MediCare AI wants to deploy a new model version to production using a blue/green strategy with minimal downtime and an automatic rollback if the new variant produces an elevated 5xx error rate. Which approach provides controlled traffic shifting and an automatic rollback mechanism within AWS SageMaker capabilities?

The correct option leverages SageMaker's native blue/green deployment capabilities. By defining two production variants (current and new model) within an EndpointConfig and initially routing all traffic to the current model (blue:100, green:0), you establish the blue/green setup. UpdateEndpointWeightsAndCapacities (or UpdateEndpoint with a new EndpointConfig) allows for controlled, incremental traffic shifting. Integrating CloudWatch alarms for 5xx errors with an EventBridge rule triggering a rollback Lambda that resets weights to the blue variant provides the essential automatic rollback mechanism, ensuring minimal downtime and quick recovery. The incorrect options have limitations: Launching a separate staging endpoint with an ALB is a viable blue/green strategy but is external to SageMaker's native endpoint management and doesn't fully utilize its built-in traffic shifting and rollback features for endpoints. The Model Registry automatic deployment feature with deploymentStrategy and errorThreshold is a planned feature but not currently available as described for automatic blue/green deployments with rollback in SageMaker. Directly calling CreateEndpoint after deregistering the old model would cause downtime and the rollback mechanism described is manual and reactive, not automatic or integrated for seamless blue/green.