Databricks-ML-Assoc Practice Test 17 — 15 Questions

Question 1

A team has several model versions registered in Unity Catalog. They want to serve a specific version through a Databricks Model Serving endpoint and later promote a newer version without changing the endpoint URL used by applications. Which approach should they use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Create the endpoint pointing to a Unity Catalog alias such as 'champion', and update the alias to reference the new model version when promoting.. Databricks Model Serving endpoints can reference a Unity Catalog model alias instead of a fixed version. Applications call a stable endpoint URL, and the team promotes new versions by moving the alias, such as from 'champion' to a newer version. This preserves the URL, avoids endpoint recreation, and keeps promotion governed through alias updates rather than client-side changes.

Answer

Create the endpoint pointing to a fixed model version, then delete and recreate the endpoint with the new version when promoting.

Answer

Create the endpoint pointing to the model name without a version, and rely on the platform to always serve the most recently created version.

Answer

Create one endpoint per model version and place a load balancer in front that routes to the newest endpoint.

Question 2

A team deploys an MLflow pyfunc model to a Databricks Model Serving endpoint. During pre-deployment testing they call the endpoint with a small batch of records and receive an HTTP 400 error stating the request payload does not match the model signature. The model was logged with an inferred signature from a pandas DataFrame. Which action most directly resolves the mismatch?

Accepted Answer

Re-log the model with an explicit input_example and signature, then register the new version and update the served entity to that version.. A signature mismatch means the logged model's expected schema differs from the payload the endpoint receives. The reliable fix is to re-log the model with an explicit signature and input example, register the new version, and repoint the served entity. Compute scaling, content-type changes, and inference tables address different concerns and do not alter schema validation.

Answer

Enable inference tables on the endpoint so failed requests are logged and the signature is auto-corrected.

Answer

Increase the endpoint's workload size from Small to Large so the request payload can be accepted.

Answer

Add a `Content-Type: text/plain` header to the request so the endpoint parses the payload differently.

Question 3

A machine learning engineer is using MLflow to log a model trained with a custom Python function. They want to ensure that the model can be loaded and served in a different environment. Which two of the following must be included when logging the model to ensure portability? (Choose two.)

Accepted Answer

The Python function code that defines the model's prediction logic.. The correct answers are the Python function code and the environment dependencies. When logging a custom Python function model, MLflow must save the function code to reconstruct the model. Additionally, the environment dependencies (conda.yaml) ensure that the required libraries are available. Together, these enable the model to be loaded and served in a different environment. The model signature, training dataset, and run ID are not strictly required for portability.

Answer

The training dataset used to fit the model.

Answer

The MLflow run ID for reference.

Answer

The model signature, specifying input and output schema.

Question 4

A machine learning engineer is configuring a Databricks Model Serving endpoint for a model that requires GPU acceleration. They set the workload size to 'GPU_Medium' but the endpoint fails to deploy. Which of the following is the most likely cause of the failure?

Accepted Answer

The Databricks workspace does not have GPU instances enabled or available.. GPU workload sizes for Model Serving require the workspace to have GPU instances enabled and available. If the workspace lacks GPU capacity or the user lacks permissions, deployment fails. Other factors like flavor or signature are unrelated to GPU provisioning. Ensuring GPU availability is a prerequisite for GPU-accelerated endpoints.

Answer

The model was not logged with the 'mlflow.pyfunc' flavor.

Answer

The model's signature does not include GPU-specific metadata.

Answer

The endpoint was configured with 'min_instances' set to 0.

Question 5

A team has an existing Databricks Model Serving endpoint serving `prod.ml.fraud_model` version 3. They register version 4, which uses a new feature set, and want to shift only 10% of traffic to version 4 while keeping version 3 for the rest. Their endpoint currently has a single served entity for version 3. What is the most appropriate approach?

Accepted Answer

Add a second served entity for version 4 to the same endpoint and configure traffic splitting between the two served entities.. Traffic splitting in Databricks Model Serving is achieved by defining multiple served entities on a single endpoint and assigning each a traffic percentage. Adding a served entity for version 4 at 10% and leaving version 3 at 90% implements the desired canary rollout. Swapping the served entity version, using an external load balancer, or renaming the model do not provide the controlled split.

Answer

Register version 4 under a new model name and point the existing served entity at the new model name.

Answer

Update the existing served entity's entity_version to 4 and rely on the endpoint's automatic gradual rollout.

Answer

Create a second endpoint for version 4 and configure a load balancer outside Databricks to split traffic 10/90 between endpoints.

Question 6

A team owns a Databricks Model Serving endpoint that receives sporadic bursts of traffic. They want to reduce cold-start latency during bursts while keeping cost predictable, and they also need to capture the request payloads and predictions for later monitoring. Which two configuration choices should they make? (Choose two.)

