Deployment CI CD

Iron Law: Never design a deployment pipeline without knowing the rollback strategy — every deploy step must be reversible or have an explicit break-glass procedure.

Deployment CI/CD

CI/CD pipeline design and deployment automation specialist covering GitHub Actions, Docker, Cloud Run, Flutter mobile distribution, and Firebase deployments.

When to Use

• Setting up a new GitHub Actions CI/CD pipeline
• Reviewing an existing deployment workflow
• Designing Docker multi-stage build optimization
• Setting up Cloud Run progressive delivery (traffic splitting)
• Automating Flutter builds and distribution to Firebase App Distribution
• Configuring Firebase Hosting, Firestore rules, or Functions deployment

Pipeline Patterns by Stack

Key Principles

• Multi-stage Docker: separate build and runtime stages; never ship dev dependencies
• Health check gate: Cloud Run deploy must pass health check before traffic shift
• Secret management: use Workload Identity Federation (not service account keys)
• Progressive delivery: canary → 10% → 50% → 100% with automated rollback on error rate spike
• Container scanning: Trivy scan before push to Artifact Registry

Process

1. Identify target stack and deployment environment
2. Load docker skill for Dockerfile patterns (if containerizing)
3. Load gcp-cloud-run skill for Cloud Run-specific patterns (if deploying to GCP)
4. Design pipeline stages: build → test → security scan → push → deploy → verify
5. Define rollback strategy for each stage
6. Output GitHub Actions workflow YAML + deployment configuration

Production Readiness Checklist

Before any production deploy, verify ALL of the following. If any item is NO → do not deploy.

Application

• /health endpoint responds 200 with DB connectivity check (see reference/health-endpoints.md)
• All required env vars validated at startup — app fails fast if missing (see reference/env-config-validation.md)
• No secrets hardcoded in source code or committed config files
• .env.example committed with all keys, placeholder values only
• Error handling: all catch blocks log + rethrow or return error state (no silent failures)

Container / Docker

• Multi-stage Dockerfile — dev deps not in production image
• Non-root user in Dockerfile (USER appuser)
• Specific image tags pinned — no :latest
• HEALTHCHECK instruction in Dockerfile
• .dockerignore excludes: node_modules, .git, .env, dist, coverage
• docker scout or trivy image scan — zero CRITICAL CVEs

CI/CD Pipeline

• Tests pass (all: unit + integration)
• Linting + type-check pass (zero errors)
• Container image pushed to Artifact Registry with commit SHA tag
• Health check gate passes before any traffic shift
• Rollback strategy defined and tested for this deploy

Cloud Run (GCP Stack)

• Workload Identity Federation configured — no service account key files
• Deploy uses --no-traffic first → canary 10% → verify → 100%
• Memory limits set (--memory flag matches actual service requirements)
• Concurrency configured (--concurrency — default 80, tune per service)
• Min instances = 0 for cost (or 1+ for latency-sensitive services)

Security

• security-reviewer agent verdict: no CRITICAL or HIGH findings
• All API endpoints require authentication (no accidental public exposure)
• CORS policy explicitly configured (not wildcard * in production)
• Secrets stored in Secret Manager, not Cloud Run env vars directly

Post-Deploy Verification

• Health endpoint returns 200 on new revision URL
• Smoke test: critical user flow works end-to-end
• Error rate < 0.1% for 5 minutes post-deploy
• Rollback procedure verified working (test once per quarter)