Skip to content
DatabricksPractice bank availableProfessional

Databricks Certified Data Engineer Professional: exam facts and practice questions

Verified exam facts from Databricks’s published guide, checked on 20 November 2026, and an original practice bank with 249 independently validated questions.

Verified exam facts

What Databricks publishes about this exam

Recorded on 20 November 2026. Confidence in the record: high. Check Databricks's own pages before you book.

Exam code
No exam code published
Exam version
Exam Published Oct 9, 2026 / Exam Guide Updated Oct 9, 2026 (new exam live in Webassessor from Oct 9, 2026)Effective 9 October 2026
Scored questions
60
Time limit
120 minutes
Unscored questions
The 9 Oct 2026 guide (p.2) says unscored questions may be included to gather statistical information, are not identified on the exam and do not affect the score, and that additional time is factored into the 120-minute limit. It states no number; the Oct 2026 transition guide (About the Exam) says up to 10.
Question format
multiple-choice
Delivery
Online proctored or test center
Languages
English
Test aids
None allowed
Exam price
US$200
Certification validity
2 years
Prerequisites
None required; related training highly recommended.
Recommended experience
One year of hands-on experience in the data engineering tasks in the exam outline is highly recommended (guide). The certification page FAQ also cites a six-month Databricks baseline.
Passing score
Databricks does not publish a passing score for this exam.CertExam does not state one. Guide does not state a passing score; the certification page says the passing score is not published, is set through statistical analysis and can change as questions are added.
Facts checked on
20 November 2026
What we could not verify (6)
  • Languages: the 9 Oct 2026 guide lists English only; the certification page still lists English, Japanese, Portuguese BR and Korean. Guide used; whether other languages are offered is not confirmed.
  • Question type: the guide says 'multiple-choice' only; whether multi-select items exist is not stated.
  • Whether the Section 8 bullet 'Use event logs and the Spark UI to debug pipelines and Spark workloads' (in the transition guide, absent from the 9 Oct guide) was removed deliberately or folded into d8.o1 is not stated by Databricks.
  • Number of unscored questions: the 9 Oct guide gives no number; 'up to 10' comes from the transition guide.
  • Passing score: not stated in the guide; the certification page says passing scores are not published. No value is recorded.
  • Exam code: none published.

Exam outline

Domains and weights

The published outline has 9 domains; the weights add up to 100 percent. Open a domain to read its objectives.

  1. 1. Developing Code for Data Processing using Python and SQL23 percent

    Objectives

    • Implement scalable Python project structures for Declarative Automation Bundles (formerly Databricks Asset Bundles / DABs) to support modular development and CI/CD integration.
    • Apply troubleshooting techniques to dependency conflicts and installation failures for external libraries (PyPI, local wheels, source archives) across serverless, pipeline, and bundle-deployed environments.
    • Develop User-Defined Functions (UDFs) using Pandas, Python, and SQL, including Unity Catalog functions for the given constraints.
    • Create production-ready data pipelines for streaming data using Lakeflow Pipelines and Auto Loader.
    • Create and automate ETL workloads using Lakeflow Jobs via the UI, API, and CLI.
    • Choose between a streaming table and a materialized view for a given latency, cost, and refresh requirement.
    • Implement CDC pipelines using AUTO CDC APIs (formerly APPLY CHANGES) in Lakeflow Pipelines, including SCD Type 1 and Type 2 via stored_as_scd_type.
    • Choose between Spark Structured Streaming and Apache Spark™ Declarative Pipelines for scalable ETL given operational constraints.
    • Create Lakeflow Jobs that use control-flow operators such as If/Else conditions and For Each loops.
    • Choose appropriate compute and configuration for environments and dependencies — including serverless compute (serverless environments, dependency management, performance mode), high-memory notebook tasks, and auto-optimization settings (e.g., disallowing retries).
    • Develop unit and integration tests using assertDataFrameEqual, assertSchemaEqual, DataFrame.transform, and testing frameworks to ensure code correctness.
    • Apply Structured Streaming stateful-processing semantics, including watermarks, output modes, foreachBatch, and checkpoints for fault-tolerant exactly-once state recovery.
  2. 2. Data Ingestion & Acquisition12 percent

