Certified Machine Learning ProfessionalDemo

Question

A Machine Learning Engineer has an existing recommendation AI model deployed on a Databricks Model Serving endpoint. Their team has created a new model and tested it thoroughly in staging. It is now ready for deployment on their platform, which receives high traffic consistently throughout every day. The engineer chooses the blue-green model-deployment method. They first configure the two models as separate entities for the Model Serving Endpoint.

What should they do next?

A Use the AI Gateway Fallback feature. The first entity should be the old model and the second one listed should be the new model.
B Use the AI Gateway Fallback feature. The first entity should be the new model and the second one listed should be the old model.
C Change the percentage of traffic where the new model gets 5% of the traffic and the old model will receive 95% of the traffic.
D Change the percentage of traffic where the new model gets 95% of the traffic and the old model will receive 5% of the traffic.

Explanation

A blue-green deployment performs an atomic switch of all traffic to the new (green) model while keeping the old (blue) model available for instant rollback—it is not a gradual traffic split. On a Databricks Model Serving endpoint with the two models configured as separate served entities, this is achieved with the AI Gateway Fallback feature, listing the new model first (primary, receiving traffic) and the old model second (fallback for automatic rollback on errors) — option B. Option A reverses the order so the old model would keep serving. Options C and D describe canary deployments (gradual percentage-based traffic splits), which is a different strategy than blue-green.

Learn more

Certified Machine Learning ProfessionalDemo

AstroTutor

Notes

Readiness

QuestionQ63

Community Discussion