JVB/Web OCTO region sharding support

[anirbandas18]

Hi,

I would like to know if this helm-chart supports a Jitsi shard and videobridge pools naturally out of the box?

jitsi shard = web + jicofo + prosody videobridge pools = jvb replicas (already supported I guess)

My use case is to have multiple shards accessing a common set of videobridges amongst them.

Will this helm-chart support my use case naturally or do i have to modify it or roll out my own helm-charts?

[spijet]

Hello @anirbandas18!

As for now, JVB multi-replica setups are supported via OCTO, and same should hold true for the Web component. Not sure about Jicofo, and Prosody is a stateful service holding all internal state of Jitsi Meet components, so sharding it should roughly be equal to deploying multiple (separate) Jitsi Meet instances, if I understand it right. See also #89 for a similar question.

There's an ongoing discussion about splitting JVB replicas to OCTO regions in #107, you can join.

[anirbandas18]

So, my idea is to apply horizontal pod auto scaling to a jitsi shard (as defined above) which should handle the stateful nature of prosody as well and use octo in conjunction between jicofo of each shard vs the common jvb in the videobridge pool by ensuring each shard sets the correct userRegion value for jicofo to select a jvb from the pool properly.

Does this sound achievable ?

[spijet]

AFAICT, HPA operates on groups of identical pods, so by definition it would be possible to use it to scale individual groups of Jitsi Web or JVB pods (as in separate auto-scaling groups for different regions or one "global" auto-scaling group), not to create new regions/shards.

[anirbandas18]

does this helm-chart support HPA of such group of pods?

what do you mean by creating new region/shards in this context?

[spijet]

Unfortunately not, this is not supported at the moment.

As for the regions/shards, here's an example from current main branch:

# templates/jvb/configmap.yaml
  {{- if .Values.octo.enabled }}
  ENABLE_OCTO: "1"
  JVB_OCTO_BIND_PORT: "4096"
  JVB_OCTO_REGION: "all"
  {{- end }}

The last variable (JVB_OCTO_REGION) tells us that all JVB pods (no matter the replica count) are assigned to the same OCTO region, called all, and all the pods will be equally eligible for use by all clients, no matter their IP/location/etc.

In order to introduce different shards/regions we have to rework this part of the chart, so that we can create separate JVB Deployments with separate ConfigMaps that specify different values for JVB_OCTO_REGION, and then do the same for the Web component as well.

[anirbandas18]

Will this be reworked as an enhancement?

[spijet]

Maybe/probably, if I come up with a good enough way to implement it. :)

[anirbandas18]

Anyway I can help?

[spijet]

PRs are always welcome, so you can try implementing it if you have an idea how to do it. :)

I'll try to do so too, but unfortunately at the moment I'm a bit caught up at my full-time job.