deployment-procedures

Follow a five-phase process: prepare by verifying tests and builds, back up the current state, deploy while monitoring, verify health endpoints and key user flows, then confirm or rollback. Never deploy untested code or skip the backup step.

Match the platform to your app type: static sites fit Vercel, Netlify, or Cloudflare Pages; simple web apps fit Railway, Render, or a VPS with PM2; microservices need container orchestration; serverless workloads fit edge functions or Lambda.

Rollback immediately when the service is down or critical errors appear, and consider it when performance degrades more than 50 percent. Only fix forward for minor issues that can be resolved quickly without compounding errors.

Rolling replaces instances one by one and suits standard releases. Blue-green switches traffic between two environments and suits high-risk changes needing easy rollback. Canary shifts traffic gradually and suits changes needing validation with real users.

On Kubernetes, run kubectl rollout undo to return to the previous revision. With Docker, redeploy the previous image tag. On Vercel or Netlify, redeploy the previous commit, and on Railway or Render use the dashboard rollback option.

Verify the health endpoint responds, scan error logs for new exceptions, test key user flows, and confirm response times are acceptable. Monitor actively for the first 5 minutes, confirm stability at 15 minutes, and review metrics the next day.