Infrastructure & Operations › CI/CD & Deployment
Deployment
Putting a new version into an environment.
Also known as: deploy, deploying, software deployment, release to production, shipping code
Deployment is putting a new version of your software into an environment, so that it’s actually running there: copying the build, starting the new processes, switching traffic to them.
A typical flow:
- A change merges, and CI builds and tests it, producing a build artifact (a container image, a bundle).
- The same artifact is deployed to staging, and checked.
- The artifact is deployed to production. Traffic moves to the new version.
- You watch error rates and health checks, and roll back if something’s wrong.
# Simplified: the artifact is built once, then the same one is promoted
docker push registry.example.com/shop:3f9c2a1
deploy --env production --image registry.example.com/shop:3f9c2a1Strategies
| Strategy | How | Notes |
|---|---|---|
| Recreate | Stop the old version, start the new one | Simple, with downtime |
| Rolling | Replace instances a few at a time | No downtime, but old and new versions run together |
| Blue-green | Run the new version beside the old, then switch traffic | Instant switch and rollback, needs double capacity |
| Canary | Send a small share of traffic to the new version first | Limits the damage of a bad release |
Rules of thumb
- Build once, deploy the same artifact everywhere, and change only configuration between environments. Don’t rebuild for production.
- Automate it. Manual deploys are slow and error-prone, and they happen less often, making each one riskier.
- Make rollback easy and practiced (rollback).
- Plan for mixed versions. During a rolling deploy, old and new code run side by side, so database changes must work with both (schema and code deploys).
- Deploy small and often. Small changes are easier to understand and undo.
- Check health after deploying: error rate, latency, key business metrics.
- Separating deployment from release (turning the feature on) with feature flags lowers the risk (deploy vs release).
Deploying on Friday evening is a classic meme for a reason: deploy when you can watch and fix what you ship.