Free study guide · Databricks
How to pass the Databricks Certified Data Engineer Professional
What the exam covers, how to plan your study, and how to read its questions. From the CertLumen course for this exam, kept up to date with the official exam guide.
- Questions
- 60
- Time
- 120 minutes
- Passing score
- Not published by Databricks; this course targets ≥ 80% on unseen mocks
- Exam fee
- USD 200 (plus applicable taxes)
- Valid for
- 2 years (recertify by taking the current live exam)
- Delivery
- Online proctored or test center
What the exam covers
A study plan that works
The default path for engineers who hold the Data Engineer Associate (or know its material) and have built a few production pipelines. It covers all 23 lessons and the 10 labs in section order, with a section drill after each section and daily flashcards. Section 1 (23%) gets five days and Section 5 (15%) two; the 5% sections share days. Mock Exam 1 on day 19 measures readiness, day 20 fixes the weakest sections and Mock Exam 2 on day 21 confirms it. Sit the exam on day 22 or 23. Plan on about 1.5–2.5 hours a day.
| Day | Focus | Time |
|---|---|---|
| 1 | Kick-off: the exam, then bundle projects and tests | 95 min |
| 2 | Dependencies, serverless environments, compute choices and UDFs | 120 min |
| 3 | Lakeflow pipelines, AUTO CDC and stateful streaming | 130 min |
| 4 | Lab: CDC pipeline, then Lakeflow Jobs and control flow | 125 min |
| 5 | Lab: UDFs, tests and a control-flow job, then the p1 drill | 110 min |
| 6 | Formats, message buses and CDC pipelines | 110 min |
| 7 | OpenSharing, Clean Rooms and Federation, then the p2 lab and drill | 125 min |
| 8 | Window functions, joins at scale and VARIANT | 110 min |
| 9 | AI Functions and expectations, then the p3 lab and drill | 130 min |
| 10 | System tables, Lakehouse alerts, APIs and event logs | 120 min |
| 11 | Lab: monitor and gate a job, then the p4 drill | 100 min |
| 12 | Managed tables, predictive optimization and layouts | 115 min |
| 13 | Caching, Change Data Feed and the query profile, then the p5 lab and drill | 135 min |
| 14 | Least privilege, ABAC, de-identification and erasure | 125 min |
| 15 | Lab: pseudonymise, mask and erase, then governance metadata | 120 min |
| 16 | Lab: tags and inheritance, then diagnosing failures | 105 min |
| 17 | Bundles and Git-based CI/CD, then the p8 lab and drill | 145 min |
| 18 | Data modeling, then the p9 lab and drill | 130 min |
| 19 | Mock Exam 1 | 160 min |
| 20 | Weak-area remediation | 120 min |
| 21 | Mock Exam 2 and final review | 150 min |
Exam strategy
How the exam works
The Databricks Certified Data Engineer Professional exam checks whether you can build, optimize and run production-grade data engineering on Databricks: secure, reliable and cost-effective ETL in Python and SQL, streaming, orchestration, observability, governance, compliance, debugging and CI/CD. Almost every item is a scenario with a constraint (latency, cost, least privilege, least operational overhead, no double counting) and four plausible options, and many show code, YAML or a CLI command. You need to know how features behave, not only that they exist.
| Fact | Detail (exam guide of October 9, 2026, exam page and certification FAQ) |
|---|---|
| Scored questions | 60 multiple-choice questions, according to the exam guide. The exam page currently says 59; prepare for 60 |
| Unscored items | May be included to gather statistics. They aren't identified and don't affect your score; extra time is already in the time limit |
| Time limit | 120 minutes, about 2 minutes per question |
| Fee | USD 200 plus applicable taxes |
| Delivery | Online proctored or at a test center. You register on Webassessor and Kryterion delivers the exam |
| Test aids | None allowed |
| Languages | The exam page lists English, Japanese, Portuguese (BR) and Korean; the October 2026 guide lists English. Check the page when you register |
| Prerequisites | None required; related training highly recommended |
| Recommended experience | One year of hands-on experience with the data engineering tasks in the exam outline |
| Passing score | Not published |
| Validity | 2 years. To recertify you take the current live exam |
The guide asks candidates to check the exam page two weeks before the exam date to make sure they prepare for the current version. This course follows the outline that went live on October 9, 2026: nine sections, with Databricks SQL, Unity Catalog, Auto Loader, Lakeflow Pipelines, serverless compute, Lakeflow Jobs, Lakehouse Federation, OpenSharing, Clean Rooms and the medallion architecture named as core platform features.
