Question

Which deployment strategy involves routing a small percentage of incoming production traffic to a new model version to validate performance before a full rollout?

Accepted Answer

The deployment strategy you are describing is called Canary Deployment. In this process, the system gradually shifts a small, predefined percentage of live production traffic from the current model version to a new model version. By exposing only a subset of users to the new model, engineers can monitor performance metrics, error rates, and system stability in a real-world environment. If the new model performs as expected, the traffic is incrementally increased until it handles one hundred percent of requests. If issues occur, the system can quickly roll back traffic to the previous version, minimizing the impact on the user base. This strategy is named after the historical practice of using canaries in coal mines to detect dangerous gases before they harmed human miners, serving as an early warning system for potential failure.

Home → All Courses → Engineering and Technology Courses → Machine Learning Engineering → Flashcard

Which deployment strategy involves routing a small percentage of incoming production traffic to a new model version to validate performance before a full rollout?

The deployment strategy you are describing is called Canary Deployment. In this process, the system gradually shifts a small, predefined percentage of live production traffic from the current model version to a new model version. By exposing only a sub....