CS Techie
Comments
courseAWS
AWS lesson

Deployment Strategies in AWS

Below are the common deployment strategies used in AWS, when to use each, and how they map to AWS services. Each entry includes a short example and key considerations.


Blue/Green Deployment

  • What: Maintain two separate environments (blue = current, green = new). Switch traffic to green when ready.
  • When: Zero-downtime upgrades, risk-averse releases, major changes.
  • AWS services: Elastic Beanstalk, CodeDeploy (blue/green), ECS with ALB + Route53, CloudFormation with separate stacks.
  • Example: Deploy new version to green environment, run tests, update Route53 or swap ALB target groups to point to green.
  • Considerations: Requires duplicate capacity. Clean rollback by switching back.


Canary Deployment

  • What: Shift a small portion of traffic to the new version, then gradually increase it.
  • When: Validate behavior in production with minimal user impact.
  • AWS services: CodeDeploy (traffic shifting), API Gateway stage variables & Canary settings, App Mesh, Lambda alias with weighted traffic.
  • Example: Send 5% traffic to new version for 30 minutes, monitor metrics, then increase to 50% then 100%.
  • Considerations: Needs robust monitoring and automated rollbacks.


Rolling Deployment

  • What: Replace instances or tasks in small batches until the fleet is updated.
  • When: Stateful deployments or when duplicate environments are not feasible.
  • AWS services: Auto Scaling Groups, ECS rolling updates, Elastic Beanstalk rolling deployments.
  • Example: Update 20% of instances at a time; ensure health checks before continuing.
  • Considerations: May expose mixed-version traffic; plan health checks and capacity.


Rolling with Additional Batch

  • What: Add extra capacity, then perform rolling updates, finally remove extra capacity.
  • When: Avoid capacity shortage during rolling updates.
  • AWS services: Auto Scaling Groups + CloudFormation, ECS with extra tasks.
  • Example: Increase ASG desired count by 20%, update instances in batches, then scale down.
  • Considerations: Uses temporary extra resources and cost.


Immutable Deployment

  • What: Create new instances (or containers) with the new version and switch traffic to them. Old infra remains unchanged.
  • When: Maximum safety and reproducibility; eliminate in-place changes.
  • AWS services: Auto Scaling with new launch template versions, CodeDeploy immutable, ECS with new tasks + new target group swap.
  • Example: Provision new ASG/launch template with the new AMI, update Load Balancer to new target group, then terminate old ASG.
  • Considerations: Higher temporary resource usage; simplifies rollback.


All-at-Once (Big Bang)

  • What: Deploy new version to all instances simultaneously.
  • When: Small environments, non-critical apps, fast releases where downtime is acceptable.
  • AWS services: Simple deploy via console/CLI, Elastic Beanstalk all-at-once option.
  • Example: Replace application on every server in one step.
  • Considerations: Highest risk; plan for quick rollback.


Shadow (Dark Launch) Deployment

  • What: Send duplicate traffic to the new version for testing without returning its responses to users.
  • When: Performance testing, telemetry, validating side-effects safely.
  • AWS services: ALB + traffic mirroring solutions, VPC traffic mirroring, App Mesh, API Gateway + Lambda (duplicate invocation).
  • Example: Mirror 100% of requests to shadow service; analyze behavior and logs.
  • Considerations: No user impact; needs tooling to capture and analyze mirrored requests.

A/B Testing and Feature Toggles

  • What: Serve different feature variants to subsets of users; use feature flags to control behavior.
  • When: UX experiments, progressive rollouts, targeted features.
  • AWS services: AppConfig (feature flags), Lambda, API Gateway, CloudFront headers + Lambda@Edge, third-party feature flagging tools.
  • Example: 10% of users see new UI controlled by AppConfig flags.
  • Considerations: Requires telemetry and experiment analysis. Keep flags short-lived and manageable.

Lambda-Specific Strategies

  • Aliases & Versions: Use function versions and aliases with weighted routing to implement canary/linear shifts.
  • Safe rollback: Repoint alias to prior version.
  • Tools: Lambda traffic-shifting with CodeDeploy or console.



CI/CD & Automation Tools (How to Implement)

  • Source control: CodeCommit, GitHub, GitLab.
  • Build: CodeBuild, Jenkins, GitHub Actions.
  • Artifact & registry: CodeArtifact, ECR, S3.
  • Deploy/orchestration: CodePipeline + CodeDeploy, CloudFormation, CDK, Elastic Beanstalk, ECS/EKS controllers.
  • Traffic control: ALB target groups, Route53 weighted records, API Gateway stages, Lambda aliases, App Mesh.



Best Practices

  • Automate deployments and rollbacks via CI/CD pipelines.
  • Use health checks and readiness probes before shifting traffic.
  • Implement robust observability: CloudWatch metrics, Logs, X-Ray, synthetic tests.
  • Define SLOs and automated rollback triggers (error rate, latency, saturation).
  • Use feature flags for behavior toggles and rapid rollback without redeploying.
  • Test strategies in staging that closely mirrors production.
  • Encrypt secrets and use IAM least privilege for deployment roles.
  • Manage capacity and cost; choose immutable or blue/green when safety outweighs cost.


Quick Decision Guide

  • Need zero downtime and safe rollback → Blue/Green or Immutable.
  • Need gradual verification in production → Canary (or linear canary).
  • Limited resources, small app → Rolling or All-at-Once.
  • Serverless (Lambda) → Alias weighted routing or CodeDeploy traffic shifting.
  • Validate without affecting users → Shadow deployments or mirrored traffic.