    Objectives

    • Develop data ingestion pipelines to ingest a variety of data formats, including Delta Lake, Parquet, JSON, Iceberg, CSV, and Binary from diverse sources such as message buses (Kafka, Kinesis, Pub/Sub) and cloud storage.
    • Configure incremental CDC pipelines using Lakeflow Pipelines with Delta or Iceberg as the target table format.
    • Configure CDC ingestion pipelines from relational database sources, including SQL Server, MySQL, and PostgreSQL, using Lakeflow Connect.
    • Implement OpenSharing (D2D and Databricks-to-Open) and Clean Rooms for privacy-preserving collaboration.
    • Configure Lakehouse Federation with governance across supported source systems, applying UC permissions and connection-level credentials.
  3. 3. Data Manipulation12 percent

    Objectives

    • Apply advanced data transformations, including window functions, joins, and aggregations, using Spark SQL and PySpark to process large datasets.
    • Develop a model and query semi-structured data using the VARIANT data type and related functions (e.g., parse_json, variant_get).
    • Apply AI functions, including ai_query, to perform model inference within data pipelines for enrichment and classification tasks.
    • Implement data quality expectations in Lakeflow Pipelines to quarantine, drop, or fail on bad records.
  4. 4. Monitoring and Alerting10 percent

    Objectives

    • Use Databricks system tables (billing, compute, access, lakeflow) for cost analysis, auditing, and workload monitoring.
    • Use Databricks REST APIs / CLI and SDK for monitoring and analyzing jobs and pipelines.
    • Use Pipeline event logs to monitor pipeline health and data quality.
    • Use Databricks Lakehouse alerts to monitor governed business metrics, data quality and anomaly signals, usage and cost, SQL warehouse/query health, audit and security events, AI agent quality, and Lakeflow Job branching.
    • Use Lakeflow Jobs UI and Jobs API to monitor job status and performance metrics.
  5. 5. Cost & Performance Optimization15 percent

    Objectives

    • Explain how Unity Catalog managed tables, Predictive Optimization, and Liquid Clustering reduce operational and maintenance overhead for a given workload.
    • Choose the appropriate Delta optimization technique (deletion vectors, Liquid Clustering, CLUSTER BY AUTO) for a given table access pattern.
    • Understand how caching (Delta cache) improves query performance for repeated reads of the same data.
    • Apply Change Data Feed (CDF) to expose row-level changes (updates/deletes) for efficient incremental downstream processing.
    • Use the query profile to identify performance bottlenecks, such as data-skipping inefficiencies, join strategies, and shuffle operations.
    • Compare Liquid Clustering vs partitioning/ZORDER for a given table size and query pattern.
  6. 6. Ensuring Data Security and Compliance8 percent

    Objectives

    • Apply least-privilege access control lists (ACLs) to secure Unity Catalog securable objects and workspace resources.
    • Apply attribute-based access control (ABAC) policies with governed tags to enforce row filters and column masks at scale.
    • Apply anonymization and pseudonymization methods — such as hashing, tokenization, suppression, and generalization — to confidential data (e.g., using column masks and related Unity Catalog features).
    • Implement a compliant data pipeline that enforces PII detection and masking controls across batch and streaming workloads using Unity Catalog features.
    • Implement a data purging strategy that satisfies data retention and deletion policies (e.g., GDPR right-to-erasure, GDPR right-to-be-forgotten deletion) using Delta Lake and Unity Catalog features.
  7. 7. Data Governance5 percent

    Objectives

    • Demonstrate understanding of Unity Catalog tags and comments as mechanisms for adding metadata to securable objects to improve data discoverability.
    • Demonstrate understanding of the Unity Catalog permission inheritance model as a mechanism for managing access control across catalogs, schemas, and objects.
  8. 8. Debugging and Deploying10 percent

