Design CI/CD pipelines: branching strategy, build/test stages, artifacts, environment…
- User
Design CI/CD pipelines: branching strategy, build/test stages, artifacts, environment promotion, canary/blue-green, rollback, and approvals; include example configs.
- Assistant
1) Guiding principles (what “good” looks like)
- One source of truth: Git + a single immutable build artifact.
- Build once, promote the same artifact across environments (no rebuilds for staging/prod).
- Fast CI on every PR; gated CD to production.
- Progressive delivery (canary/blue-green) with automatic rollback on SLO/metric breach.
- Policy-as-code: approvals, checks, and security gates enforced by the platform.
2) Branching strategy
Recommended default: Trunk-based + short-lived branches
main(protected): always releasable.feature/*(short-lived): PRs intomain.release/*(optional): only if you need stabilization for a dated release train.hotfix/*: emergency fixes, merged back tomain.
Release identifiers
- Tag releases from
main:v1.8.0. - Every pipeline run also produces a unique build ID (e.g.,
sha-<shortsha>).
Rules
- PR required to merge to
main. - Required checks: lint, unit, integration (if feasible), SAST/dep scan, container scan.
- Code owners + minimum reviewers.
- No direct pushes to
main(except break-glass).
3) CI pipeline (build/test) stages
Typical CI stages (run on PRs and
main):- Validate
- formatting/lint
- typecheck (if applicable)
- secrets scan (e.g., gitleaks)
- Unit tests
- coverage threshold gate
- Build
- compile/package
- build container image
- Integration tests
- ephemeral dependencies (docker-compose/testcontainers)
- Security
- SAST (CodeQL/Semgrep)
- dependency scan (SCA)
- container image scan
- Publish artifacts
- push container image to registry
- publish SBOM (CycloneDX/SPDX)
- sign artifacts (cosign) + provenance (SLSA)
Optimization
- Cache dependencies.
- Split tests in parallel.
- Only run expensive suites on
main/nightly if needed, but keep at least a smoke suite for PRs.
4) Artifacts (what you publish)
Immutable artifacts to publish per build
- Container image:
registry/app:<gitsha>(+ optionally:v1.8.0on tag) - Deployment bundle: Helm chart / Kustomize / manifests versioned alongside app (or separate repo)
- SBOM:
sbom.json - Provenance:
provenance.intoto.jsonl - Test reports: JUnit XML, coverage
Key rule: Promotion means “deploy the same image digest” (not “rebuild from the same commit”).
5) Environment promotion model
Example environments and gates:
Environment Trigger Purpose Gates PR ephemeral PR open/update fast feedback auto Dev merge to mainintegration w/ shared services auto Staging successful dev + manual/auto rule production-like auto tests + optional approval Prod promotion from staging real traffic approval + progressive delivery + monitoring Config separation
- Use the same artifact; change only config:
- runtime env vars
- secrets from secret manager (Vault/AWS/GCP/Azure)
- per-env values files (Helm) or overlays (Kustomize)
GitOps option
- CI builds/publishes image.
- CD repo updates “desired state” (image digest) for dev→staging→prod via PRs, with approvals.
6) Progressive delivery: canary & blue-green
Canary (recommended for continuous releases)
- Shift traffic gradually: 5% → 25% → 50% → 100%.
- Run analysis at each step:
- HTTP 5xx rate
- latency p95/p99
- business KPIs (checkout success, etc.)
- Auto-rollback if metrics regress.
Blue-green (recommended for “flip” releases)
- Deploy “green” alongside “blue”.
- Verify health checks and smoke tests.
- Switch traffic all at once.
- Rollback by switching back to blue.
7) Rollback strategy (app + data)
Application rollback
- Kubernetes: rollback to previous ReplicaSet / previous Rollout revision.
- Always keep at least N previous versions available.
Database migrations
- Use expand/contract:
- expand schema (backward compatible)
- deploy app that uses both (or new path)
- migrate data
- contract (remove old columns) later
- Avoid irreversible migrations in the same deploy as risky app changes.
Automatic rollback triggers
- Failed health checks/readiness
- Error budget burn rate / SLO breach
- Increased crash loops
- Synthetic checks fail
8) Approvals & auditability
Where approvals belong
- PR approvals for code changes into
main. - Environment approvals for production promotion/deploy (and optionally staging).
Controls
- Protected environments (GitHub/GitLab/Azure) requiring:
- specific approvers or groups
- change window restrictions
- mandatory linked ticket/change request
- “Break-glass” manual deploy path, logged and restricted.
