DP-750: Implementing Data Engineering Solutions Using Azure Databricks Practice Exam

What the DP-750 exam covers

• Set up and configure an Azure Databricks environment114 questions
• Secure and govern Unity Catalog objects114 questions
• Prepare and process data211 questions
• Deploy and maintain data pipelines and workloads211 questions

Free DP-750 practice test questions

A sample of 10 questions with answers and explanations. Sign up free to practice all 650.

1. Question 1Set up and configure an Azure Databricks environment

   A scheduled nightly transformation runs unattended and must not be affected by an analyst's interactive session. Which compute type suits it?

   • AJob compute, created for the run and terminated when it finishesCorrect
   • BAn all-purpose cluster shared with the analytics team
   • CA SQL warehouse, which is provisioned for scheduled workloads
   • DA single-node cluster left running so the job starts immediately
   ✓ Correct answer: A

   A job cluster is created when the run starts and torn down when it ends, so nothing else is running on it and no interactive user can change its libraries or exhaust its memory. It is also billed at the lower jobs rate and for only the minutes the run takes. An all-purpose cluster is the shared, interactive resource and is priced accordingly, which makes it the wrong home for unattended work.

   Why the other options are wrong

   • BA shared all-purpose cluster is exactly the interactive resource the requirement is isolating the job from.
   • CA SQL warehouse serves SQL queries and dashboards rather than running a notebook or pipeline task.
   • DLeaving a cluster running pays for idle hours and still shares it with anything else pointed at it.
2. Question 2Deploy and maintain data pipelines and workloadsSelect all that apply

   A declarative pipeline is being promoted from development to production. Which TWO differences apply in production mode? (Choose TWO)

   • ACompute is terminated after the update completes rather than kept warmCorrect
   • BFailed updates are retried according to the pipeline's policy rather than stopping immediatelyCorrect
   • CExpectations switch from dropping rows to failing the update
   • DTables are written to a different catalog automatically
   • EThe pipeline's source notebooks are compiled and cached before the run
   ✓ Correct answer: A, B

   Development mode reuses the cluster between updates so iteration is quick, and it stops on the first error so the failure can be inspected immediately. Production mode terminates compute when the update finishes, which is what stops a pipeline billing between runs, and applies retries so a transient failure does not require intervention. Everything else - the catalog, the expectations - is configuration rather than a mode difference.

   Why the other options are wrong

   • CAn expectation's action is declared in the code and does not change with the pipeline's mode.
   • DThe target catalog is configuration, typically overridden per bundle target rather than by the mode.
   • ESource notebooks are evaluated when the update runs; there is no separate compile step tied to production mode.
3. Question 3Prepare and process dataSelect all that apply

   A silver table applies type 2 history to a dimension. Which TWO columns does the pattern require? (Choose TWO)

   • AA validity range marking when each version appliedCorrect
   • BA flag or equivalent identifying the current versionCorrect
   • CA checksum of the row, used to detect changes
   • DA partition column derived from the business key
   • EAn ingestion timestamp, which serves as the surrogate key
   ✓ Correct answer: A, B

   The validity range - a valid-from and a valid-to - is what lets a fact join to the version that applied at its own date, and it is the whole point of the pattern. A current flag is the practical companion: most queries want today's row, and scanning for a null valid-to works but is slower and less obvious. A surrogate key per version is the third element, so facts reference a specific version rather than the business key.

   Why the other options are wrong

   • CA checksum is a convenience for detecting changes on load rather than part of the dimension's structure.
   • DA partition column is a physical layout choice rather than part of what makes a dimension type 2.
   • EAn ingestion timestamp records when a row was loaded and is not a surrogate key.
4. Question 4Set up and configure an Azure Databricks environment

   A build engineer must attach a notebook to compute that enforces Unity Catalog and supports Scala. What should they check?

   • AThat the access mode and runtime version support the language they needCorrect
   • BThat the cluster has the machine learning runtime installed
   • CThat the notebook's default language is set to Scala
   • DThat the catalog has been bound to their workspace
   ✓ Correct answer: A

   Which languages are available on Unity Catalog-enabled compute depends on the access mode and the runtime version, and the support matrix has changed across releases - so the practical check is the current documentation for that combination rather than an assumption carried over from an earlier version. Setting a notebook's language does nothing if the compute will not run it.

   Why the other options are wrong

   • BThe machine learning runtime adds libraries rather than determining language support under an access mode.
   • CThe notebook's language setting does not make an unsupported language available on the compute.
   • DCatalog binding governs which workspaces may reach a catalog rather than which languages run.
5. Question 5Prepare and process data

   A large join between two fact tables spills heavily and takes hours. Which change is most likely to help?

   • AFilter and aggregate each side before joiningCorrect
   • BBroadcast the larger side to avoid the shuffle
   • CIncrease the number of output files written
   • DConvert the join to a cross join with a filter
   ✓ Correct answer: A

   Applying the filters and reducing each side to the columns and grain the join actually needs shrinks what crosses the network, which is what dominates a fact-to-fact join - and it often turns a spilling stage into one that fits in memory. Broadcasting is unavailable when both sides are large, so the volume itself has to come down before tuning the shuffle is worth doing.

   Why the other options are wrong

   • BBroadcasting requires one side to be small enough to fit on every executor, which a fact table is not.
   • COutput file count affects the write rather than the shuffle that precedes it.
   • DA cross join produces every pair before filtering, which is dramatically worse.
6. Question 6Set up and configure an Azure Databricks environment

   A data engineer must create a schema whose tables are managed but whose files are held in a specific container. What must exist first?