    Objectives

    • Identify pertinent diagnostic information using Spark UI, cluster logs, system tables, and query profiles to troubleshoot errors.
    • Analyze the errors and remediate the failed job runs with job repairs and parameter overrides.
    • Deploy Databricks resources using Declarative Automation Bundles (formerly Databricks Asset Bundles).
    • Configure Git-based CI/CD workflows using Databricks Git folders (formerly Repos) to deploy notebooks and code.
  9. 9. Data Modeling5 percent

    Objectives

    • Design scalable Delta/Iceberg table layouts for large data assets by mapping partitioning to data grain, aligning clustering to relationship access patterns, and maintaining balanced file sizes through compaction.
    • Design dimensional models for analytical workloads, leveraging Materialized Views for pre-computed aggregation and Unity Catalog Metric Views for governed, reusable metric definitions, to ensure efficient querying and aggregation.

What CertExam offers

Practice bank

249 questions are published after passing automated checks and an independent blind verification by someone other than the author; 40 of them are free to try without a card. Drafted-but-unverified questions are never counted or shown.

This bank meets CertExam's own launch criteria: at least 150 independently validated questions, every exam-guide objective covered by at least two questions, enough questions for two non-overlapping full-length mock exams, and every option explained with a citation to official public documentation that a second, independent check opened. That is CertExam's own bar, not the vendor's. It does not claim complete coverage of the exam, it says nothing about your result, and it does not mean the questions resemble real exam items. All questions are original.

Which modes you can use depends on your plan. See plans and pricing.

Practice modes

  • Quick practice
  • Diagnostic
  • Topic practice
  • Full mock exam
  • Weak areas
  • Review incorrect
  • Scenario practice
  • Code and SQL
  • Adaptive practice
  • Spaced repetition

Free samples

Sample questions with answers

9 sample questions from the bank, shown with their answers and explanations. Think it through, then open the answer. Create a free account to keep your progress on the free sample set.

Sample question 1

advancedDomain 1, objective 7

Billing and CRM systems both write change records for a customers table into one feed that an AUTO CDC flow applies to an SCD Type 1 target. Each record has customer_id, seq, every data column, and changed_cols, an array<string> that names the columns this record intends to change. Columns not named in changed_cols hold null placeholders. Some records deliberately set phone to null, and the target must store that null. Each producer adds new columns over time. Which configuration applies these records correctly?

Choose one answer.

  1. Option A:

    ignore_null_updates=True, so every null in a change record leaves the existing target value in place

  2. Option B:

    ignore_null_updates_column_list naming the columns that exist today

  3. Option C:

    columns_to_update pointing at changed_cols, combined with ignore_null_updates=True as a safeguard

  4. Option D:

    columns_to_update="changed_cols"

Show answer and explanation for sample question 1

Correct answer: D

Why

When a source sends only the columns that changed, AUTO CDC must tell an absent column from an explicit null. IGNORE NULL UPDATES treats nulls as do-not-update markers for the columns it covers, which suits a fixed, known set of columns. COLUMNS TO UPDATE (columns_to_update in Python) reads an array of column names from each change record, writes those columns including explicit nulls, and preserves the rest, so it fits records that update different columns. The two mechanisms cannot be combined.

Every option

  • A(incorrect)

    This treats every null as a do-not-update marker, so it can never store the deliberate null for phone. It solves the placeholder problem but breaks the explicit-null requirement.

  • B(incorrect)

    Listed columns ignore nulls, so phone could never be nulled if listed, and columns outside the list apply explicit nulls, so placeholders would overwrite stored values. A fixed list also needs a code change whenever a producer adds a column.

  • C(incorrect)

    columns_to_update cannot be set together with ignore_null_updates or its column-list variants, so this combination is not allowed.

  • D(correct)

    Correct. Each record names the columns to write, including explicit nulls, and every unlisted column keeps its target value. It suits changing sets of updated columns across producers.

