Azure Data Fundamentals practice

{{ summaryValue }}

{{ summaryLine }}

Build a DP-900 practice round

Unofficial Microsoft Azure Data Fundamentals (DP-900) practice · English skills measured July 21, 2026 · verified September 11, 2026. Independently authored; not affiliated with or endorsed by Microsoft.

36 curated English questions cover selected concepts across all four domains. This is untimed single-choice practice, sampled uniformly without replacement. The real proctored assessment allows 45 minutes and may include interactive components. This bank does not reproduce that exam experience.

Official guide and format · Official preparation resources

Blueprint ranges: core data 25–30%; relational Azure 20–25%; non-relational Azure 15–20%; analytics 25–30%. These are scope references, not sampling weights. Concept checks and service scenarios are foundation-level practice, not calibrated exam difficulty or exhaustive coverage.

Optional. Blank uses a stable default; a custom seed changes question and option order. Up to 128 UTF-16 units; longer entries are shortened at a whole-character boundary. Do not enter personal information.

{{ seedNotice }}

{{ examSetupValidation }}

{{ answeredCount }} of {{ questionCount }} answered · {{ Math.round(progressPercent) }}%

{{ progressStatusLine }}

Question {{ examCursor + 1 }} of {{ questionCount }}

{{ currentQuestion.stem }}

{{ currentQuestion.accessible_clue }}

Answer recorded. Corrections appear when you view results.

Review this practice round

{{ resultLead }}

DP-900 · English skills July 21, 2026 · original bank 2026.09.11. This score describes only the items answered. It is not a Microsoft scaled score, pass prediction or credential. Practise with the official exam sandbox and hands-on data services as well. Microsoft and Azure are Microsoft trademarks.

Round seed: {{ examSession.seed }} · Session: {{ examSession.session_id }}

{{ readinessReviewTitle }}

{{ readinessReviewLine }}

{{ card.label }}
{{ card.value }}
{{ card.note }}

Missed answer review

  1. {{ item.prompt }}

    Your answer: {{ item.selected_answer_label }}

    Correct answer: {{ item.correct_answer_label }}

    {{ item.explanation }}

    {{ item.source.label.split(' — ')[0] }} · {{ item.source.version }}

Domain performance

Correct and missed answers by official domain for this round. Domains absent from the chart were not sampled. Use the missed-item sources to review that domain, then practise a focused set and check the full official guide.

Loading domain-performance chart…

{{ examChartStatus }}

{{ examChartExportStatus }}

Answer ledger

Question ID Prompt Topic Your answer Correct answer Outcome Explanation Source Copy
{{ row.question_id }} {{ row.prompt }} {{ row.category_label }} {{ row.selected_answer_label }} {{ row.selected_answer_id }} {{ row.correct_answer_label }} {{ row.correct_answer_id }} {{ row.outcome_label }} {{ row.explanation }} {{ row.source ? `${row.source.label.split(' — ')[0]} · ${row.source.version}` : 'Official guide' }}{{ sourceUrl(row.question_id) }}

{{ examLedgerExportStatus }}

Microsoft Azure Fundamentals Exam Prep Practise AZ-900 with 36 original questions and sourced explanations that help you review mistakes across the three exam domains.
Database Connection Pool Calculator Size database connection pools across replicas and workers while checking reserve-aware headroom, rollout surge risk and per-holder caps.
Database Storage Growth Calculator Forecast database storage runway from projected data growth, then include operational overhead and estimate threshold dates and monthly cost.
SQLite Manager Open a SQLite file in your browser and inspect schema or rows before running guarded SQL and saving deliberate edits as a new local copy.

Recording a ticket purchase and comparing three years of ticket sales place different demands on data. The purchase changes current records and must leave the remaining seat count correct. The comparison reads accumulated records to reveal patterns by month or event. Transactional processing supports the first task, while analytical processing supports the second.

The structure of the information matters too. A schema describes the fields and types a record should contain. Some records follow a fixed schema, others allow different attributes, and some content does not fit naturally into fields at all.

