CI/CD best practices: ship daily without breaking production

Short answer: keep pipelines fast, make every deploy reversible, and test in an environment that looks exactly like production. The goal is deploys so boring nobody notices them.

1. Keep the pipeline under ten minutes

Slow pipelines get skipped. Run tests in parallel, cache dependencies and build Docker images with layer caching.

2. Build once, deploy many times

Build a single immutable artefact per commit and promote it from staging to production. Never rebuild for production.

3. Use preview environments

Spin up a temporary environment for every pull request so reviewers and clients can click through changes before they merge.

4. Make deploys reversible

Use blue/green or canary deployments. Rolling back should be one click — or automatic when error rates spike.

- name: Deploy canary
  run: kubectl argo rollouts set image api api=$IMAGE
- name: Promote if healthy
  run: kubectl argo rollouts promote api

5. Treat database migrations carefully

Make migrations backwards-compatible: add columns before using them, and remove old columns only after the code that used them is gone.

6. Shift security left

Run dependency scans, secret detection and container image scans in the pipeline — not after an incident.

7. Measure it

Track the four DORA metrics: deployment frequency, lead time, change failure rate and time to restore. What gets measured improves.

Want a pipeline like this? See our CI/CD service.