Databricks ML Associate Practice Question 10: Serving Traffic Split
Databricks Certified Machine Learning Associate · Test 1 · question 10 of 10
A real-time endpoint serves version 4 of prod.recs.ranker from Unity Catalog. The team wants 10% of live requests to go to version 6 so they can compare the two on production traffic. Client applications must keep calling the same endpoint URL, and there must be no downtime while the change rolls out.
What should they do?
Show answer and explanation
Answer:
A Mosaic AI Model Serving endpoint can serve several served entities at once. In the endpoint configuration, list both model versions under served_entities and add a traffic_config with routes that give 90% of traffic to version 4 and 10% to version 6. Clients keep calling the same URL. During the update, the old configuration keeps serving requests until the new one is ready, so there is no downtime.
Option A splits traffic by client team rather than by request, which biases the comparison. It also breaks the requirement that clients keep calling the same endpoint URL.
Option B assumes an alias controls routing. Aliases are registry pointers. The Model Serving docs describe traffic splitting only through the endpoint's served entities and traffic_config, so reassigning an alias does not shift traffic on an existing endpoint.
Option C sends 100% of traffic to the untested version, which is a full cutover rather than a 10% comparison. It also leaves nothing to compare against.
Reference: Serve multiple models to a model serving endpoint