deployment-pipeline-design

Configure the production job with an environment block naming the production environment, then require reviewers in the repository's environment protection rules. GitHub pauses the workflow until an authorized reviewer approves the deployment.

Blue-green runs two full environments and switches traffic instantly, doubling infrastructure cost temporarily. Canary shifts traffic gradually (for example 10%, 25%, 50%, 100%) using tools like Argo Rollouts, reducing blast radius but requiring progressive traffic control.

Run a health check step after deployment and, on failure, execute kubectl rollout undo for the deployment. You can also roll back to a specific revision using kubectl rollout undo with the --to-revision flag after listing history.

Use feature flags when you want to deploy code without releasing it, run A/B tests, or toggle functionality per user without infrastructure changes. Canary deployments suit validating whole releases under real traffic with gradual exposure.

Track deployment frequency, lead time from commit to production, change failure rate, mean time to recovery, pipeline success rate, and average pipeline duration. These DORA-style metrics reveal both delivery speed and release stability.