A company runs an application as an Amazon ECS service on an AWS Fargate cluster.

The requirement is specifically to have ECS automatically restore the last known good deployment when a new deployment fails, while the service is using the rolling update deployment type. Amazon ECS supports this through the deployment circuit breaker with the option to rollback on failures.

When enabled, the deployment circuit breaker monitors the deployment and detects failure conditions such as tasks repeatedly failing health checks, not reaching steady state, or being unable to start successfully. With rollback enabled, ECS automatically stops the failing deployment and reverts the service to the most recently completed (stable) task definition, restoring service stability without manual intervention.

Option D matches exactly: keep the rolling deployment type and enable circuit breaker + rollback. This is minimal change and directly addresses the behavior observed.

Options A and B are unrelated: service autoscaling adjusts desired count based on metrics and does not roll back failed deployments.

Option C could also achieve rollback behavior, but switching to blue/green introduces additional components (AWS CodeDeploy integration, target groups, and traffic shifting), which is unnecessary when the requirement can be met within rolling deployments.

Therefore, enabling the ECS deployment circuit breaker with rollback on failures for a rolling deployment is the correct solution.