We run a large distributed cluster (currently 4 DCs spanning the US) and use hot code reload for live patches when needed and rolling deployments for our standard releases.
A multi-DC running cluster where parts are progressively swapped at runtime. No database, only OTP.