What the Professional adds to the Associate
This course builds on the Data Engineer Associate course on this portal. The Associate checks that you can use each product correctly; the Professional checks that you can choose between them under production constraints, read code that is subtly wrong and fix it, and operate what you deployed. Expect longer stems and more code.
| Area | Associate | Professional adds |
|---|---|---|
| Format | 45 questions in 90 minutes; about six months of experience | 60 questions in 120 minutes; about one year of experience; more code, YAML and CLI in the stems |
| Code and testing | Read databricks.yml; PySpark and SQL basics | Bundle Python projects built as wheels, dependency failures on serverless, pandas and Unity Catalog UDFs, assertDataFrameEqual and integration tests |
| Pipelines and streaming | Declarative pipelines, expectations, AUTO CDC basics | Streaming table vs materialized view by latency, cost and refresh; SCD Type 2 details; watermarks, output modes, foreachBatch, checkpoints; Structured Streaming vs pipelines |
| Ingestion | Auto Loader, COPY INTO, Lakeflow Connect overview | Kafka, Kinesis, Pub/Sub, Iceberg, database CDC connectors, OpenSharing, Clean Rooms, Lakehouse Federation |
| Transformation | Joins, aggregations, deduplication | Window frames, join strategies and skew, VARIANT, AI Functions, quarantine patterns |
| Operations | Jobs UI, Spark UI basics | System tables, REST API, CLI and SDK, event logs, Lakehouse alerts, query profiles, repair runs with parameter overrides, log delivery |
| Optimization | Liquid clustering and predictive optimization | Deletion vectors, Change Data Feed (legacy and automatic), caches, layouts by table size, query profile insights |
| Security and governance | Privileges, row filters, masks, ABAC basics | Least privilege across Unity Catalog and workspace ACLs, ABAC evaluation rules at scale, anonymisation, PII pipelines, GDPR erasure, inheritance edge cases |
| Modeling | Gold objects | Partitioning by data grain, clustering on join keys, compaction, star schemas, materialized views and metric views |
Most misses at this level come from behaviour details: an update-mode stream into a Delta table, a jobs/reset that drops the schedule, an SQL alert task that succeeds when the alert fires, an audit query filtered on workspace_id, an ABAC policy that never matches because the tag is on the table and not the column. Read the lessons' trap callouts twice.
Section weights: where the points are
Section 1 alone is almost a quarter of the exam, and Sections 1, 5, 2 and 3 together carry 62%. The question counts are this course's estimates from the published weights; they are also how the two mock exams are built. Individual exam forms vary.
| Section | Weight | ≈ Questions | Highest-yield topics |
|---|---|---|---|
| p1 Developing Code for Data Processing | 23% | 14 | Bundle project layout and wheel tasks, serverless environments and dependency errors, UDF types, streaming table vs materialized view, AUTO CDC SCD 1/2, stateful streaming, Jobs API and control flow |
| p2 Data Ingestion & Acquisition | 12% | 7 | Message bus start positions and auth, skipChangeCommits vs CDF, managed vs foreign Iceberg, Lakeflow Connect database CDC, D2D vs open sharing, Clean Rooms, federation grants |
| p3 Data Manipulation | 12% | 7 | Window frames and ranking, semi and anti joins, AQE skew and broadcast, VARIANT paths and casts, ai_query options, expectation actions and quarantine |
| p4 Monitoring and Alerting | 10% | 6 | Billing attribution, audit queries, regional vs global tables, alert states and notification frequency, alert-task branching, runs/list and event logs |
| p5 Cost & Performance Optimization | 15% | 9 | Managed tables and predictive optimization, CLUSTER BY AUTO, liquid vs partitioning vs Z-order, deletion vectors, caches, CDF, query profile |
| p6 Data Security and Compliance | 8% | 5 | Least-privilege grants, run-as identities, ABAC rules, masking tiers, pseudonymisation techniques, Data Classification, erasure and retention |
| p7 Data Governance | 5% | 3 | Comment and tag syntax and privileges, governed vs ungoverned tags, inheritance, revoking inherited grants, MANAGE vs ownership |
| p8 Debugging and Deploying | 10% | 6 | Evidence on classic vs serverless, Spark UI skew rule, repair API, bundle plan/deploy/bind, modes and variable precedence, Git folders |
| p9 Data Modeling | 5% | 3 | Partitioning thresholds, clustering on join keys, autotuned file sizes, incremental materialized views, metric views |
| Total | 100% | 60 |
How to use this course
- Learn. Read each lesson with its trap and exam callouts, answer the self-checks before you open them, and write down any key point you would have missed.
- Lab. The ten labs run on Databricks Free Edition unless a step says it needs a paid workspace. Some behaviours only stick once you have seen them, such as a downstream stream failing on an AUTO CDC target or a DELETE that leaves rows in a Parquet file.
- Drill. Take each section's drill untimed after its lessons and read every explanation, including the notes on the options you rejected. The questions in the two mocks stay out of the drills by default (*Keep the mock-exam questions unseen*), so the mocks stay a true test; the drills use all the others. After both mocks, untick the box and drill every question by section.
- Flashcards. Short daily sessions for defaults, limits and syntax: they decide many Professional questions.