Representation Typical example Important distinction
Structured Readings recorded in fixed sensor, time and temperature columns. Records follow the same defined structure.
Semi-structured JavaScript Object Notation (JSON) product documents with different optional attributes. Named fields and nesting provide structure without requiring identical fields in every document.
Unstructured The sound content of a recorded support call. A table describing the recording does not turn the audio itself into tabular data.

Relational databases organize data into tables that can be connected through keys. Non-relational databases, often called NoSQL databases, use models such as documents, key-value records or graphs. These are storage-model distinctions, not synonyms for transactional and analytical workloads. A relational database can support reporting as well as operational transactions.

A data lake collects file-based data for analysis. A data warehouse brings data into organized structures for repeated reporting, while a lakehouse combines lake storage with table-based management and querying. The choice depends on how the data arrives, how much preparation it needs and how people will query it. Simply storing a file does not produce a business report or maintain a reliable transactional table.

Timing changes the processing choice. A nightly batch can summarize accumulated sales, but a system that must react to incoming sensor readings needs streaming processing rather than waiting for that batch. Streaming reduces the wait between arrival and processing. It does not make a stream-processing service interchangeable with the database or report that receives its output.

Administration is another consequential tradeoff. A managed database service can take responsibility for engine maintenance without giving the customer access to its host operating system. Running SQL Server on a virtual machine provides that access, together with responsibility for managing the machine. A requirement to install custom host software can therefore change the appropriate service even when both options use SQL Server.

Microsoft Azure Data Fundamentals, assessed through DP-900, introduces these concepts and their relationship to Azure data services. It is aimed at people beginning to work with cloud data. Understanding why a service fits a stated requirement is more useful than memorizing its name. Foundational knowledge still needs hands-on practice before it can support database deployment, recovery or production operation.

How to Use This Tool:

This independently authored English practice is not affiliated with or endorsed by Microsoft and does not reproduce its examination questions. The initial setup is eight questions from all domains with corrections after each answer.

  1. Choose a Domain / practice focus. Use all domains for a mixed round, one of the four domains for focused study, or the cross-domain sets of 21 data concept checks and 15 Azure and Microsoft service choices.
  2. Set the Question count. All-domain rounds offer 2, 5, 8 or 36 questions. Other sets offer 2, 5, 8 or their full pool, except non-relational data, which offers 2, 5 or 7. Changing focus replaces an unsupported count with that set's default.
  3. Choose Feedback timing. After each answer reveals the correction and explanation immediately. At the end of the round records each choice but withholds corrections until results.
  4. Enter a Replay seed under Advanced only when you need a particular repeatable round, then select Start exam practice. Blank uses a stable default, not a fresh random seed.
    The seed limit is 128 UTF-16 units. Ordinary Latin characters use one unit, while many emoji use two. Longer entries are shortened without splitting a Unicode code point. Clear or replace a seed that triggers an invalid-character warning, and do not enter personal information.
  5. Select one of the four answers. Selecting it submits and locks the choice, including when feedback is deferred. Continue with Next question, then View results after the final answer.
    End round clears the current attempt and returns to setup. It does not produce a score report for a partly answered round.
  6. Review the missed explanations and the Answer ledger, which compares your choices with the correct answers and their sources. Use Retry missed to revisit errors, or Change setup to study another domain.

Interpreting Results:

During a round, the headline counts questions answered. At completion, it changes to correct answers out of all questions in that round. A full progress bar means every answer was submitted, not that every answer was correct.

Domain performance shows correct and missed answer counts, not difficulty-adjusted ratings or official domain scores. A domain marked Not sampled in this round is untested, not failed. Read the number of questions behind each result before comparing domains, then use the missed-item explanations to choose what to study.

Replay same seed reinforces the same questions and option order. Reproduction depends on keeping the focus, count, feedback mode and question-bank version unchanged. Try a new seed resamples within the same pool, so familiar questions can recur. A full-pool round cannot gain unseen questions merely by changing its seed. Better repeat scores may reflect familiarity as well as improved understanding.

