Databricks ML Professional Practice Question 8: MLflow Deployment Job Approval
Databricks Certified Machine Learning Professional · Test 1 · question 8 of 10
A team links a Databricks job to the Unity Catalog model prod.ml.fraud as its MLflow deployment job. The job has three tasks: evaluation, approval, and deployment. The job-level parameters model_name and model_version are set. When the retraining pipeline registers version 15, the deployment job starts by itself. The evaluation task succeeds, but the run then shows the approval task as failed.
What should the team conclude, and what should happen next?
Show answer and explanation
Answer:
MLflow deployment jobs start automatically when a new version of the linked Unity Catalog model is created. The human-in-the-loop step works through Unity Catalog tags. The docs say the first run of a deployment job always fails at the approval task, because the version has not been approved yet. After reviewing the evaluation metrics, an approver clicks Approve on the model version page. That adds a tag whose key is the approval task's name and whose value is Approved. It also repairs the run so it continues from the approval task into deployment.
Option A reverses the requirement. model_name and model_version must be job-level parameters, and the stem says they already are.
Option C uses the wrong mechanism. Aliases mark deployment status, such as champion. Approval for a deployment job is recorded as a Unity Catalog tag, not an alias.
Option D is wrong because the approval task is designed as a human gate. You can add automated checks in evaluation, but they do not replace the tag-based approval.
Reference: MLflow deployment jobs