Sample question 2

intermediateDomain 2, objective 1

A platform team publishes one Kafka topic per region, named orders-eu, orders-us, orders-apac, and new regions are added regularly using the same orders- naming convention. One streaming ingestion job must consume every matching topic without a code change when a region is added. Which Kafka source option fits?

Choose one answer.

  1. Option A:

    subscribePattern set to a Java regular expression such as orders-.*.

  2. Option B:

    subscribe set to a comma-separated list of the current region topics.

  3. Option C:

    assign set to a JSON string listing every topic partition.

  4. Option D:

    subscribe set to the single value orders-*, which the source expands as a wildcard.

Show answer and explanation for sample question 2

Correct answer: A

Why

The Kafka source requires exactly one way of specifying what to read: subscribe (explicit list of topics), subscribePattern (Java regex) or assign (specific topic partitions). When the set of topics follows a naming convention and grows, a pattern avoids code changes.

Every option

  • A(correct)

    Correct. subscribePattern takes a Java regex matching the topics to consume, so topics that follow the naming convention are covered by one setting.

  • B(incorrect)

    subscribe takes an explicit list of topic names, so each new region would require editing the job.

  • C(incorrect)

    assign pins the job to specific topic partitions, which is the least flexible choice and would also need editing for every new topic.

  • D(incorrect)

    subscribe expects literal topic names; pattern matching is the purpose of the separate subscribePattern option.

Sample question 3

intermediateDomain 3, objective 2

A VARIANT column event holds JSON events. Only some events contain a device object, so the path read by the query below is missing in many rows.

sql
SELECT variant_get(event, '$.device.os', 'string') AS os FROM silver_events;

What happens for rows where $.device.os does not exist?

Choose one answer.

  1. Option A:

    The query fails with an INVALID_VARIANT_CAST error for those rows.

  2. Option B:

    The query fails with a path-not-found error unless try_variant_get is used.

  3. Option C:

    The os column is the string 'null' for those rows.

  4. Option D:

    The os column is NULL for those rows and the query succeeds.

Show answer and explanation for sample question 3

Correct answer: D

Why

variant_get returns NULL when the path matches nothing. It raises INVALID_VARIANT_CAST only when the value exists but cannot be cast to the requested type, and try_variant_get converts that cast error into NULL.

Every option

  • A(incorrect)

    INVALID_VARIANT_CAST is raised when a value is found but cannot be cast to the requested type, not when the path is missing.

  • B(incorrect)

    A missing path is not an error in variant_get; try_variant_get only changes how failed casts are treated.

  • C(incorrect)

    The result is a SQL NULL of the requested type, not a text value.

  • D(correct)

    Correct. When the path does not match anything, variant_get returns NULL.

Sample question 4

advancedDomain 4, objective 4

A nightly job runs a SQL alert task named check_stock. A task restock_orders must run only when the alert is in the TRIGGERED state, and a task log_ok must run otherwise. The engineer set both downstream tasks to depend on check_stock with the default Run if setting, and both always run.

Which design implements the branching correctly?

Choose one answer.

  1. Option A:

    Set Run if to At least one failed on restock_orders, because a triggered alert makes the alert task fail

  2. Option B:

    Add an If/else task that tests {{tasks.check_stock.output.alert_state}} == TRIGGERED, with restock_orders on true and log_ok on false

  3. Option C:

    Add an If/else condition task that tests {{tasks.check_stock.result_state}} == failed to detect a triggered alert at runtime

  4. Option D:

    Add an If/else condition task that tests {{tasks.check_stock.output.alert_state}} > 0 to detect a triggered alert

Show answer and explanation for sample question 4

Correct answer: B

Why

A SQL alert task succeeds whenever the alert is evaluated, even when the alert is triggered, so Run if conditions on failure cannot branch on the alert result. Branch with an If/else condition task on the dynamic value {{tasks.<task>.output.alert_state}} (UNKNOWN, OK or TRIGGERED) using ==, and attach tasks to the true or false outcome.

