Databricks Certified Data Analyst Associate Practice Exam

What the Databricks Certified Data Analyst Associate exam covers

• Understanding of Databricks Data Intelligence Platform84 questions
• Managing Data66 questions
• Importing Data50 questions
• Executing queries using Databricks SQL and Databricks SQL Warehouses152 questions
• Analyzing Queries108 questions
• Working with Dashboards and Visualizations in Databricks120 questions
• Developing, Sharing, and Maintaining AI/BI Genie spaces92 questions
• Data Modeling with Databricks SQL50 questions
• Securing Data66 questions

Free Databricks Certified Data Analyst Associate practice test questions

A sample of 10 questions with answers and explanations. Sign up free to practice all 788.

1. Question 1Analyzing Queries

   A data analyst notices that a Databricks SQL query with heavy aggregations runs much faster on one SQL warehouse than on another warehouse of the same size. The faster warehouse has this engine enabled. What is it?

   • APhotonCorrect
   • BUnity Catalog lineage
   • CDelta Sharing
   • DAuto Loader
   ✓ Correct answer: A

   Photon is the execution engine that decides how fast a warehouse processes SQL operators once size and node count are otherwise equal. It runs natively over vectorized, columnar batches instead of routing execution through the JVM, which strips out most of the per-row overhead. That matches the scenario exactly: two warehouses of the same size differing only in whether Photon is enabled, running an aggregation-heavy query, which is precisely the workload Photon accelerates most. When two same-sized warehouses perform differently, checking whether Photon is enabled should be an early diagnostic step.

   Why the other options are wrong

   • BUnity Catalog lineage tracks how data flows between tables and dashboards for governance and impact analysis; it has no effect on how fast a query executes.
   • CDelta Sharing is an open protocol for securely sharing live data with other organizations or platforms; it does not change how queries run inside a warehouse.
   • DAuto Loader incrementally ingests new files as they arrive in cloud storage; it is an ingestion tool, not something that speeds up aggregation queries.
2. Question 2Analyzing Queries

   Which statement accurately describes how Liquid Clustering applies clustering to new data written to a table?

   • AClusters incrementally as OPTIMIZE runsCorrect
   • BRewrites every file on every insert
   • CClusters data only when queried
   • DApplies clustering only once, at creation
   ✓ Correct answer: A

   Liquid Clustering absorbs newly written data without reorganizing the whole table, then relies on OPTIMIZE, typically scheduled as a periodic maintenance job, to incrementally rewrite files according to the declared clustering columns. That incremental model is the point. A full-table rewrite on every insert would be prohibitively expensive at scale, so the table is allowed to drift slightly out of clustering between maintenance runs and each OPTIMIZE pulls it back. Clustering is maintained continuously through those periodic runs, not applied the instant data lands and not fixed once at creation.

   Why the other options are wrong

   • BRewriting every file on every insert would be extremely costly at scale and is not how Liquid Clustering is designed to operate.
   • CClustering is applied through maintenance operations like OPTIMIZE, not triggered automatically by running a SELECT query against the table.
   • DClustering can and should be reapplied over time as more data is written; it is not a single one-time action performed only at creation.
3. Question 3Working with Dashboards and Visualizations in DatabricksSelect all that apply

   An analyst tries to plot CPU utilization over 24 hours for 50 individual servers as 50 separate lines on one chart, but the result is an unreadable tangle of colors. Which TWO reasons make a heatmap (servers by hour, color as CPU percent) a better alternative? (Choose TWO)

   • AIt removes the need to collect hourly data
   • BColor lets patterns be scanned without overlapping linesCorrect
   • CIt always shows more precise values than a line chart
   • DA pie chart would organize 50 servers better
   • EIt encodes server, hour, and value in one gridCorrect
   ✓ Correct answer: B, E

   A heatmap beats 50 overlapping lines for two concrete reasons. Color intensity in a grid cell can be scanned at a glance, with no need to trace 50 tangled line colors to follow one server. And the grid encodes all three facts at once, server on one axis, hour on the other, CPU percent as color, a compactness a line chart cannot match once the series count grows that large. The tradeoff is deliberate: a heatmap gives up exact-value precision and individual-series detail in exchange for pattern recognition across many series. Reach for it when the question is where the hot spots are, not what the exact reading was.

   Why the other options are wrong

   • AThe heatmap still requires the same hourly measurements to be collected; it changes the display, not the data collection need.
   • CHeatmaps favor pattern recognition over exact-value precision; a table would show precise values more reliably than color intensity.
   • DA pie chart cannot represent two dimensions like server and hour at once, and 50 slices would be even less readable than 50 lines.
4. Question 4Developing, Sharing, and Maintaining AI/BI Genie spaces

   An insurer already has an AI/BI dashboard built on a curated set of claims tables. Which capability lets a team quickly stand up a related Genie space using those same datasets?