Accepted Answer

Set min_instances to a value greater than zero so a warm replica is always available.. Cold-start latency during sporadic bursts is best mitigated by keeping at least one replica warm with min_instances greater than zero, avoiding scale-to-zero teardown. Capturing request payloads and predictions for monitoring is done with inference tables, which write each request and response to a Delta table. Together these meet both the latency and the observability goals without conflicting.

Answer

Set scale_to_zero_enabled to true on the served entity to lower idle cost.

Answer

Increase the endpoint's workload size from Small to Large.

Answer

Attach an MLflow experiment to the endpoint to record prediction requests.

Question 7

A data scientist has registered a scikit-learn model in Unity Catalog and now wants to serve it behind a Databricks Model Serving endpoint. The model's MLflow signature records a pandas DataFrame input with three named columns. The team wants the endpoint to reject malformed requests automatically rather than silently scoring them. Which action should the data scientist take?

Accepted Answer

Create the endpoint with the registered model version; Model Serving enforces the logged MLflow signature and schema validation by default.. When a model is logged with an MLflow signature, Databricks Model Serving uses that signature to validate incoming requests. A pandas DataFrame signature with named columns causes the endpoint to check column names and types, rejecting payloads that do not conform. No custom preprocessing or inference table configuration is required, because validation happens at the serving layer before the model is invoked.

Answer

Enable inference tables on the endpoint so malformed records are flagged in the Delta table and excluded from scoring.

Answer

Register the model with a new signature that uses a tensor spec instead of a DataFrame spec so the endpoint validates each element.

Answer

Add a Python preprocessing script that parses the raw JSON and raises an error for unexpected fields, then deploy that script as the endpoint's model.

Question 8

A team operates a Databricks Model Serving endpoint with min_instances set to 0 and max_instances set to 4. During a nightly batch job, the endpoint receives a burst of requests and some clients observe elevated latency. The team wants to keep costs low during idle periods while reducing cold-start latency during bursts. Which configuration change best achieves this?

Accepted Answer

Set min_instances to 1 and leave max_instances at 4 so at least one instance is always warm.. Model Serving endpoints scale between min_instances and max_instances. With min_instances at 0, the endpoint scales to zero when idle and must cold-start a replica when traffic resumes, causing the observed latency. Setting min_instances to 1 keeps a single warm replica available for immediate responses while still allowing scale-out up to max_instances during bursts, achieving both cost and latency goals.

Answer

Set max_instances to 8 so the endpoint can scale further during the burst.

Answer

Set min_instances to 4 so the endpoint always matches the maximum capacity.

Answer

Enable scale-to-zero by setting both min_instances and max_instances to 0 and rely on queueing.

Question 9

A machine learning engineer has registered a scikit-learn model in Unity Catalog as `prod.ml.risk_model` with version 3. They want to serve it through a Databricks Model Serving endpoint that always uses this exact model version, even after new versions are registered. Which endpoint configuration should they use?

Accepted Answer

Set the served entity to `prod.ml.risk_model` with version 3 and do not configure aliases or the `latest` version.. Pinning a Model Serving endpoint to a specific registered model version is done by supplying the model name and version number in the served entity configuration. This produces an immutable reference that does not change when new versions are registered. Relying on aliases or `latest` introduces mutability, which conflicts with the requirement to always serve version 3.

Answer

Set the served entity to `prod.ml.risk_model` and set the version to `latest` so that the endpoint always uses the most recent registered version.

Answer

Set the served entity to `prod.ml.risk_model` with version 3 and enable scale-to-zero.

Answer

Set the served entity to `prod.ml.risk_model` and rely on the `@champion` alias to always resolve to version 3.

Question 10

A machine learning engineer is preparing to deploy a scikit-learn model as a Databricks Model Serving endpoint. The model expects a pandas DataFrame with specific column names and types. Which two actions should the engineer take to ensure the endpoint correctly validates and processes inference requests? (Choose two.)

Accepted Answer

Provide a representative `input_example` when logging the model so MLflow can infer and store the input schema.. Logging the model with an explicit `ModelSignature` and a representative `input_example` ensures the Serving endpoint knows the exact column names and types to expect. The signature drives request validation, while the input example documents and helps infer the schema. Together they prevent malformed requests from reaching the model and keep training and serving schemas aligned.

Answer

Convert the model to the `mlflow.pyfunc` flavor using a custom wrapper that hardcodes the expected column order.

Answer

Set the endpoint's `workload_size` to Large so the endpoint can coerce incoming data types automatically.

Answer

Enable inference tables on the endpoint so that incoming requests are automatically reformatted to match the model's schema.

Question 11

A team registered a model in Unity Catalog and wants to serve it on a Databricks Model Serving endpoint. During deployment, the build fails because the model's conda environment references a private Python package hosted on an internal PyPI mirror that the serving build cannot reach. Which approach resolves the deployment?

