Databricks Certified Machine Learning Professional Practice Questions

Databricks Certified Machine Learning Professional exam preparation

Databricks Certified Machine Learning Professional practice questions in English: 180 questions with explanations. The mock exam has 59 questions and 120 minutes.

Databricks Certified Machine Learning Professional free question 9 / 10 · Domain 3: Model Deployment

During a promotion, request traffic to a custom model endpoint jumps sharply and clients start to receive HTTP 429 errors. How does the Databricks page on optimizing Model Serving endpoints explain this, and what does it advise to reduce queuing?

  1. The client token has expired, so the clients should switch to a personal access token instead.
  2. Bursts are queued, then rejected with HTTP 429; raise the minimum provisioned concurrency for bursts.
  3. The endpoint rejects bursts on arrival, so clients should keep retrying at full speed.
  4. The endpoint has lost its model, so scale to zero should be switched on to make it reload.
Show the answer and explanation

Answer: B

Why this is correct
Autoscaling needs time to detect increased load and provision capacity, so a sudden surge can exceed the available capacity and requests queue. Model Serving endpoints allow temporary queuing to handle bursts, but beyond a defined threshold the endpoint returns HTTP 429 errors to protect system stability. To minimize queuing, the page advises setting the minimum provisioned concurrency high enough for baseline traffic plus typical bursts, enabling route optimization, and using retry logic with exponential backoff in clients.

Why the other options are wrong
"The endpoint has lost its model, so scale to zero should be switched on to make it reload." … The page on creating endpoints says scale to zero is not recommended for production endpoints, because capacity is not guaranteed when scaled to zero.
"The client token has expired, so the clients should switch to a personal access token instead." … The page explains HTTP 429 as the endpoint protecting itself when queuing passes a threshold. That is a matter of capacity, not of the type of token.
"The endpoint rejects bursts on arrival, so clients should keep retrying at full speed." … The page says endpoints allow temporary queuing for bursts, and it advises retry logic with exponential backoff, not immediate retries.

429 means the queue is past its threshold
Size minimum concurrency for bursts, and let clients back off before retrying.

Go to the Databricks Certified Machine Learning Professional practice app (10 free questions; purchase unlocks all questions)

See the full practice set