Databricks Data Engineer Associate Databricks Intelligence Platform: 49 practice questions

12 of the 49 Databricks Intelligence Platform questions in the Certsqill Databricks Data Engineer Associate bank, shown in full below. Each one carries an explanation for every option, not just the correct one — the wrong answers are where the marks go.

Preparing for Databricks Data Engineer Associate? Take the free 5-min readiness check →

1. Run dashboards on a SQL warehouse: Select TWO.

Select two. More than one option is correct — every correct one is ticked below.

1. Run the refresh on a SQL warehouse.

   A SQL warehouse supports SQL analytics but does not satisfy the required dedicated ephemeral Jobs compute requirement for the refresh.
2. Run dashboards on a SQL warehouse. ✓

   A SQL warehouse provides the separate interface for interactive SQL dashboard queries.
3. Run dashboards on all-purpose compute.

   All-purpose compute is interactive compute, not the separate SQL warehouse interface required for ad hoc dashboard queries.
4. Run the refresh on classic jobs compute.

   Classic jobs compute provides an ephemeral job lifecycle but does not satisfy the required serverless Jobs compute choice.
5. Run the refresh on serverless Jobs compute. ✓

   Serverless Jobs compute satisfies the required dedicated ephemeral execution and the stated serverless constraint.

Use serverless Jobs compute for the refresh and a SQL warehouse for dashboards.

2. ACID transactions and versioned table state: Which Delta Lake capability directly addresses this failure?

1. ACID transactions and versioned table state. ✓

   Delta Lake commits table changes transactionally and maintains versioned table state, preventing readers from observing partial committed updates.
2. Job retries with checkpoint recovery after failed tasks.

   Retries and checkpoint recovery belong to pipeline execution and do not make concurrent table writes atomic for readers.
3. Object-store versioning for deleted files and folders.

   Object-store versioning may preserve storage-object versions, but it does not provide Delta table transactions or a consistent table commit protocol.
4. Partition replacement with file renames and object-store consistency.

   File replacement and object-store behavior alone do not provide Delta's transactional commit log and consistent table versions.

Delta provides ACID transactions and versioned table state.

3. Use classic jobs compute with the custom library and Spark: Which compute choice satisfies all requirements?

1. Use serverless Jobs compute.

   Serverless Jobs compute provides the managed job lifecycle, but it cannot satisfy the required `spark.driver.extraJavaOptions` setting in this scenario.
2. Use all-purpose compute with the custom library and Spark setting for the scheduled run.

   All-purpose compute can provide customization, but it is shared interactive compute and fails the required dedicated ephemeral scheduled-job lifecycle.
3. Use a SQL warehouse for the ingestion workload.

   A SQL warehouse is designed for SQL analytics and is not the required general-purpose runtime for this customized JVM and Spark configuration.
4. Use classic jobs compute with the custom library and Spark setting. ✓

   Classic jobs compute provides the dedicated ephemeral job lifecycle and supports both the required JVM library and Spark setting.

Classic jobs compute supports the required library, setting, and ephemeral job lifecycle.

4. Associate the workspace with the Unity Catalog metastore: What prerequisite is missing?

1. Grant `USE CATALOG` on the governed catalog.

   `USE CATALOG` may be required for authorization, but it cannot replace the missing workspace-to-metastore association.
2. Grant `SELECT` on the table.

   A table grant cannot provide access through a workspace that is not associated with the relevant Unity Catalog metastore.
3. Associate the workspace with the Unity Catalog metastore. ✓

   The workspace must be associated with the metastore containing the governed catalog before its Unity Catalog objects can be accessed there.
4. Assign the analyst to a different Unity Catalog metastore.

   Metastores are associated with workspaces; assigning the analyst to another metastore does not make this catalog available in the workspace.

Associate the workspace with the relevant Unity Catalog metastore.

5. Use classic jobs compute with the dependency and Spark: Which implementation best satisfies the requirements?

1. Use classic jobs compute with the dependency and Spark setting. ✓

   Classic jobs compute supports the required dependency and Spark setting while providing the dedicated ephemeral scheduled-job lifecycle.
2. Use serverless Jobs compute and assume the unsupported Spark setting will be accepted.

   Serverless Jobs compute provides an ephemeral managed lifecycle, but it cannot satisfy the explicitly unsupported `spark.driver.extraJavaOptions` requirement.
3. Use a SQL warehouse.

   A SQL warehouse is intended for SQL analytics and cannot provide the required customized general-purpose transformation runtime.
4. Use all-purpose compute.

   All-purpose compute can support customization, but it is interactive shared compute and does not provide the required dedicated ephemeral job lifecycle.

Classic jobs compute meets the customization and ephemeral-lifecycle requirements.

6. Register the existing table as an external Unity Catalog: Which remediation is safest?

1. Create team-specific copies.

   Copies duplicate the data and do not provide one centrally governed shared table.
2. Grant both teams direct storage access.

   Direct storage access bypasses centralized table governance and complicates consistent review and revocation.
3. Create synchronized catalog copies for both teams.