Example configs
A) GitHub Actions: build once + promote with environment approvals
.github/workflows/ci.yml(PR + main build/test + publish image)name: ci on: pull_request: push: branches: [ "main" ] workflow_dispatch: permissions: contents: read packages: write id-token: write # for OIDC signing (cosign) jobs: test: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - uses: actions/setup-node@v4 with: node-version: "20" cache: "npm" - run: npm ci - run: npm run lint - run: npm test -- --ci build_image: needs: [test] runs-on: ubuntu-latest outputs: image: ${{ steps.meta.outputs.image }} digest: ${{ steps.push.outputs.digest }} steps: - uses: actions/checkout@v4 - name: Log in to GHCR uses: docker/login-action@v3 with: registry: ghcr.io username: ${{ github.actor }} password: ${{ secrets.GITHUB_TOKEN }} - name: Compute tags id: meta run: | IMAGE=ghcr.io/${{ github.repository }}/app echo "image=$IMAGE" >> $GITHUB_OUTPUT echo "sha_tag=sha-${GITHUB_SHA::7}" >> $GITHUB_OUTPUT - name: Build and push id: push uses: docker/build-push-action@v6 with: context: . push: true tags: | ${{ steps.meta.outputs.image }}:${{ steps.meta.outputs.sha_tag }} provenance: true sbom: true # Optional: sign image with cosign using GitHub OIDC - name: Install cosign uses: sigstore/cosign-installer@v3 - name: Sign image (by digest) run: | cosign sign --yes ${{ steps.meta.outputs.image }}@${{ steps.push.outputs.digest }}.github/workflows/deploy.yml(promote same digest to dev/staging/prod)Use GitHub Environments with protection rules:
production: required reviewersstaging: optional reviewersdev: no reviewers
name: deploy on: workflow_run: workflows: ["ci"] types: [completed] workflow_dispatch: inputs: image: description: "Image ref (with digest preferred)" required: true jobs: resolve: if: ${{ github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success' }} runs-on: ubuntu-latest outputs: image_ref: ${{ steps.out.outputs.image_ref }} steps: - id: out run: | if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then echo "image_ref=${{ inputs.image }}" >> $GITHUB_OUTPUT else # Example: deploy by tag (better: store digest as artifact/output and fetch it here) echo "image_ref=ghcr.io/${{ github.repository }}/app:sha-${{ github.event.workflow_run.head_sha }}" >> $GITHUB_OUTPUT fi deploy_dev: needs: [resolve] runs-on: ubuntu-latest environment: dev steps: - uses: actions/checkout@v4 - name: Deploy to dev run: | ./deploy.sh dev "${{ needs.resolve.outputs.image_ref }}" deploy_staging: needs: [deploy_dev, resolve] runs-on: ubuntu-latest environment: staging steps: - uses: actions/checkout@v4 - name: Deploy to staging run: | ./deploy.sh staging "${{ needs.resolve.outputs.image_ref }}" - name: Smoke tests run: | ./smoke-test.sh https://staging.example.com deploy_prod: needs: [deploy_staging, resolve] runs-on: ubuntu-latest environment: production # configure required reviewers in repo settings steps: - uses: actions/checkout@v4 - name: Deploy to prod (canary/blue-green handled by platform) run: | ./deploy.sh prod "${{ needs.resolve.outputs.image_ref }}"Tip: Prefer deploying by digest (
image@sha256:...) to guarantee immutability.
B) GitLab CI: stages + manual prod approval
.gitlab-ci.ymlstages: [validate, test, build, scan, deploy] variables: IMAGE: "$CI_REGISTRY_IMAGE/app" validate: stage: validate image: node:20 script: - npm ci - npm run lint unit_test: stage: test image: node:20 script: - npm ci - npm test -- --ci artifacts: when: always reports: junit: junit.xml build_image: stage: build image: docker:27 services: [docker:27-dind] script: - docker login -u "$CI_REGISTRY_USER" -p "$CI_REGISTRY_PASSWORD" "$CI_REGISTRY" - docker build -t "$IMAGE:sha-$CI_COMMIT_SHORT_SHA" . - docker push "$IMAGE:sha-$CI_COMMIT_SHORT_SHA" rules: - if: $CI_COMMIT_BRANCH == "main" deploy_dev: stage: deploy script: - ./deploy.sh dev "$IMAGE:sha-$CI_COMMIT_SHORT_SHA" environment: name: dev rules: - if: $CI_COMMIT_BRANCH == "main" deploy_staging: stage: deploy script: - ./deploy.sh staging "$IMAGE:sha-$CI_COMMIT_SHORT_SHA" - ./smoke-test.sh https://staging.example.com environment: name: staging rules: - if: $CI_COMMIT_BRANCH == "main" deploy_prod: stage: deploy script: - ./deploy.sh prod "$IMAGE:sha-$CI_COMMIT_SHORT_SHA" environment: name: production when: manual allow_failure: false rules: - if: $CI_COMMIT_BRANCH == "main"
C) Kubernetes Canary example with Argo Rollouts (can auto-rollback)
rollout.yamlapiVersion: argoproj.io/v1alpha1 kind: Rollout metadata: name: app spec: replicas: 6 selector: matchLabels: app: app template: metadata: labels: app: app spec: containers: - name: app image: ghcr.io/org/repo/app@sha256:REPLACEME ports: - containerPort: 8080 strategy: canary: canaryService: app-canary stableService: app-stable trafficRouting: nginx: stableIngress: app-ingress steps: - setWeight: 5 - pause: { duration: 60s } - analysis: templates: - templateName: success-rate - setWeight: 25 - pause: { duration: 120s } - analysis: templates: - templateName: latency - setWeight: 100Analysis template sketch (Prometheus-based) would define thresholds; Rollouts will abort and rollback if metrics fail.
What I need from you to tailor this to your stack
- Platform: GitHub/GitLab/Azure/Jenkins?
- Runtime: Kubernetes, ECS, VMs, serverless?
- Languages/build tools (Node/Java/.NET/Go)?
- Compliance constraints (SOX/PCI, change tickets, required separation of duties)?
With those, I can provide a concrete end-to-end pipeline (including SBOM/provenance/signing, ephemeral PR envs, and either GitOps promotion or direct deploy).
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.