Engineering

Cutting per-pod memory with gunicorn preload and copy-on-write

We run a real-time voice service whose containers were using too much memory and restarting during calls. Two findings: the restarts were liveness-probe timeout…