Question

A company runs a containerized order-management API on Amazon ECS with AWS Fargate behind an Application Load Balancer. Deployments are performed by pushing new task definitions manually, and roughly 15% of releases cause elevated 5xx errors that are only noticed hours later when customers complain. The operations team wants to improve the deployment process so that faulty releases are automatically detected using application health signals and traffic is reverted without manual intervention, while keeping infrastructure changes version-controlled. Which approach best meets these requirements?

Accepted Answer

Migrate the deployment to AWS CodeDeploy with an ECS blue/green deployment configuration, associate CloudWatch alarms on ALB 5xx and target group health metrics as deployment stop conditions, and manage the pipeline with CodePipeline sourced from the version-controlled task definition.

Answer

Enable ECS rolling updates with a minimum healthy percent of 100 and maximum percent of 200 so that new tasks replace old ones gradually, and rely on ECS service health checks to stop the deployment.

Answer

Configure an Amazon EventBridge rule that triggers a Lambda function whenever CloudWatch detects elevated 5xx errors, and have the Lambda function redeploy the previous task definition.

Answer

Add a CloudWatch composite alarm on 5xx errors that publishes to an SNS topic subscribed by the operations team, and document a manual runbook for reverting to the previous task definition.