Accepted Answer

Convert the private package into a wheel, log it as a model artifact, and reference it in the model's conda environment via a pip requirement pointing to the artifact path.. The robust fix is to make the model environment self-contained by logging the private wheel as part of the model artifacts and pointing the conda pip section at that relative path. The serving build then installs the dependency from artifacts it already has, with no need to reach an internal index. This preserves reproducibility and avoids exposing private infrastructure to the serving build.

Answer

Add the internal PyPI mirror URL as an environment variable on the serving endpoint configuration.

Answer

Increase the endpoint's workload size so the build has more memory for dependency resolution.

Answer

Remove the private package from the conda environment and rely on the package being preinstalled in the serving base image.

Question 12

A data scientist registers an MLflow model whose `conda.yaml` lists several Python packages. When they create a Databricks Model Serving endpoint from this model, the deployment fails during environment build. Which action is most likely to resolve the failure while preserving the model's dependency requirements?

Accepted Answer

Add the missing or incompatible packages to the endpoint's environment by specifying them in the model's `requirements.txt` or adjusting the `conda.yaml` to versions available in the Databricks runtime.. Model Serving reconstructs the model's environment from the dependency files captured at logging time. When a required package is absent or a pinned version cannot be resolved, the build fails. Adjusting the `requirements.txt` or `conda.yaml` to include the necessary packages at versions compatible with the Databricks runtime lets the environment build succeed while keeping the dependencies the model genuinely needs.

Answer

Remove all pinned package versions from the `conda.yaml` so the environment builder can pick any compatible release.

Answer

Increase the endpoint's workload size to Large so the environment builder has more memory to resolve the dependency graph.

Answer

Convert the model to a different flavor, such as replacing scikit-learn with a PyTorch implementation, so the dependencies are no longer needed.

Question 13

A machine learning engineer is deploying an MLflow model to a Databricks Model Serving endpoint. The model was trained on a Spark DataFrame and logged with MLflow using the default signature detection. During testing, the endpoint returns predictions, but the engineer notices the input schema shown in the Serving UI does not match the actual DataFrame column types used during training. Which MLflow logging step should the engineer verify first to resolve this schema mismatch?

Accepted Answer

The model was logged with `mlflow.pyfunc.log_model` without explicitly passing the `signature` argument, so MLflow inferred it from a sample that may not represent the full schema.. The logged MLflow model signature is the authoritative contract for the Serving endpoint's expected input columns and types. When a signature is missing or inferred from an unrepresentative sample, the endpoint may expose an incorrect schema. Explicitly passing a `ModelSignature` during `log_model` ensures column names and data types align with training data, resolving the mismatch before redeployment.

Answer

The model's conda environment file lists incompatible library versions, so the Serving endpoint silently coerces input types to match the environment.

Answer

The Serving endpoint was configured with `min_instances` greater than zero, which forces the endpoint to rewrite the schema for autoscaling compatibility.

Answer

The model was registered in the Model Registry under a different name than the one used in the Serving endpoint configuration, causing the endpoint to load a stale schema.

Question 14

A team is preparing to deploy a model to a Databricks Model Serving endpoint that must scale down to zero replicas when idle yet still serve bursty traffic with acceptable cold-start latency. They also need to capture the request and response payloads for later monitoring. Which two endpoint settings or features should they configure? (Choose two.)

Accepted Answer

Enable scale-to-zero on the served entity so the endpoint can drop to zero replicas during idle periods.. The scenario has two distinct requirements: scaling to zero when idle and capturing payloads for monitoring. Scale-to-zero satisfies the first by releasing all compute during idle periods, and inference tables satisfy the second by logging request and response payloads to a Delta table. Keeping a minimum replica or reserving provisioned throughput would prevent scaling to zero, and workload size does not address either requirement.

Answer

Configure a provisioned throughput endpoint to reserve dedicated capacity for the model.

Answer

Increase the workload size to Large to reduce cold-start latency when scaling up from zero.

Answer

Set the endpoint's minimum replicas to one so that a warm replica is always available.

Question 15

A team's Model Serving endpoint occasionally returns errors when the upstream feature store is slow. They want the endpoint to retry transient failures and reduce cold-start latency for bursty traffic. Which combination of endpoint settings best addresses both concerns?

Accepted Answer

Configure a non-zero min_instances for warm capacity and implement client-side retry with backoff for transient upstream errors.. Warm capacity via a non-zero minimum instance count prevents the provisioning delay that causes cold-start latency during bursts. Because the endpoint itself does not automatically retry upstream feature store calls, retry with backoff belongs in the client or model logic. Combining warm replicas with retry logic addresses latency and transient failures without sacrificing scalability or observability.

Answer

Enable inference tables and route all traffic through a single replica to simplify retry logic.

Answer

Increase the endpoint's workload size to Large and disable request timeouts on the client.

Answer

Set min_instances to zero and enable scale-to-zero to lower cost during idle periods.