Jibri setup in existing helm

[bharath-naik]

hi @minhpq331, i would like to work on jibri setup and contribute on this repo. do you have any suggestions for me or have you tried anything before for setting up and scaling the jibri. not sure but i have found one possible documentation from this site, may be this could help us click on this link

[minhpq331]

hi @bharath-naik , jibri setup is the hardest part because you must scale jibri instances based on the number of rooms, not only load. I've managed to make it work but for some reasons I'm not working on this setup anymore. There are some approaches you can try:

• Scale jibri with cloud provider (eg: aws scaling group or gg instance group), 1 jibri / 1 instance - The easiest way, support auto scaling.
• Scale jibri with multiple jibri / 1 instance - Same as your tutorial but you have to scale it manually.
• Use custom jibri with pulse audio and docker - the most complex method but support auto scaling. 1 minus point is pulse audio is less stable then alsa loopback (arcording to this thread). You must accept this trade-off if you want to deploy scalable jibri on k8s. this docker image is an example of jibri with pulse audio.

I've tested all 3 approaches and the last one seems meet my requirements. You could try that.

[marcoadasilvaa]

Hi everyone!

I am using instance scaling on GCP with a predictable estimate of the amount of jibri require on the next hour, i tried migrate to k8s but a i know the follow problems:

1 - How cant you get the currently amount meet and controller what jibri is running and available? 2 - In case GKE , is posible active modules snd and shm? In my case i tried why a Kops Cluster on GCP to settings a custom base image with ubuntu and extra package. 3 - Distribution of the jibri pod's on nodes? Because snd-aloop (loopback cards) have a limit number. For this case i used a StatefulSet with 10 replicas and included custom script inside jibri image to calcule loopback cards can used. I have not tried with pulseaudio jibri base.

I would be delighted to support this initiative.

[minhpq331]

@marcoadasilvaa hello, there are some limitations when you run jibri on GKE. I'll tell you some progress I've done so far (unfortunately I've dropped jitsi and move to other solution for my app). To run jibri on K8S, you need to:

• Change alsa loopback to pulseaudio (which run on docker without dependencies on host). You can google that jibri docker image. There are several ways to scaling jibri. I'd tried some and was working on a solution in the past. You can create your own solution with my direction:
• Scaling jibri based on load, which isn't efficient but is easiest.
• Scaling jibri based on metrics. You can setup a metric collector with prometheus to control the number of busy and idle instances.
• Scaling jibri based on custom event-controller like keda.sh

Good luck.

[marcoadasilvaa]

Thank you for tour comments.

I was working with Jitsi about 3 years with good result, maybe some expensive for a hight simultaneous concurrence.

Cant you shared what you replaced it with?

[minhpq331]

@marcoadasilvaa I dropped Jitsi because my app need more UI customization on client and low-level API access on backend. Jitsi is a well-tested product and a ready-to-go solution for conferencing but we need a custom calling service to integrate with our app.

FWI, we're testing livekit.io, a new opensource webrtc server.

[bharath-naik]

I am using instance scaling on GCP with a predictable estimate of the amount

HI @marcoadasilvaa , how are you able to predict the amount on cloud ? like how many people does your instance handle and per person how much is spent in a month. please help me out here. i need it badly. Thanks

[marcoadasilvaa]

I am using instance scaling on GCP with a predictable estimate of the amount

HI @marcoadasilvaa , how are you able to predict the amount on cloud ? like how many people does your instance handle and per person how much is spent in a month. please help me out here. i need it badly. Thanks

Hi @bharath-naik in my case, in video conference has appointment. i have a crontab to evaluate amount of appointment and provision jibri instance via autoscaling group

[bharath-naik]

hi @marcoadasilvaa do you have time to network with me. I am here . iam interested in your jibri setup.