Free DP-750 practice test: 20 questions
Twenty questions across the four DP-750 skill areas, weighted roughly as the real exam is: three on environment setup, three on security and governance, seven on preparing and processing data and seven on deploying and maintaining pipelines. These are original scenario questions. Find the constraint in each scenario before you look at the options.
Set up and configure an Azure Databricks environment
Question 1. Interactive development clusters are often left running overnight, and the bill is growing. Developers still want to attach notebooks to their own compute during the day. What is the simplest setting change?
- A. Disable autoscaling on the cluster
- B. Move to a larger node type
- C. Set an auto-termination period for idle time
- D. Remove the attach permission from developers
Show answer
Answer: C
An auto-termination period shuts the cluster down after a set idle time, removing the idle cost while keeping the cluster available on demand. Disabling autoscaling does not stop idle charges. A larger node type raises the hourly price. Removing permissions would stop developers from working.
Want more questions like this? Full DP-750 practice tests →
Question 2. A job starts a new cluster for every run and the 5-minute start time delays a pipeline that runs every 15 minutes. Which feature reduces the start time?
- A. Enabling Photon acceleration
- B. Using an instance pool
- C. Installing more libraries on the cluster
- D. Selecting a larger driver node
Show answer
Answer: B
A pool keeps ready idle instances that clusters can take, which cuts start time. Photon speeds up query execution, not cluster start. Adding libraries to the cluster can make start slower. A larger driver does not shorten provisioning.
Want more questions like this? Full DP-750 practice tests →
Question 3. A team wants production data isolated from development data, with different owners and the ability to grant access per environment. Which Unity Catalog layout fits best?
- A. One schema with environment prefixes on table names
- B. One table with an environment column
- C. A volume per environment
- D. A catalog per environment
Show answer
Answer: D
Separate catalogs per environment give a clean isolation boundary with their own owners and grants. A single schema with name prefixes provides no real separation. Tables with an environment column mix the data. Volumes hold files and are not an isolation mechanism by themselves.
Want more questions like this? Full DP-750 practice tests →
Secure and govern Unity Catalog objects
Question 4. Support staff may query the customers table but must see only the last four digits of the phone column. Compliance staff must see the full value. What should you implement?
- A. A column mask that checks group membership
- B. A row filter on the customers table
- C. Revoke SELECT from support staff
- D. A masked copy of the table refreshed weekly
Show answer
Answer: A
A column mask on the phone column can return the full value for the compliance group and a masked value for everyone else. A row filter hides rows, not parts of a column. Revoking SELECT from support removes all access. A second copy of the table doubles storage and needs syncing.
Want more questions like this? Full DP-750 practice tests →
Question 5. A nightly pipeline currently runs under an engineer's personal account. The engineer is leaving. What is the recommended replacement identity?
- A. A shared password stored in the team wiki
- B. Another engineer’s personal account
- C. A service principal with least-privilege grants
- D. A guest user account
Show answer
Answer: C
A service principal is a non-human identity intended for automation and can be granted only the privileges the pipeline needs. A shared team password is insecure and untraceable. A different engineer’s account repeats the problem. A guest user is still a person and may lose access.
Want more questions like this? Full DP-750 practice tests →
Question 6. Before changing a column in a core table, an engineer needs to know which downstream tables, notebooks and jobs use it. Where should they look?
- A. The cluster event log
- B. The lineage view in Catalog Explorer
- C. The Spark UI of the latest run
- D. The VACUUM history of the table
Show answer
Answer: B
Lineage in Catalog Explorer shows upstream and downstream tables, columns, notebooks and jobs for an object. The cluster event log lists compute events. The Spark UI describes one run, not dependencies across objects. A table’s VACUUM history shows maintenance operations, not consumers.
Want more questions like this? Full DP-750 practice tests →
Prepare and process data
Question 7. Event data arrives on Azure Event Hubs and must be processed continuously into a Delta table with fault tolerance after a restart. Which combination fits?
