This message was deleted.

cc @able-gigabyte-21598 (also Ben please keep messages in one thread, it’s easier for us to keep track of this way 🙏 )

apologies

no worries

Context: calling StartTrackCompositeEgress from the js sdk

It means gstreamer stopped responding. There is a bug where this is occasionally happening to h264 (video) tracks, which https://github.com/livekit/egress/pull/449 aims to fix

i hear gstreamer is a pain in the ass

We encounter same issue ,too(pipeline frozen) thank you for reporting.

@acoustic-engineer-41666 The egress service was working well all day and outputting to s3, but just started having pipeline frozen responses again after I tried to send a screenshare and a video track to it at the same time (I have since noticed only one egress can work at once on a free account). I was in the process of doing a demo of the service to my company. We are trying to determine whether we can use this for our video requirements. Is there any way I can get the egress working again for a demo by EOD?

@clean-belgium-47556 If you submit a quota increase request I’ll approve it right away, that should at least let you have >1 egress

Okay, so to summarize where we are at after testing today: 1. Recording two track composite egresses at once fails on both free and paid plans (which say they have 50 concurrency by default) 2. After a recording fails, it seems to take a random amount of time before the egress succeeds again There's also an additional thing I'd like to ask about, the first frame of the capture seems really blurry, and then the video goes to full quality. Is that possible to remove from the saved videos?

I can attest @acoustic-engineer-41666, we're super impressed with what you guys are building, these are the only key issues preventing us from moving forward with a production implementation.

You mentioned one screenshare and one video track - track composite egress is built to handle at most one video and one audio track. It sounds like you're sending two video tracks, which would require a room composite

I need fine grained control over which tracks get recorded when, and how many files get created from doing so. My assumption was that the room egress recorded the whole chatroom as a single video, with exactly one command to turn it on and off. I chose an implementation where i bundle camera and mic into one track, and grab a second screenshare track when needed, and sent them as separate startTrackCompositeEgress commands. Is there something in the architecture that forbids this?

I see, room composite is just a single video as you said. The gstreamer freezes are my top priority this week, there definitely shouldn’t be any issues running two track composite at once. As for the bit of blurred video at the start, there is a tradeoff between having a clean start and recording as quickly as possible to capture as much data as possible, since a clean start would mean not starting the recording until we receive a keyframe. There might be a way for me to show a black screen at the start instead, but our priority for now is making sure these never fail

Understood, thanks for the explanation. How far apart are the key frames?

I think it depends on the client, chrome by default sends one every 3000 frames with vp8. When egress joins the room, it sends a keyframe request through the SFU to the client, so the lag at the beginning is the time it takes for the client to receive that request and create one

we are working on something that will make recording each participant a lot easier. basically you don't have to worry about track lifecycles anymore, where the output is a single recording for each participant (and optionally a secondary recording for their screen share)

@dry-elephant-14928 that sounds awesome (I work with Ben), is their an eta as to when that will be shipped?