Microsoft Certified: Azure Databricks Data Engineer Associate
Independent: not affiliated with Microsoft. Free to try; the study bundle adds a study guide and two practice exams.
5 original questions, none of them in the practice exams.
Pick an answer. Show answer explains every option and links its Microsoft source.
5 questions / Explanations link Microsoft sources / Last updated Oct 26, 2026
Question 1
Multiple choice
Domain 1 / Task 1.1 Select and configure compute in a workspace
A bank runs its payment settlement jobs on classic job compute. Its policy says that the runtime a production job uses may change only through stability and security fixes, that the runtime's end-of-support date must be published when the job is deployed, and that each new job should stay on a supported runtime for as long as possible. The compute form offers Databricks Runtime 19 (released June 2026, not marked LTS), 18 LTS, 17.3 LTS and 14.3 LTS, among others. Which runtime should the team select for the new jobs?
Show answerHide answer
Answer: C
Why
During a version's GA feature-development phase, which lasts about six months, Azure Databricks releases new features and fixes under the same version number. The version then moves to long-term support (LTS), in which major stability and security fixes are backported for three years. The supported-versions table marks the LTS versions and lists their end-of-support dates, while the date for Databricks Runtime 19 reads "Determined at LTS transition". Of the versions offered, 18 LTS meets the first two conditions and has the latest end-of-support date, June 10, 2029.
Why not the others
A17.3 LTS meets the first two conditions, but its support ends on October 22, 2028, before that of 18 LTS.
B14.3 LTS meets the first two conditions, but its support ends on February 1, 2027, the earliest end-of-support date in the table.
DDatabricks Runtime 19 is still in its feature-development phase, so new features ship under its version number, and its end-of-support date is not set yet.
Investigators on a bank's security team read the audit logs of the finance workspace, but the records they see do not include the text of the commands that ran. For future investigations they need the text of every notebook cell that a user or a job runs, and of every query sent to a SQL warehouse, in the audit records. What should a workspace admin do?
Show answerHide answer
Answer: D
Why
Verbose audit logs are additional records generated whenever a command or query runs in the workspace, and they record its text. They are off by default; a workspace admin turns them on next to Verbose Audit Logs on the Advanced tab of the admin settings page. The audit logs then include runCommand events from the notebook service (cells run by users) and from the jobs service (cells run by job runs), each with commandText, and commandSubmit events with commandText for commands sent to Databricks SQL. Turning the setting on or off is itself logged as a workspace event.
Why not the others
AThe databricks_pii_access group unmasks the SQL definitions that some audit events store (function_info, view_definition, definition_json, managed_definition); it cannot show command text that was never recorded.
BCompute log delivery copies the driver, worker and event logs of that compute to the volume; it adds nothing to the audit logs and does not cover queries sent to SQL warehouses.
CUnity Catalog captures lineage automatically, down to the column level, and lineage shows which tables and columns queries read and write, not the text of each command.
Domain 3 / Task 3.3 Cleanse, transform and load data
A data engineer receives bronze.web_events, a 42-column table with about 300 million rows, and must assess every column before writing the silver logic: how many distinct values each categorical column holds, which values occur most often, and the percentiles and distribution of the numeric columns. The engineer wants this from one command on a Spark DataFrame in a Python notebook. Which command should the engineer run?
Show answerHide answer
Answer: A
Why
summarize(df, precise) calculates and displays summary statistics of a Spark or pandas DataFrame, in Python, Scala and R. With precise at its default of false, the distinct-value counts for categorical columns, the frequent-value counts, the histograms and the percentile estimates are approximated to cut run time; with precise=True everything except the histograms and percentiles is exact. The command analyzes the complete DataFrame, so it can be expensive on very large tables. The data utility is in Public Preview.
Why not the others
Bdescribe() returns only count, mean, stddev, min and max per column; it has no distinct counts, frequent values, percentiles or histograms.
CANALYZE TABLE collects estimated statistics that the query optimizer uses to plan queries; it is not a profiling command and displays no summary of the data in the notebook.
DRenders 1,000 rows, so anything judged from them describes a sample rather than the table.
Domain 3 / Task 3.4 Implement and manage data quality constraints
A nightly notebook appends a DataFrame to the Delta table prod.silver.shipments with df.write.mode("append").saveAsTable("prod.silver.shipments"). The carrier has started sending a new field, carrier_code, and the DataFrame now carries that column, so the append fails because the table's schema does not include it. The team wants the column added to the table, the existing rows kept with carrier_code null, and no other write in the session affected. What should the data engineer do?
Show answerHide answer
Answer: B
Why
With schema evolution enabled on the write, columns present in the source but missing from the target table are added as part of the write transaction, at the end of the schema, and existing rows receive NULL for the new column. Databricks recommends enabling evolution per write operation, with the mergeSchema option or the WITH SCHEMA EVOLUTION syntax, rather than through a session-wide Spark configuration; an option set on the write takes precedence over the configuration.
Why not the others
AThe append succeeds, but the new field is discarded.
CIt would add the column, but it enables schema evolution for every write in the session, which Databricks does not recommend for production because it can cause unintended schema changes across operations; the team asked that other writes be unaffected.
DoverwriteSchema with overwrite mode replaces the table's schema and its data with the DataFrame, so the existing rows are lost.
A Lakeflow Job that builds prod.gold.daily_sales normally finishes in 20 minutes, but some runs stall for hours waiting on a lock in an upstream database. The team wants a message in its Microsoft Teams channel, which an administrator has already set up as a system notification destination, when a run passes 45 minutes, and wants any run that passes 3 hours to end automatically. Which two settings should the data engineer configure on the job? Choose two.
Show answerHide answer
Answers: C and D
Why
Under Duration and streaming backlog thresholds in the Job details pane, the Warning field for the Run duration metric sets the job's expected completion time; when a run exceeds it an event is triggered, and a notification with Duration warning selected sends it to an email address or a system destination such as Microsoft Teams. The Timeout field sets the maximum completion time: a run that does not complete in that time is set to Timed Out.
Why not the others
AStreaming backlog metrics measure how far a Structured Streaming source is behind, and a batch build waiting on a lock has no streaming backlog; a backlog threshold also only triggers a notification, so it would not end the run.
BRetries re-run a task after it fails; a run waiting on a lock has not failed, so nothing is retried, notified or stopped.
EA 45-minute timeout ends the run at 45 minutes and sets it to Timed Out; the team wants only a message at 45 minutes and the run ended at 3 hours.
The study bundle has two practice exams with 120 original questions, an answer key and an explanation for every answer, plus the 63-page study guide. Practice exam explanations are shorter than the ones on this page: a short paragraph per question, with no source links.
Study bundle
$19
120 practice questions / Study guide / Answer keys
Every bundle includes 6 months of access: download the PDFs, and any new editions, until 6 months after your purchase date. Your receipt has a license key. Enter it at overqual.com/download for the three PDFs, stamped with your email and order number. They are yours to keep.