Databricks Data Engineer Pro Sample Questions

About these questions

These five questions are original. We wrote them to match the topics in the Databricks exam guide dated October 9, 2026, covering five of its nine sections. They are not real exam questions, and the real exam will word things differently. All five are single-answer multiple choice.

How to use these

Try each question without looking at your notes, and pick an answer before you open the explanation. Then read the explanation even when you were right, because the reason the closest wrong option fails is where the learning is. If you slipped on a topic, find its section in domains explained.

Question 1 (Developing Code for Data Processing using Python and SQL)

A pipeline applies change events from a customer-loyalty system to a target table with AUTO CDC. Analysts must be able to report which loyalty tier any customer held on any past date. Which configuration meets the requirement?

  • A. SCD Type 1, because it keeps the latest tier and the pipeline can compare it with the source
  • B. SCD Type 1 with an extra timestamp column added to the source events
  • C. SCD Type 2, so each tier change creates a new row version with validity boundaries
  • D. SCD Type 2 on the source table, with SCD Type 1 on the target
Show the answer and explanation

Correct answer: C. SCD Type 2 keeps every version of a record and marks when each version was valid, so a report can ask for the tier as of a given date. In Lakeflow Pipelines you request this with `stored_as_scd_type` set to 2 on the AUTO CDC flow. SCD Type 1, the closest wrong option, overwrites the row with the newest values, so older tiers are lost even if the source carries a timestamp. Option D is not a real configuration, since the history is built in the target of the flow.

Question 2 (Data Ingestion & Acquisition)

A research group has no Databricks account. It needs read access to a governed table and plans to load it into a notebook with an open source client. Which approach fits?

  • A. Share the table using Databricks-to-open sharing, which gives the group credentials for an open client
  • B. Share the table using Databricks-to-Databricks sharing
  • C. Create a Lakehouse Federation connection to the group’s laptop
  • D. Export the table nightly to a storage bucket and give the group the bucket keys
Show the answer and explanation

Correct answer: A. Databricks-to-open sharing is meant for recipients outside Databricks. The provider issues access details that an open source Delta Sharing client can use, and the data stays governed from the provider’s side. Databricks-to-Databricks sharing, the closest wrong option, requires the recipient to use Databricks with Unity Catalog, which this group does not. Federation works in the other direction: it lets Databricks query an external database. The nightly export hands out storage keys and creates a copy that drifts from the source.

Question 3 (Data Manipulation)

A shipments pipeline must keep loading when some rows have no tracking number. Those rows must never be written to the target table, and the number of rows removed must appear in the pipeline’s quality metrics. Which approach fits?

  • A. An expectation with the default behavior, which records violations but keeps the rows
  • B. An expectation that fails the update when any row violates the rule
  • C. A filter in the query’s WHERE clause, because filtered rows are counted automatically as violations
  • D. An expectation that drops violating rows
Show the answer and explanation

Correct answer: D. An expectation with a drop action removes violating rows before they reach the target, lets the pipeline continue, and records how many rows were dropped in the quality metrics. The default expectation behavior is the closest wrong option: it counts violations but still writes the rows, so bad data lands in the target. Failing the update stops the load, which breaks the requirement to keep loading. A WHERE clause removes rows silently and leaves no violation count.

Question 4 (Cost & Performance Optimization)

Change Data Feed is already enabled on a Delta table of warehouse stock levels. A downstream team wants every row that changed since table version 12, including the value before and after each update. Which query gives them that?

  • A. `SELECT * FROM stock VERSION AS OF 12`
  • B. `SELECT * FROM table_changes(‘stock’, 12)`
  • C. `DESCRIBE HISTORY stock`
  • D. `SELECT * FROM stock TIMESTAMP AS OF ‘2026-10-01’`
Show the answer and explanation

Correct answer: B. `table_changes` reads the change feed from the starting version onward and returns the changed rows along with metadata columns, including a change type that separates the old and new values of an update. Time travel in options A and D returns a snapshot of the table, not a list of changes, so the team would have to compare snapshots themselves. Option C, the closest wrong option, lists commits and their metrics but returns no row content.

Question 5 (Debugging and Deploying)

A nightly job has seven tasks. Tasks 1 to 4 finished successfully, task 5 failed because of a short network outage, and tasks 6 and 7 depend on task 5 so they were skipped. The inputs have not changed. What is the most efficient way to finish the run?

  • A. Start a new run of the whole job and discard the first run
  • B. Clone the job and delete the first four tasks from the clone
  • C. Repair the failed run so that task 5 and its dependent tasks run again
  • D. Rewrite task 5 so that it retries forever
Show the answer and explanation

Correct answer: C. A repair run reruns only the failed task and the tasks downstream of it, within the same run, and reuses the successful results of tasks 1 to 4. A repair can also take parameter overrides if a fix needs a different value. Starting a full new run repeats work that already succeeded, which makes it the closest wrong option. Cloning and editing a job changes your job definition for a one-off problem, and an endless retry hides real faults instead of fixing them.

More guides for DB-DE-PRO

Next step

If a topic felt shaky, the domains explained page covers each section with typical traps. For more questions in the same style, the original practice sets have 60 questions each, with an explanation for every option.

Try these practice exams