- A. A daily batch spark.read of the hub
- B. COPY INTO from the event hub
- C. A manual CSV export loaded weekly
- D. Structured Streaming with a checkpoint location
Show answer
Answer: D
Structured Streaming reading Event Hubs through its Kafka-compatible endpoint, with a checkpoint location, resumes from where it stopped after a restart. A one-off batch read misses new events. COPY INTO reads files, not an event stream. Exporting events to CSV by hand adds delay and manual work.
Want more questions like this? Full DP-750 practice tests →
Question 8. Two Delta tables, current_customers and churned_customers, share a customer_id column. You need ids that appear in current_customers but not in churned_customers. Which operator do you use?
- A. EXCEPT
- B. INTERSECT
- C. UNION
- D. INNER JOIN
Show answer
Answer: A
EXCEPT returns rows from the first query that do not appear in the second. INTERSECT returns rows found in both. UNION combines rows from both. A plain inner join returns matches, the opposite of what you need.
Want more questions like this? Full DP-750 practice tests →
Question 9. A pipeline must stop writing and report an error whenever a batch contains a negative order amount, instead of loading it. Which rule type fits a Lakeflow Spark Declarative Pipeline?
- A. An expectation with the drop-row action
- B. An expectation with no action
- C. An expectation with the fail-update action
- D. VACUUM before each update
Show answer
Answer: C
An expectation with the fail-update action stops the update when any row violates the rule. The drop action silently removes bad rows and continues. An expectation without an action only records metrics. VACUUM does not validate data.
Want more questions like this? Full DP-750 practice tests →
Question 10. A source starts sending an extra column. Your Delta table rejects the write. You want the new column added automatically in a controlled way, without breaking other loads. What is the concept involved?
- A. Z-ordering the table
- B. Enabling schema evolution on the write
- C. Time travel to an older version
- D. Running VACUUM
Show answer
Answer: B
Schema enforcement is what rejects the mismatched write, and schema evolution, enabled deliberately on the write or in Auto Loader, lets the table accept the new column. Z-ordering is a layout technique. Time travel reads old versions. Vacuuming removes files and does not alter schemas.
Want more questions like this? Full DP-750 practice tests →
Question 11. Large CSV exports are delivered once a month into a folder. A SQL-only team wants to load only files not loaded before, with no streaming code. Which statement suits them?
- A. CREATE OR REPLACE TABLE over the folder
- B. INSERT INTO selecting the whole folder
- C. MERGE INTO with the folder as source
- D. COPY INTO the target table
Show answer
Answer: D
COPY INTO loads new files from a location and skips files it has already loaded, and it is a SQL command. CREATE OR REPLACE TABLE rewrites the table every time. A plain INSERT of the whole folder would reload old files. A MERGE needs a source with keys and does not track files.
Want more questions like this? Full DP-750 practice tests →
Question 12. A finance table stores amounts as strings, and sums sometimes fail or round oddly. Which change is most appropriate?
- A. Store the amounts as DECIMAL with fixed precision and scale
- B. Store the amounts as DOUBLE
- C. Keep strings and cast in every query
- D. Store the amounts as BOOLEAN
Show answer
Answer: A
Money values should use a DECIMAL type with a defined precision and scale, which allows exact arithmetic. DOUBLE can introduce rounding error. Keeping strings and casting in every query hides the problem. A boolean cannot hold amounts.
Want more questions like this? Full DP-750 practice tests →
Question 13. You must ingest tables from a supported enterprise SaaS application on a schedule with the least custom code. Which Databricks option should you evaluate first?
- A. A custom Python API client in a notebook
- B. Lakeflow Connect
- C. Manual exports uploaded to a volume
- D. Auto Loader pointed at the application
Show answer
Answer: B
Lakeflow Connect provides managed connectors for supported sources, so you configure the ingestion instead of writing it. A custom Python client in a notebook works but costs maintenance. Manual exports are slow and error-prone. Auto Loader reads files from storage, not application APIs.
Want more questions like this? Full DP-750 practice tests →
Deploy and maintain data pipelines and workloads
Question 14. A job has task B that must run only after task A succeeds, and task C that must run after both. Where do you define this?
