The primary Prometheus metric that livekit-server exposes to help trigger scaling or launching of an egress pod and instance is livekit_egress_available. This metric indicates whether an egress instance is available to accept new requests based on its CPU utilization. It provides a more accurate assessment of instance availability compared to average CPU or memory usage. This metric is commonly used in Kubernetes Horizontal Pod Autoscaling (HPA) configurations to maintain a minimum number or percentage of available instances, ensuring sufficient capacity for egress workloads. For example, you can configure autoscaling to maintain at least 3 available egress instances or 30% availability of the total instances based on this metric. This metric is exposed when the

prometheus_port

is configured on livekit-server, allowing Prometheus to scrape these metrics for autoscaling purposes. Sources: Self-hosting the Egress Service | LiveKit Docs | config-sample.yaml