   Synchronized copies add duplication and do not provide one governed shared data asset.
4. Register the existing table as an external Unity Catalog table and grant groups access. ✓

   Registering the existing storage-backed table in Unity Catalog provides governed discovery, centralized permissions, and lineage without copying data.

Register the existing table in Unity Catalog and grant group access.

7. Use classic jobs compute: Which implementation is most appropriate?

1. Attach the scheduled task to an always-running all-purpose cluster.

   All-purpose compute is interactive shared compute and violates the requirement for dedicated ephemeral job execution.
2. Use classic jobs compute. ✓

   Classic jobs compute supports an ephemeral job lifecycle while allowing the required pinned runtime and supported custom Spark configuration.
3. Use a SQL warehouse and execute the transformation manually.

   A SQL warehouse and manual execution do not provide the required configurable Spark job runtime or scheduled ephemeral job implementation.
4. Use serverless Jobs compute despite the unsupported key.

   This choice cannot satisfy the explicitly required Spark configuration because the specified key is unavailable on serverless Jobs compute.

Use classic jobs compute for the required custom Spark configuration.

8. VACUUM can remove files required by older versions: Select TWO accurate statements about time travel.

Select two. More than one option is correct — every correct one is ticked below.

1. Time travel succeeds when only the old transaction-log entry remains.

   The log identifies the historical state, but the referenced data files are also required to read that version.
2. VACUUM preserves every file needed by all historical versions.

   VACUUM can remove unreferenced old data files, which can make corresponding historical versions unavailable.
3. VACUUM can remove files required by older versions. ✓

   By deleting unreferenced old data files, VACUUM can prevent reads of historical versions that depend on those files.
4. Delta time travel guarantees every historical version indefinitely.

   Retention cleanup can remove transaction-log entries or data files, so historical recovery is not guaranteed indefinitely.
5. Time travel can read the older version when its log and data files remain. ✓

   A historical read requires the transaction-log information and the data files needed for that table version to remain available.

Time travel needs retained logs and data files; VACUUM can remove old files.

9. No; it does not guarantee either fix: Does increasing the autoscaling range alone guarantee resolution?

1. No; it does not guarantee either fix. ✓

   Adding workers does not necessarily eliminate a single skewed task or prevent driver memory exhaustion caused by collect.
2. Yes; worker autoscaling always increases driver memory.

   Autoscaling workers does not guarantee a larger driver, so it does not by itself resolve driver memory exhaustion during collect.
3. No; autoscaling is unavailable for product-metrics workloads.

   Autoscaling can support suitable parallel workloads; its limitation here is the skew and driver-side failure pattern, not universal unavailability.
4. Yes; more workers automatically repartition the skewed task.

   Autoscaling changes available capacity but does not automatically repartition or eliminate data skew in the failing task.

Autoscaling does not guarantee fixes for skew or driver-side collect OOM.

10. The compute plane: Which architectural component is responsible for executing the workload?

1. The control plane.

   The control plane manages workspace services but does not execute the customer-events transformations.
2. Unity Catalog.

   Unity Catalog provides governed metadata and access control, but it does not execute notebook transformations.
3. The compute plane. ✓

   The compute plane runs notebook commands and data-processing workloads.
4. The cloud object store.

   Cloud object storage holds data files but does not provide the compute resources that execute Spark commands.

The compute plane executes notebook and data-processing workloads.

11. Run A costs $3 per completed report: Which conclusion is correct?

1. Run A costs $3 per completed report, while Run B costs $2. ✓

   Cost per completed workload divides total run cost by successfully completed reports, making Run B more efficient.
2. Run A is cheaper because its total run cost is lower.

   Total run cost alone ignores how many completed reports each workload produced.
3. Run B is more expensive per report because its total cost is higher.

   Run B costs more overall but produces enough additional reports to lower its per-report cost.
4. Both runs cost $20 per completed workload because they use reporting compute.

   Compute type does not determine cost per workload; completed output count determines the denominator.

Run A costs $3 per report; Run B costs $2, so Run B has lower unit cost.

12. Adopt a documented medallion convention: What prerequisite should the design add?

1. Define independent retention schedules for bronze, silver, and gold inventory tables.

   Retention schedules address lifecycle management, not the missing architectural definitions for layer responsibilities and consumers.
2. Assign owners to each inventory layer.

   Ownership is useful, but it does not define the responsibilities, transformations, organization, and consumers required by the design.
3. Adopt a documented medallion convention for responsibilities and governed table organization. ✓

   A documented medallion convention defines the purpose, transformation boundary, table organization, and consumption expectations of each layer.
4. Place each layer in a separate Unity Catalog catalog.

   Separate catalogs are not required and would not by themselves define layer responsibilities or transformations.

Add a documented medallion convention defining layer responsibilities, organization, and consumption.

Other Databricks Data Engineer Associate domains

• Data Transformation and Modeling — 180 questions →
• Data Ingestion and Loading — 172 questions →
• Working with Lakeflow Jobs — 129 questions →
• Governance and Security — 123 questions →
• Implementing CI/CD — 82 questions →
• All 817 Databricks Data Engineer Associate questions →
• Databricks Data Engineer Associate certification: requirements, cost and exam format →