- A. Cluster tags
- B. Library settings
- C. Task dependencies in the job
- D. The Git branch name
Show answer
Answer: C
Task dependencies in the job definition set the order, so B depends on A and C depends on A and B. Cluster tags label compute and do not control order. Library settings install packages. A Git branch name has no effect on task order.
Want more questions like this? Full DP-750 practice tests →
Question 15. A job should start automatically whenever new files land in a storage location, without a fixed time schedule. Which trigger do you configure?
- A. A cron schedule every minute
- B. A manual run by an operator
- C. Auto-termination on the compute
- D. A file arrival trigger
Show answer
Answer: D
A file arrival trigger starts the job when new files appear in a monitored location. A cron schedule runs at fixed times whether data arrived or not. A manual run needs a person. Auto-termination ends compute and does not start jobs.
Want more questions like this? Full DP-750 practice tests →
Question 16. Intermittent network errors fail a task a few times a week, and a retry usually succeeds. What should you configure?
- A. Automatic retries on the task
- B. A larger cluster for the job
- C. Disable the failure alerts
- D. Remove the task dependency
Show answer
Answer: A
Automatic task retries rerun the task after a failure, which handles transient errors without human action. A larger cluster does not fix network errors. Disabling alerts hides problems. Removing the dependency changes the pipeline logic.
Want more questions like this? Full DP-750 practice tests →
Question 17. Two engineers edited the same notebook on different branches, and merging produces a conflict. What is the correct way to proceed?
- A. Force-push your branch over main
- B. Resolve the conflict in the pull request, test and merge
- C. Delete the other branch
- D. Copy the notebook under a new name
Show answer
Answer: B
Resolve the conflict in the merge or pull request, keep the intended changes from both, test, and complete the merge after review. Force-pushing overwrites the other engineer’s work. Deleting the branch loses changes. Copying the notebook to a new name bypasses version control.
Want more questions like this? Full DP-750 practice tests →
Question 18. A Spark stage shows one task running for 40 minutes while the others finish in seconds. The join key has a few extremely common values. What is the problem?
- A. Disk spill
- B. A missing library
- C. Data skew
- D. An expired credential
Show answer
Answer: C
One very slow task among many fast ones is the signature of data skew caused by uneven key distribution. Spill would show disk usage in every heavy task. A missing library gives errors, not a slow task. An expired credential would fail the job, not slow one task.
Want more questions like this? Full DP-750 practice tests →
Question 19. Queries on a table slow down after thousands of small appends. You want to compact the files without removing history. Which command should you run?
- A. VACUUM
- B. DROP TABLE
- C. DESCRIBE HISTORY
- D. OPTIMIZE
Show answer
Answer: D
OPTIMIZE rewrites many small files into fewer larger ones to speed up reads, leaving the table’s history intact. VACUUM removes unreferenced old files and does not compact. DROP TABLE deletes data. DESCRIBE HISTORY only shows operations.
Want more questions like this? Full DP-750 practice tests →
Question 20. Operations wants an alert in Azure when a Databricks workload logs repeated errors, using a central log store. What should you set up?
- A. Log streaming to Log Analytics with an Azure Monitor alert rule
- B. Emailing log files to operations each morning
- C. A longer Delta table retention period
- D. Enabling Photon on the compute
Show answer
Answer: A
Streaming Databricks logs to a Log Analytics workspace allows KQL queries and Azure Monitor alert rules on error counts. Emailing logs by hand is not an alert. Increasing the retention of Delta tables concerns data, not logs. Photon speeds queries and has no monitoring role.
Want more questions like this? Full DP-750 practice tests →
How did you do?
Sixteen or more correct suggests you are close. Below fourteen, use the domain guides to find the gaps. Questions 7 to 13 are the largest block, so weak results there point to the data processing guide. Questions 14 to 20 cover pipelines and operations; if you missed more than two, read the pipelines guide. Questions 4 to 6 map to Unity Catalog security and 1 to 3 to the environment guide.
Repeat Microsoft’s practice assessment afterwards. If both results point at the same domain, you know where your remaining time should go.