Operations

The person on call needs three things from the install: a metric, a log line, and a trace for the one stream that failed.

​

What the process exposes

• Prometheus metrics for the host and the sessions
• Structured logs
• A trace you can follow per stream

Watch the protocols you actually serve. A WebRTC session and an HLS crowd stress the machine in different ways. Size the instance for the concurrency you sold, then read the metrics before you add a second machine.

​

Where it sits

The same install runs in a datacenter, on a cloud VM, or on a network that never touches the public internet. See Install for Docker, the binary, and the package.

​

Restart

Drain viewers, then restart. Shutdown hooks exist so you leave a live room on purpose. A warm standby is a second machine you keep ready. You put your own load balancer in front of the pair.