deployment-pipeline-design

Define sequential stages for build, test, staging deploy, and production, then gate production behind an environment protection rule. In GitHub Actions, set required reviewers under Settings, Environments, production; in GitLab use a manual job; in Azure Pipelines use ManualValidation.

Canary suits high-traffic services needing metric-driven gradual rollout with automated analysis, while blue-green fits high-risk releases or database migrations needing instant switchover. Blue-green temporarily doubles infrastructure cost; canary requires Argo Rollouts or a service mesh.

Run a post-deployment health check script against a deep readiness endpoint, then trigger kubectl rollout undo on failure. Argo Rollouts can abort and roll back automatically when an AnalysisTemplate metric like error rate breaches its failure condition.

The pipeline is likely hitting a shallow ping endpoint that returns 200 even when dependencies are down. Use a deep readiness endpoint that verifies database, cache, and queue connections and returns 503 when any check fails.

The AnalysisTemplate metric query may return no data, leaving the analysis inconclusive and stalling promotion. Set inconclusiveLimit so the rollout fails fast instead of hanging, and verify the Prometheus metric names still exist.

Keep migrations backward-compatible using the expand/contract pattern so old code works with the new schema for at least one release cycle. Version undo scripts alongside forward migrations and avoid destructive changes until the old code is fully retired.