MLA-C01 Question 251
Select 3You are deploying a new version of a machine learning model to a SageMaker endpoint that is already serving live traffic. You want to minimize downtime and ensure that if issues are detected with the new version, you can quickly revert to the previous version. Which of the following deployment strategies should you consider implementing?
- A
Blue/Green deployment strategy
- B
Canary deployment strategy
- C
Direct replacement of the endpoint with the new model
- D
A/B testing deployment strategy
- E
Rollback to a previous endpoint configuration using SageMaker's model versioning
Show answer and explanation
Correct answers: A, B, E
Explanation
When deploying a new version of a model, it is critical to minimize downtime and have a rollback strategy in place. Blue/Green and Canary deployments are established best practices for achieving these goals, as they allow for testing and validation before fully routing traffic to the new version. Additionally, SageMaker's support for model versioning provides a straightforward rollback mechanism. Avoid direct replacement of the endpoint, as it can lead to unplanned downtime and lacks a robust rollback plan. A/B testing is not a deployment strategy but rather a method for comparing different models.
- A. Correct.
Blue/Green deployment is a best practice for minimizing downtime. It involves deploying the new version to a separate environment and switching traffic only after validating the new version.
- B. Correct.
Canary deployment gradually shifts traffic to the new version, allowing you to monitor performance and roll back if issues are detected, making it a reliable strategy.
- C. Incorrect.
Direct replacement of the endpoint with the new model introduces significant risks, including downtime and no immediate rollback option if issues arise.
- D. Incorrect.
A/B testing is focused on comparing the performance of two or more models rather than ensuring a reliable deployment with rollback capabilities.
- E. Correct.
SageMaker's model versioning allows you to track and manage different versions of your model, enabling a quick rollback to a previous version in case of deployment issues.