A production SageMaker endpoint is configured with automatic scaling. During testing, the team observes that new instances are launched but traffic is routed to them before they finish initializing. You need to prevent new instances from receiving traffic until they are ready. What change should you make?

The issue describes new instances receiving traffic before they are fully initialized, leading to potential errors or performance degradation. Increasing the cooldown period for scale-out activities ensures that after a scaling event, newly launched instances have sufficient time to initialize and become ready to serve traffic before Amazon SageMaker Auto Scaling routes requests to them. Decreasing the cooldown period for scale-in activities and increasing maximum capacity would not address the problem of instances receiving traffic prematurely. A SageMaker multi-model endpoint is designed for hosting multiple models on a single endpoint, not for managing instance readiness during scaling. Fronting the endpoint with API Gateway and Lambda adds complexity and custom logic that is unnecessary when SageMaker Auto Scaling provides a direct solution through cooldown periods.