   • ACreating a Genie space directly from the existing dashboardCorrect
   • BExporting the dashboard as a PDF and re-uploading it as a table
   • CCloning the dashboard's SQL warehouse into a new catalog
   • DConverting the dashboard into a Delta Live Tables pipeline
   ✓ Correct answer: A

   Databricks provides a shortcut for exactly this situation: generating a Genie space directly from an already-published dashboard carries that dashboard's curated datasets across automatically, so the team never has to re-select the same claims tables and views by hand. Beyond saving setup time, it keeps the new conversational space grounded in data that has already been vetted, scoped, and governed for the dashboard. What Genie needs is a set of live, queryable datasets to generate fresh SQL against, which is why moving compute alone or capturing a rendered snapshot does not get the team there.

   Why the other options are wrong

   • BExporting the dashboard as a PDF and re-uploading it as a table would only capture a static, already-rendered snapshot, not a live, queryable dataset Genie could use.
   • CCloning the SQL warehouse into a new catalog only moves compute configuration; it does not carry over the dashboard's curated dataset selection that a Genie space needs.
   • DDelta Live Tables is used to build data transformation pipelines; converting a dashboard into one has no relationship to creating a conversational Genie space.
5. Question 5Developing, Sharing, and Maintaining AI/BI Genie spacesSelect all that apply

   Before rolling a Genie space out to the entire company, which TWO practices help establish a trustworthy accuracy baseline? (Choose TWO)

   • ARunning a benchmark set of questions with known-correct answersCorrect
   • BAssuming accuracy is fine since setup was easy
   • CSkipping review since users will report problems
   • DHaving domain experts review a sample of generated answersCorrect
   • EHiding the space from all stakeholders indefinitely
   ✓ Correct answer: A, D

   These two practices complement each other before a company-wide launch. Running a benchmark set of questions with known-correct answers gives an objective, repeatable measurement of accuracy that can be re-run later to catch regressions, turning validation into a process rather than a guess. Having domain experts review a sample of generated answers adds the qualitative half: someone who actually understands the business can catch a technically plausible answer that is still wrong in a way an automated check would miss. Together they produce a baseline you can defend and re-measure.

   Why the other options are wrong

   • BHow easy the space was to configure says nothing about whether the answers it produces are actually correct.
   • CWaiting for users to report problems after a wide rollout is reactive, not a way to establish a baseline before one.
   • EHiding the space from stakeholders indefinitely blocks both the feedback and adoption a rollout needs, and is not a valid strategy.
6. Question 6Managing Data

   An analyst notices that a phone_number column has inconsistent leading and trailing whitespace that is breaking exact-match joins to another table. Which function directly removes this extra whitespace?

   • ALENGTH(phone_number)
   • BTRIM(phone_number)Correct
   • CCAST(phone_number AS INT)
   • DROUND(phone_number, 0)
   ✓ Correct answer: B

   TRIM(phone_number) strips extra spaces from the start and end of a string value, which is a frequent and often invisible cause of failed exact-match joins between tables that hold the same logical value. Whitespace differences are impossible to spot when scanning data visually, yet they break the strict equality comparison a join depends on. Applying TRIM in the SELECT list or directly inside the JOIN condition fixes the comparison without altering the stored table. Takeaway: reach for TRIM whenever an exact-match join unexpectedly fails even though the values look the same.

   Why the other options are wrong

   • ALENGTH returns the character count of a string value; it measures whitespace but does not remove it.
   • CCasting a phone number to an integer strips formatting characters unpredictably and is not the correct fix for stray whitespace.
   • DROUND applies to numeric values and rounding precision, which has no meaning for a text column like phone_number.
7. Question 7Data Modeling with Databricks SQL

   In a well-formed star schema for an airline's flight bookings, the fact table's foreign key columns should reference which kind of column in each dimension table?

   • AThe dimension surrogate keyCorrect
   • BDescriptive text attribute
   • CNatural/business source key
   • DThe dimension load timestamp
   ✓ Correct answer: A

   A well-formed star schema joins fact rows to dimensions through the surrogate key: a stable, warehouse-generated identifier that stays valid even when the source system changes or reuses its own key. That stability is also what makes SCD Type 2 workable, since each historical version of a dimension row gets its own surrogate key for facts to point at. The alternatives fail on either stability or purpose. A natural business key belongs to a system outside the warehouse's control, descriptive attributes are values to report on rather than join on, and load metadata records when a row arrived rather than identifying which row it is.