A Perfect round is not an exam-readiness rating. These rounds are untimed and their difficulty is not calibrated to DP-900. The official assessment is proctored, allows 45 minutes and may include interactive components. Microsoft reports technical-exam scores on a scale of 1 to 1,000, with 700 or greater required to pass. That is not a raw 70% threshold, and no conversion from this practice score is supported. Use the official preparation resources, exam sandbox and hands-on exercises alongside question review.

Technical Details:

A correct data-service answer must satisfy the requirement in the question, not merely name a product that can store data. Each practice question has one authored correct choice and three alternatives. Every correct selection earns one point, while any other offered choice earns zero.

Accuracy and completion progress

Accuracy counts correct answers against the entire round. Progress instead counts all submitted answers, whether correct or incorrect.

Points = C Accuracy (%) = 100×CN Progress (%) = 100×AN

C is the correct-answer count, A is the answered count and N is the positive total number of questions in the round. Possible points equal N, so scores range from 0 to N points and accuracy from 0% to 100%. There is no partial credit, negative marking or domain weighting.

Unanswered questions remain in the accuracy denominator but are not counted as incorrect. Completion requires every question to be answered. For example, three answers in an eight-question round give 37.5% progress, shown as 38% beside the progress bar. If two are correct, the practice accuracy calculation is 25%, with one incorrect answer and five unanswered questions. That is an unfinished attempt, not a completed result.

Raw accuracy and underlying progress are unrounded. Visible progress text rounds to the nearest whole percent, with half-percent ties upward; the bar width and accessible progress value retain the raw percentage. The result headline preserves the exact correct/total count. Completed-round review labels follow these rules rather than a pass threshold.

Review label Condition Meaning
Perfect round C = N No answers in this round need correction.
Start with the missed answers 0 ≤ C < N At least one submitted choice was incorrect and is included in missed-answer review.

Objective scope and question selection

The 36-question bank is dated September 11, 2026 and references the English DP-900 skills measured as of July 21, 2026. The skills-measured date identifies an objective edition, not an exam launch date or exact live question composition. The official percentages describe examination scope. The item counts below describe only this curated practice bank.

Domain Official share Bank items Selected practice topics
Core data concepts 25% to 30% 10 Data representations, file formats, transactions, workloads and data roles.
Relational data on Azure 20% to 25% 9 Keys, normalization, SQL, views, Azure SQL choices and PostgreSQL.
Non-relational data on Azure 15% to 20% 7 Object, file, entity and queue storage, documents and an existing Gremlin application.
Analytics workloads on Azure 25% to 30% 10 Data ingestion, warehouses, Fabric, Databricks, streaming and Power BI.

Questions are sampled uniformly without replacement within the selected pool, with seeded question and option order. There are no official domain quotas. A short mixed round may omit a domain entirely. Covering selected concepts in all four domains does not establish exhaustive objective coverage, and the English objective date should not be assumed to match every localized exam update.

Transactions, files and relational structure

ACID names four transaction properties: atomicity, consistency, isolation and durability. Atomicity requires the whole transaction to succeed or be rolled back. Consistency preserves valid data rules. Isolation governs interactions between concurrent transactions. Durability preserves committed changes. An all-or-nothing transfer therefore concerns atomicity, not simply whether a successful transfer remains stored.

File layout and transaction guarantees are different concerns. Parquet groups column values so analytical readers can retrieve selected columns efficiently. Avro is row-oriented. Delta Lake adds a transaction log over Parquet files to support reliable table changes and version history. A column-oriented file format alone is not the same thing as transactional table management.

Normalization reduces duplication by separating facts about different entities. A primary key identifies a row, and a foreign key links related records. Storing a customer's current address once and referring to the customer from orders avoids conflicting copies of that same address. A standard view serves a different purpose: it exposes a saved query as a tabular result rather than requiring another manually maintained copy of its rows.

Structured Query Language (SQL) distinguishes operations on data from operations on structure. SELECT retrieves rows, INSERT adds rows, UPDATE changes them and DELETE removes them. CREATE TABLE is data definition language because it defines a table and its columns. A question asking to read data must not be answered with a command that changes that data.

Matching services to requirements

The Azure SQL family shares the SQL Server engine, but its hosting choices expose different levels of control. The requirement for database-level features, instance-level compatibility or host access is more informative than the presence of the word SQL.

