Databricks Data Analyst practice test
Twenty questions across the eight sections of the Databricks Data Analyst Associate exam, weighted roughly as the real exam is. Querying with Databricks SQL carries the most weight, so it gets the most questions here.
Executing queries using Databricks SQL and SQL Warehouses
Question 1. A dashboard query on a SQL warehouse is slow because it scans a very large table on every run, though analysts only ever look at the last 30 days. What is the most effective change?
- A. Increase the warehouse size
- B. Filter on the date column so file skipping eliminates older data
- C. Convert the query to a view without changing the filter
- D. Re-run the query more often so it stays warm
Show answer
Answer: B
Filtering on a partitioned or clustered date column lets the engine skip files outside the range. Increasing warehouse size spends money to scan data that need not be read at all.
Question 2. Several analysts run short interactive queries throughout the day, with long idle gaps. Which warehouse configuration controls cost best?
- A. The largest available size with auto-stop disabled
- B. A small size that never stops
- C. A size matched to the workload with a short auto-stop interval
- D. A new warehouse per analyst
Show answer
Answer: C
A modest size with a short auto-stop shuts the warehouse down during idle gaps while remaining responsive when used. Disabling auto-stop bills continuously.
Question 3. What is the practical difference between a serverless and a classic SQL warehouse for an analyst?
- A. Serverless starts much faster, suiting intermittent interactive use
- B. Serverless supports different SQL syntax
- C. Classic warehouses cannot run joins
- D. Serverless cannot query Delta tables
Show answer
Answer: A
Serverless starts in seconds because compute is pre-provisioned by the platform, which suits intermittent interactive work. Query syntax and results are unchanged.
Question 4. A query returns the right rows but takes far longer than a colleague's equivalent query. Where should you look first?
- A. The notebook’s Python version
- B. The workspace’s region
- C. The number of dashboard viewers
- D. The query profile, to see where time and data volume are spent
Show answer
Answer: D
The query profile shows where time is spent, such as scanning, shuffling or spilling, and directs the fix. Guessing at syntax changes wastes effort.
Creating Dashboards and Visualizations
Question 5. A dashboard must show the same figures to executives every Monday morning without anyone running it manually. What should you configure?
- A. A reminder in the team calendar
- B. A scheduled refresh with subscription delivery
- C. A note in the dashboard title
- D. A longer auto-stop on the warehouse
Show answer
Answer: B
A schedule refreshes the dashboard and can notify recipients automatically. Asking a person to run it is not a control.
Question 6. Viewers must be able to switch the dashboard between regions without editing the query. What should you add?
- A. A dashboard parameter bound to the query filter
- B. One dashboard per region
- C. A text box explaining how to edit the SQL
- D. A larger warehouse
Show answer
Answer: A
A parameter bound to the query lets viewers change the filter through the interface. Duplicating the dashboard per region multiplies maintenance.
Question 7. Which visualisation best shows the contribution of each product category to total revenue over twelve months?
- A. A gauge showing total revenue
- B. A scatter plot of revenue against product id
- C. A stacked area or stacked bar chart over time
- D. A table sorted by revenue
Show answer
Answer: C
A stacked area or stacked bar over time shows both the total and each category’s contribution. A scatter plot and a single gauge show neither.
Analyzing Queries
Question 8. A query aggregates sales by customer and then joins to a customer dimension table of 50 rows. The plan shows a large shuffle. What usually helps?
- A. Increase the number of shuffle partitions
- B. Broadcast the small dimension table
- C. Sort both tables before joining
- D. Convert the join to a cross join and filter
Show answer
Answer: B
A broadcast join sends the tiny dimension to every executor and removes the shuffle. Increasing shuffle partitions manages but does not eliminate it.
Question 9. Results differ between two analysts running what appears to be the same query. What is the most likely cause to check first?
- A. One used a larger warehouse
- B. One ran the query later in the day
- C. One is in a different cloud region
- D. They are querying different catalogs, schemas or table versions
Show answer
Answer: D
Different catalogs, schemas or table versions produce different results from identical SQL. Warehouse size and region do not change results.
Question 10. A query on a Delta table must reproduce the result as it stood last Tuesday. Which capability supports this?
