AI-300 Practice Question 54: Online Endpoint Traffic Split
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 6 · question 4 of 10
A managed online endpoint serves a model from a deployment named blue. The team deployed an improved model as green and wants 10 percent of live requests to be answered by green while blue keeps the other 90 percent. Users must actually receive responses from both deployments. Which configuration meets the requirement?
Show answer and explanation
Answer:
A managed online endpoint can hold several deployments and split live traffic among them with the endpoint's traffic values, for example az ml online-endpoint update --traffic "blue=90 green=10". The values must total 100 or less, and each user request is answered by the deployment that receives it (option B). This is the basis of a gradual blue/green rollout.
- A is wrong because mirrored requests are copied to green, but its responses are discarded. Every live response would still come from blue. - C is wrong because instance counts set capacity, not routing. Requests are distributed by the endpoint's traffic settings, not by how many instances each deployment has. - D is wrong because this moves routing into every client and creates a second endpoint to manage. An endpoint already splits traffic across its deployments.
Reference: Safe rollout for online endpoints