Every option

  • A(incorrect)

    Wrong. The SQL alert task reports Succeeded whenever the alert evaluates, whether or not the condition is triggered; it fails only on errors such as a warehouse or query problem.

  • B(correct)

    Correct. The alert task exposes alert_state (UNKNOWN, OK or TRIGGERED) as a dynamic value, and an If/else task can compare it as a string to branch the run.

  • C(incorrect)

    Wrong. A triggered alert does not change the task's result state; the task still succeeds, so the condition would never be true for a triggered alert.

  • D(incorrect)

    Wrong. The > operator performs a numeric comparison and alert_state is a string such as TRIGGERED; the string comparison operators == and != are the ones that apply.

Sample question 5

intermediateDomain 5, objective 3

Several analysts run exploratory queries repeatedly against the same large Delta tables on a shared cluster with SSD-backed workers that support disk caching. One engineer proposes adding df.cache() to every notebook and an unpersist() at the end. Another suggests simply relying on the cluster as configured.

Which recommendation is most appropriate?

Choose one answer.

  1. Option A:

    Add df.cache() everywhere, because only the Spark cache can accelerate repeated reads of Delta tables.

  2. Option B:

    Rely on the disk cache, which is enabled by default on supported workers and populates automatically on first read.

  3. Option C:

    Disable the disk cache so that the Spark cache is the only cache and results stay consistent.

  4. Option D:

    Run CACHE SELECT for each table at the start of every session to pre-load the SSDs on Runtime 14.3.

Show answer and explanation for sample question 5

Correct answer: B

Why

The disk cache is automatic: files are cached on first read, stale entries are evicted when files change, and no code is needed. Databricks recommends it for repeated reads of Parquet and Delta data, while the Spark cache needs code changes and is lost when a worker is decommissioned. CACHE SELECT is ignored on current runtimes and SQL warehouses.

Every option

  • A(incorrect)

    The disk cache already accelerates repeated reads of Delta and Parquet data, with no code changes.

  • B(correct)

    Correct. Databricks recommends the disk cache for repeated reads of Parquet and Delta data.

  • C(incorrect)

    The two caches are not in conflict, and disabling the disk cache would only remove an automatic benefit.

  • D(incorrect)

    On SQL warehouses and Runtime 14.2 and above, CACHE SELECT is ignored in favour of an enhanced caching algorithm.

Sample question 6

intermediateDomain 6, objective 1

An engineer holds the MANAGE privilege on the Unity Catalog view prod.finance.v_revenue, which they do not own. The current owner has left the company and the engineer tries to transfer ownership of the view to a named colleague (an individual user in the account) with ALTER VIEW prod.finance.v_revenue OWNER TO \colleague@example.com``. What is the outcome, and what should the engineer do?

Choose one answer.

  1. Option A:

    The transfer is denied, because a MANAGE holder can transfer a view only to themselves or to a group they belong to; a metastore admin is needed.

  2. Option B:

    The transfer succeeds, because MANAGE lets its holder assign ownership of the view to any individual user in the account.

  3. Option C:

    The transfer succeeds once the colleague has been granted SELECT on the view and USE SCHEMA on the parent schema first.

  4. Option D:

    The transfer is denied and cannot be fixed, because a view whose owner has left cannot be reassigned to another principal.

Show answer and explanation for sample question 6

Correct answer: A

Why

All Unity Catalog securables have an owner, and owners and MANAGE holders can normally transfer ownership. To prevent privilege escalation, for views, functions and models, owners and MANAGE holders can transfer ownership only to themselves or to a group they belong to; only a metastore admin can transfer these objects to any other principal. Making a group the owner of a view is therefore a practical pattern.

Every option

  • A(correct)

    Correct. For a view, function or model, owners and MANAGE holders can transfer ownership only to their own username or to a group they are a member of; only a metastore admin can transfer it to any other user, service principal or group.

