MLA-C01 Question 351
Single answerYou are deploying a new version of a machine learning model on Amazon SageMaker. The model serves real-time predictions and has strict uptime requirements. To minimize the risk to production traffic during deployment, you decide to send a small percentage of the traffic to the new version while continuously monitoring its performance. What deployment strategy should you use?
- A
Blue/Green deployment
- B
Canary deployment
- C
All-at-once deployment
- D
Linear deployment
Show answer and explanation
Correct answer: B
Explanation
Canary deployment is the best choice for this scenario as it allows you to send a small percentage of traffic to the new model initially, monitor its performance, and then gradually increase traffic. This minimizes the risk to production traffic while ensuring continuity of service.
- A. Incorrect.
Blue/Green deployment involves creating a completely separate environment for the new version and switching all traffic to it at once. While it ensures rollback is fast, it does not involve gradually shifting traffic, which is required in this scenario.
- B. Correct.
Canary deployment sends a small percentage of traffic to the new version initially, allowing you to monitor its performance before routing more traffic. This approach is suitable for minimizing risks to production traffic, as described in the scenario.
- C. Incorrect.
All-at-once deployment replaces the old version with the new version immediately. This approach is risky for scenarios requiring uptime and gradual traffic shifts, as it does not allow incremental monitoring.
- D. Incorrect.
Linear deployment gradually increases traffic to the new version over a specified period. While it achieves gradual traffic shifts, it does not start with a small percentage of traffic like Canary deployment, which is explicitly required in this scenario.