What changed
AWS has introduced a reusable CDK construct that addresses deployment failures when updating SageMaker AI endpoints with auto scaling configurations. The issue arises when updating endpoint configurations through AWS CloudFormation or AWS CDK, as the system fails to track Application Auto Scaling resources during deregistration.
What this means in practice
The new CDK construct replaces native CloudFormation resources with custom resources that orchestrate deregistering and re-registering auto scaling during endpoint updates. This allows users to update both endpoint configurations and auto scaling parameters in a single deployment without manual intervention. The solution ensures that the endpoint update process waits for the auto scaling resources to be properly deregistered and re-registered, preventing deployment failures and rollback issues.
Limitations and availability
The solution is available through a GitHub repository and requires AWS CDK v2, Python 3.12 or later, and specific AWS services such as SageMaker AI, Lambda, IAM, and Application Auto Scaling. The deployed stack includes multiple Lambda functions and Step Functions state machines, which are generated by the provider framework and should be accounted for during resource reviews. The construct is designed for users who need to manage auto scaling configurations alongside endpoint updates in a scalable and automated manner.
