MLA-C01 Question 249
Select 3You are deploying a machine learning model on AWS SageMaker and want to ensure minimal downtime during updates, while also being able to roll back to a previous version if needed. Which of the following practices should you implement?
- A
Use SageMaker Endpoint versioning to manage different versions of the model.
- B
Deploy the new model to a new SageMaker Endpoint and perform a weighted traffic shift.
- C
Manually update the existing SageMaker Endpoint with the new model without creating backups.
- D
Leverage Blue/Green deployment by creating a new endpoint configuration and switching traffic gradually.
- E
Implement Canary deployment by routing a small portion of traffic to the new model version before full rollout.
Show answer and explanation
Correct answers: B, D, E
Explanation
To ensure minimal downtime and allow rollback for machine learning model deployments in AWS SageMaker, you should use strategies like weighted traffic shifting, Blue/Green deployment, or Canary deployment. These approaches allow you to validate new models incrementally while keeping the previous version available for fallback. Avoid directly overwriting existing endpoints or relying solely on version management without proper deployment strategies.
- A. Incorrect.
SageMaker does not directly support endpoint versioning. While you can manage versions externally (for example, tagging models in S3), this is not sufficient for deployment or rollback strategies.
- B. Correct.
Deploying the new model to a new SageMaker Endpoint and gradually shifting traffic allows you to validate the new model while maintaining the existing version for rollback purposes.
- C. Incorrect.
Manually updating the existing endpoint without backups introduces risks because there is no easy way to roll back to the previous model in case of issues.
- D. Correct.
Blue/Green deployment is a best practice for ensuring safe model updates. It allows you to create a new endpoint configuration and gradually redirect traffic, reducing downtime and enabling rollback.
- E. Correct.
Canary deployment is a recommended strategy where you test the new model with a small portion of traffic before rolling it out fully, ensuring that potential issues can be caught early.