Databricks Data Engineer Professional format
Here is what Databricks publishes about the Data Engineer Professional exam. The figures are for the version that went live on 9 October 2026.
| Certification | Databricks Certified Data Engineer Professional |
| Scored questions | 60 |
| Unscored questions | Up to 10 may be included, unmarked; extra time is already in the limit |
| Duration | 120 minutes |
| Question format | Multiple choice |
| Registration fee | $200, plus local taxes |
| Validity | 2 years, then recertification |
| Prerequisites | None (related training highly recommended) |
| Recommended experience | One year hands-on with the exam guide’s data engineering tasks |
| Delivery | Online proctored or test centre |
| Languages | English, Japanese, Portuguese (Brazil), Korean (Databricks’ exam page; it does not say whether the October 9 outline is localized yet) |
| Test aids | None permitted |
| Passing score | Not published by Databricks |
The outline changed on 9 October 2026
Databricks replaced the outline on 9 October 2026. Anyone who sat the exam up to 8 October took the old version: 59 scored questions across ten sections, offered in English, Japanese, Portuguese (Brazil) and Korean. From 9 October the exam has 60 scored questions across nine sections. Databricks’ exam page still lists the same four languages and does not say whether the new outline is localized yet. Time, fee and validity did not change.
What moved:
- Data Sharing and Federation is no longer a section. OpenSharing, the new name for Delta Sharing (Databricks-to-Databricks and Databricks-to-Open), Lakehouse Federation and the new Clean Rooms objective now sit in Data Ingestion & Acquisition, which grows from 7% to 12% and also adds CDC from SQL Server, MySQL and PostgreSQL through Lakeflow Connect, and Iceberg as a format and target.
- Data Transformation, Cleansing, and Quality became Data Manipulation (10% to 12%). It adds the VARIANT type for semi-structured data and AI functions such as
ai_query; quarantining bad data is now framed as pipeline expectations that quarantine, drop or fail. - Developing Code grows from 22% to 23%. New items: SQL UDFs and Unity Catalog functions, SCD Type 1 and 2 through AUTO CDC, serverless compute settings, and stateful Structured Streaming (watermarks, output modes,
foreachBatch, checkpoints). - Cost & Performance Optimisation grows from 13% to 15%. It adds Predictive Optimization,
CLUSTER BY AUTOand the Delta cache, and takes over the Liquid Clustering versus partitioning and Z-ordering comparison from Data Modelling. - Security and Compliance falls from 10% to 8% and adds attribute-based access control (ABAC) with governed tags. Governance falls from 7% to 5% and now names tags and comments. Data Modelling goes to 5% and adds Iceberg layouts, materialised views and Unity Catalog metric views.
- Monitoring and Alerting and Debugging and Deploying stay at 10% each.
On 9 October the certification web page still showed the old weights. The exam guide PDF dated October 2026 carries both outlines and is the source for everything here. Its sample questions follow the new outline.
Section weighting
| Section | Weight | Approx. scored questions |
|---|---|---|
| 1. Developing Code for Data Processing using Python and SQL | 23% | ~14 |
| 2. Data Ingestion & Acquisition | 12% | ~7 |
| 3. Data Manipulation | 12% | ~7 |
| 4. Monitoring and Alerting | 10% | ~6 |
| 5. Cost & Performance Optimisation | 15% | ~9 |
| 6. Ensuring Data Security and Compliance | 8% | ~5 |
| 7. Data Governance | 5% | ~3 |
| 8. Debugging and Deploying | 10% | ~6 |
| 9. Data Modelling | 5% | ~3 |
Section numbers follow the exam guide. Counts are derived from Databricks’ percentages against 60 scored questions and rounded.
Nine sections changes how you prepare
Most certifications have three to five domains and a dominant one. This has nine, and only one above 15%.
The practical consequences:
Breadth is compulsory. Security, governance and modelling are 8%, 5% and 5%: about eleven questions together. Individually small; collectively 18% of the exam. Skip all three and you have given away nearly a fifth of the paper.
No single area saves you. Being excellent at code development covers 23% and leaves 77% elsewhere.
Small sections deserve proportionate time, not zero. Governance is about three questions on tags, comments and permission inheritance. Learn them properly in an hour; do not spend a weekend.
Two minutes per question
60 questions in 120 minutes. Comfortable, and the same effective pace as the Associate despite the harder content.
Code is read, not written
Multiple choice throughout, so even the 23% code section presents you with Python and SQL and asks what it does, what is wrong with it, or which version is correct. You need fluent code-reading rather than the ability to produce working code under time pressure, but the fluency required is real.
The passing score is not published
Databricks does not publish one. Aim for consistent 80%+ on realistic practice material rather than trusting an unsourced figure.
Two-year validity
Recertification every two years, and Databricks says you must take the current live exam rather than the version you passed.
What the exam is really about
The Associate asks whether you can build a working pipeline. The Professional asks whether you can build one that is efficient, observable, secure and debuggable, and whether you can read the code that does it.
If your work is notebooks that run once and are never tuned, this is a stretch. If you own pipelines other teams depend on, it describes your job.
Booking
Book through Databricks. Nothing gates registration, and the Associate is not a prerequisite.