Databricks Data Engineer Associate: The 45-Question Split and What Each Domain Costs You
Forty-five scored questions in 90 minutes for $200. Databricks publishes exact domain weights, so you can work out that ingestion and transformation alone are worth about 19 questions - here is the full split.
- 45 scoredQuestions
- 90 minTime
- $200Cost
- 2 yearsValid
Table of Contents
The Databricks Certified Data Engineer Associate paper is 45 scored questions in 90 minutes and costs $200. Ingestion at 21% and transformation at 22% are 43% of the paper between them - roughly 19 of 45 questions. Everything else is a supporting act at associate level.
What follows is the question split by domain, then five practice questions written to Databricks’s published objectives, each with the reasoning for the right answer and for every wrong one. These are original practice items, not real exam content – the point is to show you how the questions are shaped and where the traps sit.
How the 45 scored questions split
Databricks publishes percentage weights for this exam, which means you can convert them into the number of questions you will actually face. On a 45-question paper the split looks like this.
| Domain | Weight | Questions | |
|---|---|---|---|
| Databricks Intelligence Platform | 6% | ~3 | |
| Data Ingestion and Loading | 21% | ~9 | |
| Data Transformation and Modeling | 22% | ~10 | |
| Working with Lakeflow Jobs | 16% | ~7 | |
| Governance and Security | 15% | ~7 | |
| Implementing CI/CD | 10% | ~4 | |
| Troubleshooting, Monitoring, and Optimization | 10% | ~4 |
Read this as a revision budget. The domains at the top of the table are where marks are won and lost. Time spent on the bottom rows has a much lower ceiling.
Five practice questions
Answer each one before reading the key underneath it.
New JSON files land continuously in cloud storage and must be ingested exactly once without rescanning the whole directory. Which approach fits?
- A A nightly batch job that reads the full directory and overwrites the target table
- B Auto Loader, which tracks which files have already been processed
- C A COPY INTO run that reprocesses every file each time
- D Manually listing new files and passing them to a read job
Answer: B
Auto Loader is built for incrementally ingesting files as they arrive, keeping its own record of processed files so a growing directory is never rescanned in full.
Why the others fail
A A full-directory nightly read gets slower every day and is not incremental.
C COPY INTO is idempotent for files it has seen, but the option as written describes reprocessing everything, which is the behaviour the question rules out.
D Manual file listing is exactly the bookkeeping Auto Loader exists to remove, and it breaks the moment a run is missed.
In a bronze, silver and gold layout, what belongs in the silver layer?
- A Raw ingested data in its original form
- B Cleaned, conformed and deduplicated data ready to be joined and aggregated
- C Business-level aggregates that feed dashboards
- D Archived data no longer queried
Answer: B
Silver is the cleaned and conformed layer. Raw arrives in bronze, silver applies quality and structure, and gold holds the business aggregates.
Why the others fail
A Raw, unmodified data is the definition of the bronze layer.
C Business aggregates for reporting are the gold layer.
D The medallion layers describe refinement stages, not an archive tier.
Two workspaces must share one governed catalogue of tables with consistent permissions. Which capability provides this?
- A Workspace-local table ACLs configured identically in both
- B Unity Catalog, which governs data across workspaces from a single metastore
- C Cluster policies restricting who may start compute
- D Mounting the same storage location in both workspaces
Answer: B
Unity Catalog provides a single metastore with centrally managed permissions that apply across workspaces, which is precisely the stated requirement.
Why the others fail
A Duplicating ACLs in two places guarantees they will drift apart.
C Cluster policies govern compute configuration, not access to data objects.
D A shared mount exposes files without governing tables, permissions or lineage.
A job has three tasks where the third must run only if both earlier tasks succeed. How is this expressed?
- A Three separate jobs each on its own schedule, timed to not overlap
- B One job with task dependencies, so the third task depends on the first two
- C A single notebook containing all three steps in sequence
- D Three tasks with no dependencies and a retry policy on the third
Answer: B
Jobs support multi-task workflows with explicit dependencies, so a task runs only after the tasks it depends on have succeeded.
Why the others fail
A Timing-based coordination is fragile - a slow first job silently breaks the ordering.
C One notebook works but loses per-task retries, isolation and observability, which is why multi-task jobs exist.
D Retries do not create ordering. Without dependencies all three start together.
A Delta table has accumulated a very large number of small files and queries have slowed. What is the appropriate remedy?
- A Drop and recreate the table from source each night
- B Compact the small files, for example with OPTIMIZE
- C Increase the cluster size until queries return quickly again
- D Disable transaction logging on the table
Answer: B
The small-file problem is fixed by compaction. OPTIMIZE rewrites many small files into fewer larger ones, which is the standard Delta remedy.
Why the others fail
A Rebuilding nightly is expensive, loses history, and the small files simply come back.
C More compute masks the symptom while the per-file overhead stays exactly the same.
D The transaction log is what makes the table reliable. Disabling it is not an optimisation.
What catches people out
- Databricks publishes weights but not a pass mark. Plan for a comfortable margin rather than aiming at a number nobody has stated.
- The exam guide is versioned and the domain names change - Lakeflow Jobs is the current naming. Study against the current guide, not a two-year-old blog.
- Governance is 15%, larger than CI/CD or optimisation. Unity Catalog is not optional background knowledge at associate level.
- Recertification is every two years, so factor the renewal into the cost, not just the $200 first sitting.
What it really costs to pass
The sticker price is $200, but that is the cost of passing first time. The number worth budgeting is different.
- First attempt: $200.
- Resit: Databricks does not publish a discounted resit alongside the headline fee, so budget a second full fee unless the official page says otherwise when you book.
- Renewal: the credential lasts 2 years, so the honest cost is the fee divided across that period, plus whatever renewal Databricks requires at the end of it.
- Your time: 90 minutes in the chair, and realistically several weeks of preparation before it. That is the largest cost on this list and the only one you cannot pay to avoid.
There is no published pass mark. Databricks does not publish one for this exam, so you cannot calculate how many mistakes you can afford. Prepare for a comfortable margin rather than aiming at a threshold nobody has stated.
Who should sit it: there are no formal prerequisites, though hands-on data engineering experience is recommended. Going in well under that bar is usually how people end up paying the fee twice.
How to prepare
A workable sequence for this exam, assuming you already work near the material:
- Read the official guide first. Everything on this page is secondary to what Databricks publishes at the link in the sidebar, and guides get revised.
- Work the heaviest domain until you can teach it. Use the table above to decide which one that is.
- Practise under the clock. 90 minutes for 45 scored questions is roughly 120 seconds each. Knowing the material and being able to apply it at that pace are different skills.
- Review every wrong answer to the level of the distractor. Being able to say why the other three options fail is what turns a 60% into a pass.
Frequently asked questions
How many questions are on the Databricks Data Engineer Associate exam?
Forty-five scored questions in 90 minutes. Additional unscored items may be included for statistical purposes.
What is the passing score for the Databricks Data Engineer Associate?
Databricks does not publish a passing score for this exam. Prepare for a comfortable margin rather than a specific threshold.
How much does the Databricks Data Engineer Associate cost?
A $200 registration fee, with the exam delivered online or at a test centre.
How long is the certification valid?
Two years, after which recertification is required to keep the credential active.
Sources
Every figure on this page comes from the certifying body rather than a third-party summary:
Checked September 2, 2026. Formats and fees change – confirm against Databricks before you pay for anything.