- Mocks. The two mock exams follow the exam's section mix (60 questions, 120 minutes). Take each once, timed, in one sitting, with no notes.
- Review loop. Keep an error log (topic, why you missed it, the rule, where it lives) and retest the topic 48 hours later.
The guide recommends the instructor-led *Advanced Data Engineering with Databricks* course, whose four modules are also self-paced in Databricks Academy: *Advanced Techniques with Apache Spark Declarative Pipelines*, *Databricks Data Privacy*, *Databricks Performance Optimization* and *Automated Deployment with Declarative Automation Bundles*. It also points to the Databricks AI Prep Guide, whose *Hands-on minimum* step generates 6–10 Free Edition tasks mapped to the objectives. Use them alongside this course.
Readiness criteria
- ≥ 80% on both mock exams on the first attempt, each taken timed in one sitting.
- ≥ 70% in every section across both mocks combined. For p7 and p9, which have only three questions per mock, use the section drill score instead.
- ≥ 80% on every section drill.
- You finish a mock in ≤ 100 minutes, leaving time for a review pass.
- Without notes, you can rebuild these cheat-sheet tables: streaming table vs materialized view, output modes and sinks, Jobs API calls, layout by table size, ABAC evaluation rules, repair API fields and bundle modes.
- You can name today's product for every entry in the renames table.
Databricks doesn't publish the passing score. Aiming for 80% on unseen mocks leaves a margin for a harder exam form, unfamiliar wording and nerves.
Time management
120 minutes for 60 scored items plus any unscored ones is about 2 minutes per question. Recall items take under a minute; code and multi-constraint items can take three. Use three passes.
- First pass (≈ 85 minutes). Answer everything you can within about 90 seconds. If two options survive, pick the better one, flag the question and move on. Never leave a question blank.
- Second pass (≈ 25 minutes). Return to flagged questions. Re-read the constraint words in the stem before you re-read the options.
- Final check (≈ 10 minutes). Make sure nothing is unanswered. Change an answer only for a concrete reason, such as a constraint you missed.
| Checkpoint | Target elapsed time |
|---|---|
| Question 20 | ≤ 35 minutes |
| Question 40 | ≤ 70 minutes |
| Question 60 | ≤ 100 minutes, then flagged questions only |
Unanswered questions earn nothing, and unscored items look like scored ones. When time runs short, answer every remaining question first, then go back to the flags.
Reading Professional-level scenario questions
- Read the last sentence (the ask) first: *which change*, *what is the cause*, *which configuration meets both requirements*.
- Find the compute: serverless or classic, pipeline or notebook, SQL warehouse or job. Many answers flip on it (Spark UI, triggers, Spark configs, init scripts).
- Find every constraint word and predict the answer before you read the options.
- When code is shown, check it line by line against the checklist below; the bug is usually one keyword.
- Eliminate options that break any constraint, then pick the managed, governed option among those left.
Words that point to the answer
Phrases in a question that usually point to one answer. The course has the full list.
| When the question says | Think of |
|---|---|
| gold aggregate must always reflect the full history, including late-arriving updates to prior days | A materialized view |
| append-only events must be queryable within seconds, and each update should process only new rows | A streaming table in a continuous pipeline |
| bound state growth from very late events and resume after a driver failure without double counting | A watermark on the event-time column plus the query's checkpoint |
| keep only the latest running total per key in a Delta table from a streaming aggregation | foreachBatch with a MERGE on the key |
| a micro-batch appeared twice in the target after a restart from the same checkpoint | Write with txnAppId and txnVersion = batch_id, or a sequenced MERGE |
| the team changed the aggregation's grouping key and the restart fails with a state schema error | A new checkpoint location, plus a backfill |
processingTime trigger fails on serverless with INFINITE_STREAMING_TRIGGER_NOT_SUPPORTED | trigger(availableNow=True) scheduled by a Lakeflow Job, or a continuous pipeline |
| custom per-key state machine with timers that fire when no event arrives | Structured Streaming with transformWithState / transformWithStateInPandas |
SCD Type 2 history with minimal code; a change only to last_login must not create a new version | AUTO CDC ... STORED AS SCD TYPE 2 with TRACK HISTORY ON * EXCEPT (last_login) |
| change events arrive out of order and two events can share a timestamp | SEQUENCE BY STRUCT(event_ts, tie_breaker) |
Revise from the cheat sheet
The course's cheat sheet has a page for each of these:
- Code, pipelines and jobs
- Ingestion sources, CDC and sharing
- Windows, joins, VARIANT, AI Functions and expectations
- System tables, alerts, APIs and event logs
- Managed tables, layouts, caches, CDF and the query profile
- Least privilege, ABAC, de-identification and erasure
- Tags, comments and inheritance
- Diagnosis, repair, bundles and Git folders
- Table layouts and dimensional models
Practise with the real exam's mix
340 exam-style questions, 4 timed practice exams, flashcards and an AI tutor.