This message was deleted.

you can do this using the prometheus metric

livekit_egress_available

or by CPU

Ooh, that's a great idea! I take it there are no ready-made solutions at the moment? Either Kubernetes or Prometheus to find out if additional resources are needed for recording and raise them up. Right?

yeah, nothing ready-made but we just use kube autoscaling with that prometheus metric to make sure there are always some instances available

All right, I'll probably raise the kube after all. Thanks. so far it is not clear what resourses to give to node, because the number of video streams increases the resources that are used. 4 cores. 1 users +- 30% (here it is clear that the chrome shell is running, etc.) 25 users (bots) load in egress +- 80%. 50 everything hangs and writing is impossible. Although there are not so many users on the screen, and I thought that between 10 users and 1000 should not be a big difference at all.

if you know the workload, 4 cores might be ok, but it would be better to use 6 or 8 if you don’t know the workload