- A. Delta time travel by timestamp or version
- B. Restoring the workspace from backup
- C. Re-importing the source files
- D. Increasing the retention of query history
Show answer
Answer: A
Time travel queries a table as of a version or timestamp. Recreating from backups is slower and may not match exactly.
Developing, Sharing, and Maintaining AI/BI Genie spaces
Question 11. Business users ask questions in natural language and get inconsistent answers because column names are cryptic. What improves this most?
- A. Renaming the Genie space
- B. Granting more users access
- C. Adding descriptive comments, instructions and example queries to the space
- D. Using a larger warehouse
Show answer
Answer: C
Clear table and column comments, plus example questions and defined metrics, give the system the semantics it needs. Renaming the space does nothing.
Question 12. Who should curate the tables exposed in a Genie space?
- A. Every user, self-service
- B. A curator who understands the business definitions and the data model
- C. Nobody; expose all tables in the catalog
- D. The workspace administrator only, without business input
Show answer
Answer: B
Someone who understands both the business meaning and the data model must curate, so answers reflect correct definitions. Exposing everything invites wrong joins and wrong metrics.
Understanding of Databricks Data + AI Platform
Question 13. What does the medallion architecture describe?
- A. Progressive refinement of data through raw, cleansed and business-level layers
- B. A licensing tier structure
- C. A cluster sizing standard
- D. A dashboard colour scheme
Show answer
Answer: A
Bronze, silver and gold layers describe progressive refinement from raw ingestion through cleansing to business-level aggregates. It is not a security or pricing model.
Question 14. Which statement about Delta Lake is correct?
- A. It is a proprietary database that replaces object storage
- B. It cannot handle concurrent writes
- C. It stores data only in memory
- D. It adds ACID transactions and schema enforcement over files in object storage
Show answer
Answer: D
Delta provides ACID transactions and schema enforcement over object storage, which plain file formats do not. It is not a proprietary database engine.
Managing Data
Question 15. An analyst needs a reusable, always-current logical result set without storing a second copy of the data. What should they create?
- A. A copy of the table
- B. A view
- C. A downloaded CSV
- D. A dashboard
Show answer
Answer: B
A view stores the query definition and evaluates it on read, so it is always current with no duplicate storage. A table copy duplicates and goes stale.
Question 16. A table accumulates many small files from frequent writes and queries slow down. What addresses this?
- A. Compacting small files into larger ones
- B. Adding more columns to the table
- C. Granting access to more users
- D. Renaming the table
Show answer
Answer: A
Compacting small files into larger ones reduces file listing and read overhead. Adding columns or viewers has no effect on file layout.
Securing Data
Question 17. Analysts in one region must see only rows for their own region, using a single shared table. What should be implemented?
- A. A separate table per region
- B. A dashboard parameter defaulting to their region
- C. A row filter on the table keyed to the user’s group
- D. An instruction in the table description
Show answer
Answer: C
A row filter applied to the table restricts rows per user or group centrally. Separate tables per region multiply maintenance and drift.
Question 18. A column contains national identification numbers that most analysts must not see, though they need the rest of the row. What applies?
- A. Drop the column
- B. A column mask that redacts the value for unprivileged users
- C. Revoke access to the whole table
- D. Rename the column
Show answer
Answer: B
A column mask redacts the sensitive column for unprivileged users while leaving the row accessible. Dropping the column removes it for everyone.
Importing Data
Question 19. A recurring CSV drop must be loaded incrementally, processing only new files each run. Which approach fits?
- A. Reload the entire folder every run
- B. Manually move processed files by hand
- C. Load once and assume no new files arrive
- D. Incremental file ingestion that tracks processed files
Show answer
Answer: D
Incremental ingestion that tracks which files have been processed avoids reprocessing and duplication. Reloading everything grows more expensive each run.
Question 20. An uploaded CSV has a column of numbers that lands as strings, breaking aggregations. What is the correct fix?
- A. Define an explicit schema or cast the column on read
- B. Correct each value in the dashboard
- C. Sort the data before importing
- D. Increase the warehouse size
Show answer
Answer: A
Defining an explicit schema or casting on read produces the right types. Fixing values one by one in the visualisation does not change the data.
How did you do?
Sixteen or more correct suggests you are close. Below fourteen, work through the section guides here first. Note that Databricks does not publish a passing score for this exam, so treat any specific number you see quoted elsewhere with suspicion.