CD & Deployment Strategies#

Continuous Delivery (CD) ensures that code is always in a deployable state. Continuous Deployment takes this further by automatically releasing every change that passes the pipeline. The deployment strategy you choose determines how risk is managed during releases.

Continuous Delivery vs Continuous Deployment#

graph LR
    A[Commit] --> B[Build & Test]
    B --> C[Deploy to Staging]
    C --> D{Manual Approval?}
    D -->|Continuous Delivery| E[Deploy to Prod]
    D -->|Continuous Deployment| E
    
    style D fill:#fff3cd

Deployment Strategies#

Blue-Green Deployment#

Maintain two identical production environments. Route traffic entirely to one (blue) while deploying to the other (green). Switch when ready.

graph TD
    LB[Load Balancer]
    LB -->|100% traffic| Blue[Blue Environment v1.2]
    LB -.->|0% traffic| Green[Green Environment v1.3]
    
    style Blue fill:#e3f2fd
    style Green fill:#e8f5e1

Canary Deployment#

Route a small percentage of traffic to the new version. Gradually increase if metrics are healthy.

graph TD
    LB[Load Balancer]
    LB -->|95% traffic| V1[v1.2 - Stable]
    LB -->|5% traffic| V2[v1.3 - Canary]
    
    style V1 fill:#e3f2fd
    style V2 fill:#fff3cd

Rolling Deployment#

Replace instances of the old version with the new version one at a time (or in batches).

graph TD
    subgraph "Rolling Update (4 instances)"
        A["Instance 1: v1.3 ✓"]
        B["Instance 2: v1.3 ✓"]
        C["Instance 3: v1.2 → v1.3"]
        D["Instance 4: v1.2"]
    end
    
    style A fill:#e8f5e1
    style B fill:#e8f5e1
    style C fill:#fff3cd
    style D fill:#e3f2fd

Recreate (Big Bang)#

Stop all old instances, deploy new ones. Simple but causes downtime.

Strategy Comparison#

Feature Flags#

Feature flags decouple deployment from release. Code is deployed but features are enabled/disabled via configuration:

# Feature flag example
from feature_flags import is_enabled

def get_search_results(query: str, user: User) -> Results:
    if is_enabled("new-search-algorithm", user=user):
        return new_search(query)
    else:
        return legacy_search(query)

Feature Flag Types#

Feature Flag Best Practices#

• Clean up — Remove flags after full rollout (technical debt otherwise)
• Default safe — Flag off = old behaviour (safe fallback)
• Monitor — Track metrics per flag state
• Limit scope — Don’t nest flags or create combinatorial explosion

A/B Testing in Deployments#

A/B testing uses deployment infrastructure to measure business impact:

Rollback Strategies#

When a deployment goes wrong, you need a fast path back to stability:

Automated Rollback Conditions#

# Example: Auto-rollback criteria
rollback:
  triggers:
    - metric: error_rate_5xx
      threshold: "> 1%"
      window: 5m
    - metric: latency_p99
      threshold: "> 2000ms"
      window: 5m
    - metric: health_check
      condition: "3 consecutive failures"
  action: route_traffic_to_previous_version

Database Migrations in CD#

Database changes are the hardest part of continuous deployment because they’re often irreversible:

Expand-Contract Pattern#

graph LR
    A["Step 1: Expand
    Add new column (nullable)"] --> B["Step 2: Migrate
    Code writes to both old & new"] --> C["Step 3: Contract
    Remove old column"]

Deployment Automation#

Environment Promotion#

graph LR
    A[Build Artifact] --> B[Deploy to Dev]
    B --> C[Auto-tests pass]
    C --> D[Deploy to Staging]
    D --> E[Smoke tests + approval]
    E --> F[Deploy to Production]
    F --> G[Post-deploy verification]

Deployment Pipeline Configuration#

# Environment promotion example
deploy-staging:
  stage: deploy
  environment: staging
  script:
    - deploy --env staging --version $CI_COMMIT_SHA
    - run-smoke-tests --env staging
  rules:
    - if: $CI_COMMIT_BRANCH == "main"

deploy-production:
  stage: deploy
  environment: production
  script:
    - deploy --env production --version $CI_COMMIT_SHA
    - run-post-deploy-checks --env production
  rules:
    - if: $CI_COMMIT_BRANCH == "main"
      when: manual  # Human approval gate
  needs: [deploy-staging]

Progressive Delivery#

Progressive delivery combines deployment strategies with observability:

Tools that enable progressive delivery:

• Argo Rollouts — Kubernetes-native progressive delivery
• Flagger — Automated canary analysis for Kubernetes
• AWS CodeDeploy — Canary/linear deployment for ECS and Lambda
• LaunchDarkly / Unleash — Feature flag management

Key Takeaways#

• Blue-green gives instant rollback at the cost of double infrastructure
• Canary provides gradual risk reduction with granular traffic control
• Rolling is the simplest zero-downtime strategy for stateless services
• Feature flags decouple deployment from release — deploy anytime, release when ready
• Automated rollback based on health metrics is the safety net every CD pipeline needs
• Database migrations require the expand-contract pattern for safe continuous deployment
• Choose your strategy based on: risk tolerance, infrastructure cost, and observability maturity