   • AAn external location for that container, set as the managed locationCorrect
   • BA volume in that container, which the schema uses for its tables
   • CA foreign catalog registered against that container
   • DA mount point mapping the container into the workspace
   ✓ Correct answer: A

   Unity Catalog reaches storage through an external location backed by a storage credential, and a schema's managed location names one of those - so the container has to be registered before the schema can be created against it, and the creating principal needs the privilege to use it. Managed tables in that schema then write there rather than to the catalog's default.

   Why the other options are wrong

   • BA volume holds files under a location rather than serving as a schema's managed storage.
   • CA foreign catalog federates a database system rather than registering a storage container.
   • DMounts are the legacy mechanism and are not used for managed locations.
7. Question 7Prepare and process dataSelect all that apply

   A streaming pipeline must deduplicate a stream. Which TWO facts about that operation matter? (Choose TWO)

   • AIt holds state per key, which grows without a watermarkCorrect
   • BIt removes only the duplicates that fall within the retained stateCorrect
   • CIt guarantees global uniqueness across the stream's whole lifetime
   • DIt requires the target table to have a primary key constraint
   • EIt can only be applied to a batch read
   ✓ Correct answer: A, B

   Deduplication tracks the keys already emitted, so state grows with distinct keys until a watermark allows old entries to be dropped. The consequence is that a duplicate arriving after its key has been evicted is not caught - deduplication is bounded by the state window rather than absolute. Where absolute uniqueness matters, a merge into the target on the key is the durable answer.

   Why the other options are wrong

   • CUniqueness holds within the retained state rather than over the stream's whole lifetime.
   • DDelta does not enforce primary key constraints, so none is required or relied upon.
   • EDeduplication applies to streaming queries as well as batch ones.
8. Question 8Set up and configure an Azure Databricks environmentSelect all that apply

   A catalog is being created for a new domain. Which TWO decisions should be made at creation? (Choose TWO)

   • AIts managed storage location, which is fixed at creationCorrect
   • BWhich workspaces it is bound to, if the environment must be isolatedCorrect
   • CWhich tables it will eventually contain
   • DThe clustering keys for its future tables
   • EThe runtime version its consumers will use
   ✓ Correct answer: A, B

   The managed storage location determines where the catalog's managed tables physically live and is set at creation, so getting it right avoids a migration later. Binding decides which workspaces can address the catalog at all, which is how environments are separated and is far easier to establish up front than to impose once people are already using it from everywhere.

   Why the other options are wrong

   • CTables are created as they are needed rather than declared with the catalog.
   • DClustering is a per-table decision made when each table is created.
   • ERuntime versions are a compute concern unrelated to the catalog.
9. Question 9Deploy and maintain data pipelines and workloads

   A workspace admin must judge whether to store pipeline configuration in the bundle or in a table. What favours the bundle?

   • AIt changes with the code and should be reviewed and deployed alongside itCorrect
   • BIt is edited by business users without engineering involvement
   • CIt changes several times a day
   • DIt differs per row of the data being processed
   ✓ Correct answer: A

   Cluster sizes, schedules, target catalogs and library versions change when the pipeline changes and should be reviewed in the same pull request - so they belong in the bundle, where a deployment applies them atomically with the code. Business parameters that change independently and are maintained by non-engineers belong in a governed table instead.

   Why the other options are wrong

   • BValues edited by business users belong in a table rather than in a deployment definition.
   • CSomething changing several times a day should not require a deployment each time.
   • DPer-row values are data rather than configuration.
10. Question 10Secure and govern Unity Catalog objects

    A group holds SELECT on a schema through membership of an analytics group, and one table within it must be hidden from them. What is the correct approach?

    • AMove the table to a schema they have no privilege on, since there is no denyCorrect
    • BAdd a deny on that table for the group
    • CGrant a narrower privilege on the table that overrides the schema grant
    • DRemove the group's members individually from the table's access list
    ✓ Correct answer: A

    Unity Catalog accumulates privileges and offers no mechanism to subtract one, so a schema-level grant reaches every table in it including any added later. Meeting the requirement means relocating the sensitive table to a schema the group has nothing on, or granting at table level throughout instead of at schema level - both structural changes rather than an exception.

    Why the other options are wrong

    • BThere is no deny mechanism in the privilege model.
    • CGrants are additive; a narrower grant does not subtract from a broader one.
    • DThere is no per-table exclusion list to remove members from.

Who this DP-750 practice exam is for

This practice set is for anyone preparing for the DP-750: Implementing Data Engineering Solutions Using Azure Databricks exam at the intermediate level - from first-time candidates building a foundation to experienced Microsoft practitioners doing a final review before test day. If you learn best by working through realistic questions and reading why each answer is right or wrong, it is built for you.

How to use this DP-750 practice exam

1. Start with the free sample questions above to gauge your current baseline.
2. Read the full explanation on every question, including why each wrong option is wrong.
3. Track your weak domains and focus your study where you are losing the most marks.
4. Once you are scoring consistently well, take a timed, full-length mock exam.
5. Use your readiness score to decide when you are ready to book the real DP-750 exam.

Related Microsoft resources

• Microsoft practice examsAll Microsoft
• Certification pathWhere this fits
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• How these questions are written and reviewedMethodology
• Report a problem with a questionCorrections
• DP-800 practice examRelated
• DP-900 practice examRelated
• MB-310 practice examRelated

DP-750 practice exam FAQ