deployment-pipeline-design – skill DevOps & Hạ tầng cho agent AI

Deployment Pipeline Design

Architecture patterns for multi-stage CI/CD pipelines with approval gates and deployment strategies.

Do not use this skill when

• The task is unrelated to deployment pipeline design
• You need a different domain or tool outside this scope

Instructions

• Clarify goals, constraints, and required inputs.
• Apply relevant best practices and validate outcomes.
• Provide actionable steps and verification.

Purpose

Design robust, secure deployment pipelines that balance speed with safety through proper stage organization and approval workflows.

Use this skill when

• Design CI/CD architecture
• Implement deployment gates
• Configure multi-environment pipelines
• Establish deployment best practices
• Implement progressive delivery

Pipeline Stages

Standard Pipeline Flow

┌─────────┐   ┌──────┐   ┌─────────┐   ┌────────┐   ┌──────────┐
│  Build  │ → │ Test │ → │ Staging │ → │ Approve│ → │Production│
└─────────┘   └──────┘   └─────────┘   └────────┘   └──────────┘

Detailed Stage Breakdown

1. Source - Code checkout

2. Build - Compile, package, containerize

3. Test - Unit, integration, security scans

4. Staging Deploy - Deploy to staging environment

5. Integration Tests - E2E, smoke tests

6. Approval Gate - Manual approval required

7. Production Deploy - Canary, blue-green, rolling

8. Verification - Health checks, monitoring

9. Rollback - Automated rollback on failure

Approval Gate Patterns

Pattern 1: Manual Approval

# GitHub Actions
production-deploy:
  needs: staging-deploy
  environment:
    name: production
    url: https://app.example.com
  runs-on: ubuntu-latest
  steps:
    - name: Deploy to production
      run: |
        # Deployment commands

Pattern 2: Delayed Deployment

# GitLab CI
deploy:production:
  stage: deploy
  script:
    - deploy.sh production
  environment:
    name: production
  when: delayed
  start_in: 30 minutes
  only:
    - main

Pattern 3: Multi-Approver

Configure the platform's actual protected-environment approval policy and verify who may approve or bypass it. For Azure, bind manual validation to the installed task version and its supported agentless job rather than placing it in an arbitrary deployment hook. Record required approvers and evidence; notifications alone are not approval.

Deployment Strategies

1. Rolling Deployment

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
spec:
  replicas: 10
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxSurge: 2
      maxUnavailable: 1

Characteristics:

• Gradual rollout
• Availability depends on readiness, capacity and compatible changes
• Easy rollback
• Best for most applications

2. Blue-Green Deployment

# Blue (current)
kubectl apply -f blue-deployment.yaml
kubectl patch service my-app --type merge -p '{"spec":{"selector":{"app":"my-app","version":"blue"}}}'

# Green (new)
kubectl apply -f green-deployment.yaml
# Test green environment
kubectl patch service my-app --type merge -p '{"spec":{"selector":{"app":"my-app","version":"green"}}}'

# Rollback if needed
kubectl patch service my-app --type merge -p '{"spec":{"selector":{"app":"my-app","version":"blue"}}}'

Characteristics:

• Instant switchover
• Easy rollback
• Doubles infrastructure cost temporarily
• Good for high-risk deployments

3. Canary Deployment

apiVersion: argoproj.io/v1alpha1
kind: Rollout
metadata:
  name: my-app
spec:
  replicas: 10
  strategy:
    canary:
      steps:
      - setWeight: 10
      - pause: {duration: 5m}
      - setWeight: 25
      - pause: {duration: 5m}
      - setWeight: 50
      - pause: {duration: 5m}
      - setWeight: 100

Characteristics:

• Gradual traffic shift
• Risk mitigation
• Real user testing
• Requires service mesh or similar

4. Feature Flags

from flagsmith import Flagsmith

flagsmith = Flagsmith(environment_key="API_KEY")

if flagsmith.has_feature("new_checkout_flow"):
    # New code path
    process_checkout_v2()
else:
    # Existing code path
    process_checkout_v1()

Characteristics:

• Deploy without releasing
• A/B testing
• Instant rollback
• Granular control

Pipeline Orchestration

Multi-stage implementation contract

Build once, record the commit and immutable artifact digest, run tests/security checks on that artifact, deploy to authorized staging, run integration checks, obtain configured production approval, then deploy that same digest with bounded progressive traffic. Each job must establish its own checkout, locked tools, scoped credentials and artifact transfer. Verify rollout and application metrics before promotion; do not immediately promote a canary and thereby skip its observation window. Notifications require an explicitly configured authorized recipient and destination.

Pipeline Best Practices

1. Fail fast - Run quick tests first

2. Parallel execution - Run independent jobs concurrently

3. Caching - Cache dependencies between runs

4. Artifact management - Store build artifacts

5. Environment parity - Keep environments consistent

6. Secrets management - Use secret stores (Vault, etc.)

7. Deployment windows - Schedule deployments appropriately

8. Monitoring integration - Track deployment metrics

9. Rollback automation - Auto-rollback on failures

10. Documentation - Document pipeline stages

Rollback Strategies

Automated Rollback

deploy-and-verify:
  steps:
    - name: Deploy new version
      run: kubectl apply -f k8s/

    - name: Wait for rollout
      run: kubectl rollout status deployment/my-app

    - name: Health check
      id: health
      run: |
        for i in {1..10}; do
          if curl -sf https://app.example.com/health; then
            exit 0
          fi
          sleep 10
        done
        exit 1

    - name: Rollback on failure
      if: failure()
      run: kubectl rollout undo deployment/my-app

Manual Rollback

# List revision history
kubectl rollout history deployment/my-app

# Rollback to previous version
kubectl rollout undo deployment/my-app

# Rollback to specific revision
kubectl rollout undo deployment/my-app --to-revision=3

Monitoring and Metrics

Key Pipeline Metrics

• Deployment Frequency - How often deployments occur
• Lead Time - Time from commit to production
• Change Failure Rate - Percentage of failed deployments
• Mean Time to Recovery (MTTR) - Time to recover from failure
• Pipeline Success Rate - Percentage of successful runs
• Average Pipeline Duration - Time to complete pipeline

Integration with Monitoring

Define a measured baseline, minimum traffic, observation interval and justified thresholds. Query a verified metrics endpoint with bounded timeouts and validate the provider status, result shape and numeric values. Missing/empty/malformed metrics are an incomplete verification, never a healthy result. Stop promotion and escalate when verification is unavailable. A fixed sleep and a shell comparison cannot establish deployment health.

Approval and orchestration checklist

Bind the artifact to an immutable commit/image digest, verify the exact target and existing deploy identity, require protected environment approval for production, and record health criteria and an explicit rollback revision. Every job runs on its own checkout/runtime; transfer verified artifacts rather than assuming the previous runner filesystem remains available. These are schematic examples, not a complete credential or orchestration template. No unbundled playbook is needed.

Related Skills

• github-actions-templates - For GitHub Actions implementation
• gitlab-ci-patterns - For GitLab CI implementation
• secrets-management - For secrets handling

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.