  • B(incorrect)

    MANAGE does not allow this for a view. Only a metastore admin can transfer ownership of a view to another user.

  • C(incorrect)

    Granting privileges to the recipient does not change the rule. For a view, a MANAGE holder is still restricted to transferring to themselves or to a group they belong to.

  • D(incorrect)

    All Unity Catalog securables have an owner, and ownership can be reassigned. A metastore admin can transfer the view to the colleague.

Sample question 7

intermediateDomain 7, objective 1

A data steward needs to add the tag domain = 'sales' to the table prod.sales.orders. The steward does not own the table or its parent objects, and domain is an ordinary (ungoverned) tag. Which set of permissions allows the steward to set the tag?

Choose one answer.

  1. Option A:

    SELECT and MODIFY on the table, plus USE SCHEMA and USE CATALOG on its parents

  2. Option B:

    BROWSE on the catalog and READ METADATA on the table, plus USE SCHEMA on the schema

  3. Option C:

    APPLY TAG on the table, USE SCHEMA on the schema and USE CATALOG on the catalog

  4. Option D:

    CREATE TABLE on the schema and MODIFY on the table, plus USE CATALOG on the catalog

Show answer and explanation for sample question 7

Correct answer: C

Why

To add a tag to a securable object, a principal must own the object or hold APPLY TAG on it together with USE SCHEMA on the parent schema and USE CATALOG on the parent catalog. SELECT and MODIFY cover reading and writing data, and BROWSE and READ METADATA only expose metadata; editing tags is the separate APPLY TAG privilege.

Every option

  • A(incorrect)

    SELECT and MODIFY cover reading and writing the table's data. Adding tags needs APPLY TAG, which this set does not include.

  • B(incorrect)

    BROWSE and READ METADATA let a principal view an object's metadata, including its tags, but not change it; editing tags needs APPLY TAG.

  • C(correct)

    Correct. A non-owner needs APPLY TAG on the object plus USE SCHEMA on its parent schema and USE CATALOG on its parent catalog.

  • D(incorrect)

    CREATE TABLE creates new tables and MODIFY changes data. Neither is APPLY TAG, and USE SCHEMA is also missing.

Sample question 8

intermediateDomain 8, objective 3

A bundle declares a variable and sets a value for the prod target.

yaml
variables:
  catalog:
    description: Target catalog
    default: dev_catalog

targets:
  prod:
    variables:
      catalog: prod_catalog

The project also contains .databricks/bundle/prod/variable-overrides.json with {"catalog": "sandbox_catalog"}. The CI runner has BUNDLE_VAR_catalog=staging_catalog set in its environment and runs databricks bundle deploy -t prod with no --var option. Which value does ${var.catalog} resolve to?

Choose one answer.

  1. Option A:

    prod_catalog, because the target-level value is the most specific setting for the prod target.

  2. Option B:

    dev_catalog, because the default in the top-level declaration is always used at deploy time.

  3. Option C:

    staging_catalog, because the BUNDLE_VAR_ environment variable outranks the overrides file and the target values.

  4. Option D:

    sandbox_catalog, because a variable-overrides.json file for the target replaces all other sources.

Show answer and explanation for sample question 8

Correct answer: C

Why

The CLI looks for a variable's value in a fixed order and stops at the first hit: --var option, BUNDLE_VAR_ environment variable, variable-overrides.json for the target, the target's variables mapping, and finally the declared default. With no --var option, the environment variable wins. Knowing this order explains why a value in the YAML can be silently superseded in CI.

Every option

  • A(incorrect)

    Target-level variables mappings rank below environment variables and the overrides file, so they apply only when nothing higher supplies a value.

  • B(incorrect)

    The default is the last resort in the lookup order, used only when no other source supplies a value.

  • C(correct)

    The CLI checks --var options first, then BUNDLE_VAR_ environment variables, so the environment variable is found first and the search stops.

  • D(incorrect)

    The overrides file is third in the order, below --var options and BUNDLE_VAR_ environment variables, so it is not reached when the environment variable exists.

