Zero-Downtime Deployments: How We Keep CoDuck Always Online

The Requirement

CoDuck users are building applications in real-time through conversation. An AI generation can take 30-60 seconds. If we restart the server during a generation, the user loses their work.

The Solution: PM2 Cluster Mode

We run our API server in PM2 cluster mode with 2 instances. When we deploy a new version, PM2 performs a rolling restart.

The Result

We deploy multiple times per day with zero user-facing downtime.