AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsHard
A data science team is developing a critical machine learning model for real-time anomaly detection. They are using an AWS CodePipeline-based MLOps pipeline for continuous integration and continuous delivery. A new model version is ready for deployment. The team wants to ensure that the new model undergoes a rigorous performance and stability check in a production-like environment, processing actual production traffic, but without directly impacting the real-time responses to end-users. Only after a successful observation period will the model be considered for full deployment. Which deployment strategy should be integrated into the CodePipeline?
- ABlue/Green Deployment
- BCanary Deployment
- CShadow Deployment
- DA/B Testing
Show answer & explanationAnswer & explanation
Correct answer: C. Shadow Deployment
Shadow deployment is the most suitable strategy here. It routes a copy of production traffic to the new model for evaluation without impacting the live responses to end-users. This allows for rigorous performance and stability checks in a production-like environment before any user-facing changes are made.
Why the other options are wrong
- A. Blue/Green deployment involves routing all (or a significant portion) of live traffic to the new model, which directly impacts end-users, contrary to the requirement of 'without directly impacting the real-time responses'.
- B. Canary deployment routes a small percentage of live user traffic (with user-facing impact) to the new model, gradually increasing it. While it's a gradual rollout, it still impacts users from the start, which isn't the primary goal here.
- D. A/B testing serves a small percentage of actual user responses from the new model, directly impacting those users, which is not desired for the initial rigorous check without impact.
Shadow Deployment
A deployment strategy where a new model processes a copy of live production traffic in parallel with the existing model, but its predictions are not used for actual user responses, allowing for non-disruptive validation.
- Routes a copy of production traffic to the new model.
- New model's predictions are not returned to users.
- Enables non-disruptive performance and stability testing.
Memory trick: Shadows secretly test new paths without changing the main road.