Requirement Matching option Boundary
A managed relational database without host access or instance-level requirements. Azure SQL Database Database-as-a-service, not an operating system to administer.
Broad SQL Server instance compatibility without managing the operating system. Azure SQL Managed Instance Managed instance features do not grant host operating-system access.
Custom software must be installed on the SQL Server host. SQL Server on Azure Virtual Machines Host control includes responsibility for the virtual machine.

Engine compatibility matters separately. Azure Database for PostgreSQL provides a managed PostgreSQL engine with automated patching and backups. It is not another SQL Server hosting option simply because both engines accept SQL.

Among Azure Storage services, Blob Storage holds objects such as media files, while Azure Files supplies managed shares, including Server Message Block (SMB) access. Table Storage holds non-relational key/attribute entities, despite its name. Queue Storage holds messages so a producer can hand work to a consumer for later processing. A mounted file share, a document database and a work queue solve different access requirements.

Azure Cosmos DB for NoSQL supports flexible JSON documents and global distribution for operational applications. Graph data instead describes vertices and the edges connecting them. The Gremlin question concerns that model in an existing Azure Cosmos DB for Apache Gremlin application, not a recommendation for a new graph deployment.

From incoming data to a report

Extract, transform, load (ETL) prepares data before loading it into the destination. Extract, load, transform (ELT) loads first and uses the destination's processing capabilities to transform it. The distinguishing point is where and when transformation happens, not whether the final result can be reported on.

Microsoft Fabric combines analytics capabilities around OneLake, its shared logical data lake. Azure Databricks supports lakehouse work and Apache Spark, a distributed data-processing engine. These platforms can support several stages of analysis, so a service scenario must be read for its stated requirements rather than reduced to a single keyword.

For incoming events, Azure Stream Analytics provides managed processing with SQL-like queries and time-based aggregation. Fabric eventstreams provide ingestion, transformation and routing to supported destinations, including no-code authoring. Neither an ordinary file share nor a report page supplies the stream-processing requirement just because it can participate in the wider solution.

A Power BI semantic model organizes data, relationships and calculations for reporting. In a star schema, fact tables contain observations such as sales at a defined grain, meaning what one row represents. Dimension tables describe products, dates and other entities used to filter or group those observations. Numeric facts and descriptive dimensions have different roles even when both appear in the same report.

Privacy and Saved Attempts:

Answer checking takes place in your browser, but the page address can contain an unencrypted saved attempt, including the seed and recorded choices. A working saved link can restore those choices for its recipient. The address can remain in browser history and be sent to the website when reopened or reloaded, so local scoring does not make the attempt private.

Page resources and chart or document features can use the network, and opening a reference visits the linked website. Downloaded answer ledgers also contain your responses. Do not put personal information in a replay seed or share an attempt address when its answers should remain private. Fully offline operation is not guaranteed.

A damaged saved link or one from an incompatible question-bank version may show Saved attempt unavailable and return to setup. Start a new round rather than treating that warning as a score. If saving an active attempt in the address fails, keep the page open to continue or return to setup with another seed. Clearing the current attempt does not remove previously copied links or browser-history entries.

Worked Examples:

One transaction mistake in a full-bank round

Choose all domains and 36 questions. Suppose every answer is correct except the transfer question, where you select durability instead of atomicity. The completed headline is 35/36, and core data shows nine correct answers and one miss. Retry missed creates a one-question round containing that error, even though the normal setup menu starts at two questions. Choosing atomicity then earns 1/1 for the retry. That result records correction of the specific mistake, not mastery of all data concepts.

Reading prices without changing them

In a relational practice round, the question asking for product names with prices greater than 100 is answered by SELECT ProductName FROM Products WHERE Price > 100;. For illustrative rows priced at 99, 100 and 125, that condition selects only the row priced at 125. Selecting the matching answer earns one point, with its explanation shown immediately or at completion according to feedback timing. An UPDATE answer changes data rather than reading it, and a less-than comparison selects the wrong rows. The practice checks your selected answer; it does not execute SQL against a database.

References: