Question

A team deploys a containerized microservice to Amazon ECS on Fargate using an AWS CodePipeline that invokes a standard ECS deploy action (rolling update via the ECS service, not CodeDeploy blue/green). Recently a bad image passed unit tests but crashed on startup in production due to a missing runtime config. The failed tasks looped indefinitely, the deployment hung, and engineers had to manually roll back. The team wants the platform to automatically detect a failed rollout and revert the ECS service to the last healthy task definition without adding a separate CodeDeploy blue/green setup. What is the MOST effective change?

Accepted Answer

Enable the ECS deployment circuit breaker with rollback on the service's deployment configuration so ECS automatically stops and reverts to the last successful task definition when new tasks fail to reach a steady state.

Answer

Add a CloudWatch alarm on service CPUUtilization and configure the CodePipeline stage to fail when the alarm triggers, prompting a manual redeploy of the previous revision.

Answer

Increase the ECS service minimum healthy percent to 100 and maximum percent to 200 so failing tasks are replaced before old tasks are drained.

Answer

Configure a CodeDeploy canary traffic shift with an alarm-based rollback on the ECS service to catch the startup crash before full rollout.