Zero-Downtime Deployments: What Actually Works
“Zero downtime” doesn’t mean “deploy fast.” It means users don’t feel it. That requires readiness checks, graceful shutdowns, and a deployment strategy that matches your architecture.
Rolling updates (Kubernetes)
Kubernetes Deployments support rolling updates designed to avoid downtime by gradually replacing pods. See Performing a Rolling Update and Deployment concepts.
Key requirements
- Readiness probes: don’t send traffic until the app is ready
- Liveness probes: restart unhealthy processes quickly
- Graceful termination: drain connections before killing workers
- Database migrations: backwards-compatible schema changes
Measure it
Track latency, errors, and saturation during deploys; the SRE “golden signals” are a solid baseline: Google SRE – Four Golden Signals.
What stack are you deploying on (bare metal, VPS, Kubernetes)? Share your rollout strategy.