   Why the other options are wrong

   • BDescriptive text attributes, like a product name, are never used as the join key between a fact table and its dimension.
   • CA natural or business source key can change or be reused by the source system over time, making it unreliable as a long-term join target.
   • DA load timestamp is metadata recording when a row was loaded; it is never used as the key joining a fact table to a dimension.
8. Question 8Executing queries using Databricks SQL and Databricks SQL Warehouses

   Which clause restricts a query to only rows where status equals 'shipped'?

   • AHAVING status = 'shipped'
   • BORDER BY status = 'shipped'
   • CWHERE status = 'shipped'Correct
   • DGROUP BY status = 'shipped'
   ✓ Correct answer: C

   WHERE evaluates its condition against each row before any grouping or aggregation happens, keeping only the rows where status equals 'shipped' and discarding everything else. HAVING looks similar but operates after GROUP BY, filtering aggregated groups rather than raw rows, so it does not fit a simple row-level condition with no aggregation involved. ORDER BY only controls the sequence in which surviving rows are returned and cannot filter anything out, and GROUP BY defines which columns to aggregate by rather than accepting a filter condition like this one.

   Why the other options are wrong

   • AHAVING filters aggregated groups after GROUP BY runs; it is not the right clause for a simple row-level condition with no aggregation.
   • BORDER BY only controls the sort order of the returned rows; it cannot filter any rows out of the result.
   • DGROUP BY defines the columns used for aggregation; it does not accept a filter condition like this one.
9. Question 9Executing queries using Databricks SQL and Databricks SQL WarehousesSelect all that apply

   Which two statements about Delta Lake time travel retention are correct? (Choose TWO)

   • ADelta tables retain every historical version forever with no possibility of losing access
   • BHow far back time travel can go depends on log and data file retention settingsCorrect
   • CTime travel retention is controlled exclusively by the SQL warehouse's auto-stop setting
   • DRunning VACUUM can remove data files needed for older versions, limiting time travel rangeCorrect
   • ETime travel is unlimited only for external tables, never for managed tables
   ✓ Correct answer: B, D

   Delta Lake's time travel range is governed by configurable retention settings covering both the transaction log and the underlying data files. Running VACUUM physically removes data files that older versions still depend on, which directly shortens how far back time travel can successfully reach. Those two facts together are why retention is a design decision a team has to manage actively, balancing storage cost against how far back they might ever need to look. A table with aggressive VACUUM settings or short log retention can lose the ability to read a version that still appears in its history, regardless of whether the table is managed or external.

   Why the other options are wrong

   • ADelta tables do not retain every historical version forever with no possibility of losing access; retention is governed by configurable thresholds and can be shortened by VACUUM.
   • CSQL warehouse auto-stop controls compute idling; it has no bearing on Delta version retention.
   • ETime travel availability depends on log/data retention settings, not on whether a table is managed or external.
10. Question 10Understanding of Databricks Data Intelligence Platform

    A data engineering team wants to define a multi-stage ETL pipeline declaratively, with built-in data quality checks and automatic handling of pipeline errors. Which component is designed for this?

    • ALakeflow Jobs
    • BDelta Live TablesCorrect
    • CMosaic AI
    • DAI/BI Dashboards
    ✓ Correct answer: B

    Delta Live Tables is built for exactly this: a team declares the tables and transformations for each stage of a multi-stage ETL pipeline along with data quality expectations, and the framework manages dependency ordering, incremental processing, retries, and error handling instead of the team hand-writing imperative orchestration code. That declarative definition plus managed execution is what separates DLT from the scheduling, modelling, and visualization components around it. Declare the pipeline and its quality rules, and let DLT run them.

    Why the other options are wrong

    • ALakeflow Jobs schedules and orchestrates tasks, potentially including a DLT pipeline, but is not itself the declarative pipeline framework with built-in quality checks.
    • CMosaic AI is for building and serving machine learning models, not for declaring ETL pipeline transformations.
    • DAI/BI Dashboards visualize query results for end users; they do not build or manage ETL pipelines.

Who this Databricks Certified Data Analyst Associate practice exam is for

This practice set is for anyone preparing for the Databricks Certified Data Analyst Associate exam at the intermediate level - from first-time candidates building a foundation to experienced Data 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 Databricks Certified Data Analyst Associate 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 Databricks Certified Data Analyst Associate exam.

Related Data resources

• Databricks Certified Data Analyst Associate study guideKey concepts
• Data practice examsAll Data
• Certification pathWhere this fits
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• How these questions are written and reviewedMethodology
• Report a problem with a questionCorrections
• Databricks Certified Data Engineer Associate practice examRelated
• Databricks Certified Data Engineer Professional practice examRelated
• Databricks Certified Generative AI Engineer Associate practice examRelated

Databricks Certified Data Analyst Associate practice exam FAQ