AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsMedium
A financial services company is deploying a new fraud detection model. Due to the critical nature of the application, they need to validate the new model's performance against the existing production model using real-world traffic without impacting current users. They want to gradually shift a small percentage of live traffic to the new model, monitor its performance, and then progressively increase traffic if successful. Which deployment strategy is best suited for this requirement?
- AShadow Deployment
- BRolling Update
- CBlue/Green Deployment
- DCanary Deployment
Show answer & explanationAnswer & explanation
Correct answer: D. Canary Deployment
Canary deployment involves gradually rolling out a new version of an application or model to a small subset of users or traffic, monitoring its performance, and then incrementally increasing the traffic to the new version if it performs well. This matches the requirement of gradually shifting traffic and monitoring.
Why the other options are wrong
- A. Shadow deployment sends a copy of live traffic to the new model for evaluation without impacting the production model's responses, meaning no live users are served by the new model. The requirement is to 'gradually shift a small percentage of live traffic'.
- B. Rolling update is a general software deployment strategy where instances are updated one by one, but it doesn't specifically address the ML model validation and traffic shifting aspect as precisely as Canary deployment.
- C. Blue/Green deployment involves running two identical production environments (blue and green) and switching traffic completely from one to the other, which doesn't allow for gradual traffic shifting.
Canary Deployment
A deployment strategy where a new version of an application or model is incrementally rolled out to a small subset of users or traffic, monitored for performance, and then gradually expanded.
- Minimizes risk by limiting exposure of new version
- Allows for real-world performance validation
- Requires robust monitoring and rollback capabilities
- Traffic is gradually shifted, not copied (like shadow)
Memory trick: Canary flies first, then the flock follows.