Question

A GenAI engineer manages a production Model Serving endpoint that is actively serving a customer-facing chatbot. Leadership requires updating the endpoint to serve a newly registered model version, but the update must not cause any service interruption or dropped requests during the transition. Which approach correctly achieves a zero-downtime update using Databricks Model Serving?

Accepted Answer

Update the endpoint's served entity configuration to the new model version; Model Serving provisions the new version and shifts traffic only after it becomes ready, keeping the old version serving until then.

Answer

Delete the existing endpoint and immediately recreate it pointing to the new model version so the new deployment starts fresh.

Answer

Manually stop the endpoint, swap the model artifact in Unity Catalog, and restart the endpoint with the same name.

Answer

Set the endpoint to scale-to-zero so it shuts down, then bring it back up bound to the new model version.