MLA-C01 Question 252
Select 3You are deploying a new version of your machine learning model using Amazon SageMaker. To minimize the risk of downtime and ensure you can quickly revert to the previous version if issues occur, which deployment strategies should you consider?
- A
Blue/Green deployment
- B
Canary deployment
- C
Direct replacement deployment
- D
Shadow testing
- E
All-at-once deployment
Show answer and explanation
Correct answers: A, B, D
Explanation
Blue/Green deployment, Canary deployment, and Shadow testing are considered best practices when deploying machine learning models, as they minimize risk and provide mechanisms for testing and rollback in case of issues. Direct replacement and All-at-once deployments are riskier and do not align with deployment best practices.
- A. Correct.
Blue/Green deployment creates two separate environments (blue and green), allowing you to test the new model in the green environment while the blue environment remains live. It also makes rollback straightforward by switching traffic back to the blue environment if issues arise.
- B. Correct.
Canary deployment gradually shifts a small percentage of traffic to the new model, allowing you to monitor its performance before rolling it out completely. This strategy minimizes risk and provides an easy rollback option.
- C. Incorrect.
Direct replacement deployment involves replacing the existing model with the new one immediately. This approach has a high risk of downtime and does not provide a straightforward rollback strategy.
- D. Correct.
Shadow testing involves sending traffic to the new model without impacting the live environment. While this is not a deployment strategy that directly serves traffic, it is useful for testing the new model’s performance and ensuring it behaves as expected before deployment.
- E. Incorrect.
All-at-once deployment sends 100% of traffic to the new model immediately. While this approach is fast, it carries a high risk of impacting users if issues occur, and rollback can be complicated.