Sample question 9

intermediateDomain 9, objective 1

A team is creating a Unity Catalog managed Delta table for clickstream events. The table currently holds about 600 GB and grows by roughly 20 GB per month. Nearly every query filters on a date range of event_ts. A teammate proposes PARTITIONED BY (event_date, event_hour) so that date filters are fast.

Based on Databricks guidance on partitioning, which design should the team choose?

Choose one answer.

  1. Option A:

    Partition by event_date and event_hour, so that date-range filters prune whole directories of data.

  2. Option B:

    Leave the table unpartitioned, rely on ingestion time clustering, and add liquid clustering on event_ts if needed.

  3. Option C:

    Add CLUSTER BY (event_ts) and also PARTITIONED BY (event_date), so that each date partition is clustered internally.

  4. Option D:

    Partition by event_date and schedule OPTIMIZE ... ZORDER BY (event_ts) to order data within each date partition.

Show answer and explanation for sample question 9

Correct answer: B

Why

Databricks states that most tables under 100 TB do not need partitioning, that tables under 1 TB should not be partitioned, and that each partition should contain at least 1 GB. Unpartitioned Delta tables are automatically clustered by ingestion time, which gives performance similar to date partitioning without tuning. For anything beyond that, liquid clustering is the recommended layout feature and it cannot be combined with partitioning.

Every option

  • A(incorrect)

    Wrong. Tables under 1 TB should not be partitioned, and each hourly partition here would hold only tens of megabytes, far below the 1 GB per partition that Databricks recommends. Partitioning below these minimums is likely to hurt queries.

  • B(correct)

    Correct. With less than 1 TB of data Databricks says not to partition. Unpartitioned Delta tables are automatically clustered by ingestion time, which gives partitioning-like performance for date filters, and liquid clustering is the recommended layout feature if more is needed.

  • C(incorrect)

    Wrong. Liquid clustering is not compatible with partitioning, so the two cannot be combined on one table. The table is also too small for partitioning to help.

  • D(incorrect)

    Wrong. This keeps the partitioning that guidance advises against for tables under 1 TB, and Databricks recommends liquid clustering over Z-ordering for all new tables.

FAQ

Data Engineer Professional: common questions

How many questions and how much time does the Data Engineer Professional exam have?

Under the guide we checked (Exam Published Oct 9, 2026 / Exam Guide Updated Oct 9, 2026 (new exam live in Webassessor from Oct 9, 2026)), Databricks describes 60 scored questions and a 120-minute time limit. The 9 Oct 2026 guide (p.2) says unscored questions may be included to gather statistical information, are not identified on the exam and do not affect the score, and that additional time is factored into the 120-minute limit. It states no number; the Oct 2026 transition guide (About the Exam) says up to 10. We checked these facts on 20 November 2026; confirm them with Databricks before you book.

What is the passing score for the Data Engineer Professional exam?

Databricks does not publish a passing score for this exam, so CertExam does not state one. Figures quoted on forums or by other providers are guesses unless Databricks publishes them.

Does CertExam offer practice questions for the Data Engineer Professional exam?

Yes. The bank has 249 independently validated questions, 40 of them free to try. The bank meets CertExam's own launch criteria: at least 150 independently validated questions, every exam-guide objective covered by at least two questions, enough questions for two non-overlapping full-length mock exams, and every option explained with a citation to official public documentation that a second, independent check opened. That is CertExam's own bar, not the vendor's. It does not claim complete coverage of the exam, it says nothing about your result, and it does not mean the questions resemble real exam items.

Where do these exam facts come from?

From Databricks's own published pages and exam guide, read and recorded on 20 November 2026. Links to the official sources are on this page. Vendors revise exams, so treat Databricks's pages as the final word.

Is CertExam affiliated with Databricks?

No. CertExam is an independent provider, not affiliated with or endorsed by Databricks. Databricks and the certification name are trademarks of their owner and are used only to describe the exam.