Deploying code is the riskiest moment in the software lifecycle — and the teams that do it safely dozens of times per day are not braver than everyone else, they have better patterns. PhonePe and Swiggy ship to production without incidents at high frequency because they have mastered the deployment strategies that bound blast radius, automate rollback, and make every change fully auditable. This pillar covers GitOps with ArgoCD and the deployment patterns that make production releases safe at scale.
What This Pillar Covers
- Implementing GitOps with ArgoCD — declarative Kubernetes deployments where Git is the single source of truth
- Blue-green deployments — smoke-testing before traffic switch, instant rollback by reverting the Service selector
- Canary deployments with Argo Rollouts — progressive traffic shifting with automated Prometheus metric gates
- Environment promotion pipelines — dev auto-deploy, staging auto-deploy, production behind manual approval gate
- Immutable artifact promotion — the same image digest that passed staging goes to production, never rebuilt
- Rollback strategies — kubectl rollout undo, ArgoCD Git-revert, automated metric rollback, and database migration safety
Who This Is For
Platform engineers and senior DevOps engineers managing production Kubernetes deployments who need patterns that are safe, auditable, and fast to recover from when something goes wrong.
Why This Matters in Production
During a Swiggy flash sale, a bad deploy reaching 100% of production traffic immediately can mean thousands of failed orders per minute before anyone notices. A canary at 5% with an automated Prometheus gate catches that before it reaches the rest of users and rolls back without waking anyone up. GitOps ensures every change to the cluster is traceable to a Git commit, a pull request, and a named approver — which is the audit trail compliance teams require and incident reviewers need.
Prerequisites
- Kubernetes Deployments, Services, and basic workload management
- CI/CD pipeline fundamentals — stages, artifacts, and environment promotion concepts
- Familiarity with Helm or Kustomize for Kubernetes manifest management