Custom Metrics for Autoscaling

Hey everyone, I am working with a group of other summer software interns for a small startup. We have been looking into how to scale in Kubernetes. Currently, we have our jicofo, prosody, and web containers all in one jitsi deployment, and we have our jvb in a stateful set with a HPA that scales when CPU utilization reaches 50%.

In our cluster, we have Prometheus running, and were wondering whether anyone used custom metrics when autoscaling their videobridges? Did anyone come across a combination of metrics that worked out better than just CPU utilization and/or memory usage?

Hi Marco,

I have just finished today extensively customized deployment of jitsi stack on GKE. And introduced I think the most interesting metric for scaling jvbs … the stress level :slight_smile:

Hope to report more this week how does HPA works with this custom metric