handled by one pc, and the downloading of other participants' packets is handled by another pc?

yeah, at a high level :slightly_smiling_face:

y'all! just released 0.4.2, with active speaker updates. LiveKit now detects speakers that are active and sends out `RoomEvent.ActiveSpeakersChanged` with current speakers.

can you share more about "websocket disconnections"? the logs themselves aren't terribly helpful.. they show that client WS has terminated. what would be interesting to know:

1. what was going on before that happens? 
2. what actions have been triggered on the client? 
3. what do the client logs say when this happens?

we'll try to round up more logs. for me it's the refresh issue that comes up. i'll try to add client logs onto that issue.

Abe saw more disconnects pretty regularly now with his test bots

k, sounds good.. when you refresh.. the logs would be showing that the previous websocket connection has disconnected.

does it not reconnect after a refresh? (i.e. when it's not a bot doing it)

on refresh here's what happens:

User connects -&gt; works good
User refreshes &amp; connects again -&gt; doesn't successfully connect. The client returns so we move forward but nothing works.
User refreshes &amp; connects again -&gt; works

<https://github.com/livekit/livekit-server/issues/4> is the issue that shows the server logs

Abe has other issues that are popping up so he'll report those when he has time but this is the one I know about

i'll get you some client logs on that issue tonight

I see. the server logs in the ticket is very limited, and doesn't really tell us anything other than the client refreshed... would need ~10 lines before and after the error to get a better idea

Hmm I thought it included the previous disconnect I’ll get you a new full set client + server 

The next major issue is the websocket disconnects, we'd like LK to auto retry and/or give us an event onDisconnect or something (where we could handle it)

<@U01KZ61GV5Z> any ideas on how I can access the peer connection on the client side? I'm interested in calling this to validate how many tracks actually have data flowing vs what we expect <https://developer.mozilla.org/en-US/docs/Web/API/RTCPeerConnection/getStats>

you can do `room.engine.publisher` and `room.engine.subscriber`

the `.pc` member gives you access to it.. tho.. an easier way is just to see if you've gotten a subscribed event for all of the tracks that are published (since auto subscribe it the current behavior)

well in our case a `OnTrackSubscribed` event sometimes results in video not streaming, but that might be a rendering issue on our end

so by checking the stats directly to see if the bytes are flowing through, we can determine its a rendering problem on our end

I have a general question. What was the thinking around having a single node and multi node option when a multi node can be of size 1 anyways? Was there a use case (or use cases) you guys thought of when adding the single node option?

Maybe dedicated nodes for large dedicated calls that can be spun up and down independently of a dependency like redis?

I could see having a system where each call, as it starts for the first time, spins up a dedicated single node instance that can spin down when the call is over. Then again, that might be a poor experience if the call is global (poor latency for users further away from the single box)

Perhaps the same could be said for multi node if you spun up a set of nodes + redis for each call

I haven’t thought about it (single node benefits over multi node) enough so I’m interested to learn the motivation

I think multi-node means there is a redis (that’s how i really think about it). So if no redis then it operates in “single”. Can just stub the redis calls and abstract a “bring your own DB” (in memory, redis, aerospike, etc)

Ya I understand the differences clearly. I’m interested in why it was decided to be an offering 

Screen Shot 2021-02-16 at 2.19.12 PM.png

hah, you've been collecting these.. nice!

webrtcsamples/tutorials sounds like good sites for content marketing :slightly_smiling_face:

I got a bunch more but saw this list thought it’d be fun to share 

the idea with single node was really to have a version of LiveKit that you could run with zero dependencies. It's designed to be simple to start, so people can try it out without the friction of Redis/configuration/etc

it basically stubs out redis methods with golang channels

right now we use --dev in production for the log_level, probably break that out into it's own flag?

I haven’t had much a chance to test today but will later tonight