Overview
Domain 1 is the vocabulary layer of DP-900. Microsoft groups it into four functional groups. The first asks you to tell structured, semi-structured and unstructured data apart. The second covers common file formats such as CSV, JSON, Parquet and Avro, the features of data stores including databases, and which Azure datastore suits a given use case.
The third group contrasts transactional workloads, which record many small changes with strong consistency, against analytical workloads, which read large volumes of history to answer business questions. The fourth group maps tasks to the database administrator, data engineer and data analyst roles. As of October 31, 2026, DP-900 is an active exam with no retirement or replacement announced, and the outline used here is skills measured as of July 21, 2026.
How it's tested
Expect short scenarios or statements that describe a dataset, a file, a workload or a job task, then ask you to classify it. Many items are yes/no statement sets or drag-and-drop matches, so precise definitions matter more than product configuration detail. Watch for one adjective, such as columnar or schema-on-read, that settles the answer.
-
Classify the data
A description of rows and columns, tagged JSON documents, or images and audio, and you name it structured, semi-structured or unstructured.
-
Name the workload
Many small inserts with ACID guarantees point to transactional processing, while large aggregations over history point to analytical processing.
-
Match the role
Backups and permissions belong to the database administrator, pipelines to the data engineer, and reports and models to the data analyst.
A question says a company stores sensor readings as JSON messages where each device sends different fields. Rows and columns with a fixed schema would make it structured, and raw video would make it unstructured. Because each message carries its own field names and nesting but no enforced table schema, the data is semi-structured. A store that accepts flexible documents, such as Azure Cosmos DB for NoSQL, follows naturally from that classification.
Functional groups 4
Official functional groups from the Microsoft outline, in outline order.
1.1 Describe ways to represent dataStructured, semi-structured and unstructured data and how each is organized.
- What it means
- This group covers describing the features of structured data, semi-structured data and unstructured data, and recognizing which category a given dataset belongs to.
- What to understand
- Structured data follows a fixed schema of tables, rows and columns, so every record has the same fields and types, as in a relational database. Semi-structured data carries its own organization through tags or key-value pairs but allows records to differ; JSON, XML and YAML are the usual examples, and nesting is common. Unstructured data has no internal model a database can query directly, such as images, video, audio, PDFs and free text. Classification depends on how the data is organized, not on where it is stored: a JSON file in Blob storage is still semi-structured.
- How it shows up
- Expect a dataset description and a choice of category, often with JSON or CSV as the deciding example.
- Common confusion
- Calling a CSV file unstructured because it is a plain text file; its consistent columns make it structured.
1.2 Identify options for data storageFile formats, data store features and Azure datastores for common use cases.
- What it means
- This group covers common formats for data files, the features of common data stores including databases, and identifying which Azure datastore fits a given use case.
- What to understand
- Delimited text such as CSV is human-readable and widely supported. JSON and XML hold hierarchical records. Avro is a row-based binary format with an embedded schema that suits streaming and write-heavy ingestion. Parquet and ORC are columnar binary formats that compress well and speed analytical reads of a few columns, and Delta Lake adds transaction logs on top of Parquet. Data stores include file stores, relational databases and non-relational databases such as key-value, document, column-family and graph stores. On Azure, relational needs map to Azure SQL or Azure Database for PostgreSQL or MySQL, flexible documents to Azure Cosmos DB, and files or blobs to Azure Storage and Data Lake Storage Gen2.
- How it shows up
- Expect a use case such as fast column scans, schema-flexible documents or bulk file storage, and you choose the format or Azure store.
- Common confusion
- Treating Avro and Parquet as interchangeable; Avro is row-oriented, Parquet is columnar.
1.3 Describe common data workloadsTransactional and analytical workloads and their characteristics.
- What it means
- This group covers describing the features of transactional workloads and the features of analytical workloads.
- What to understand
- Transactional systems, also called online transaction processing or OLTP, capture business events such as orders and payments as they happen. They rely on ACID properties: atomicity, consistency, isolation and durability. Data is usually normalized to avoid update anomalies, and operations touch few rows quickly. Analytical workloads read large volumes of historical data, often copied from operational systems through extract, transform and load or extract, load and transform processes into a data warehouse, lakehouse or data lake. Their schemas are often denormalized into fact and dimension tables to make aggregation fast, and they favor read performance over write performance.
- How it shows up
- Expect workload descriptions with words like insert, update, concurrency, aggregate or trend, and you classify them as transactional or analytical.
- Common confusion
- Assuming analytical systems must be real time; many run on batch loads refreshed daily or hourly.
1.4 Identify roles and responsibilities for data workloadsDatabase administrator, data engineer and data analyst responsibilities.
- What it means
- This group covers describing the responsibilities of database administrators, data engineers and data analysts in data workloads.
- What to understand
- A database administrator keeps operational databases available, secure and performant: installing or provisioning, patching, backup and restore, high availability, user permissions and monitoring. A data engineer builds and maintains the pipelines and stores that move and shape data, including ingestion, cleansing, transformation and lake or warehouse design. A data analyst explores and models prepared data to find insights, building semantic models, reports and dashboards, commonly in Power BI. Overlap exists in practice, but the exam expects the primary owner of each task.
- How it shows up
- Expect a list of tasks to drag to the correct role, or a statement about who is responsible for a given activity.
- Common confusion
- Assigning pipeline building to the data analyst or backup configuration to the data engineer.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Structured data
Data with a fixed tabular schema.
- Remember
- Every row has the same columns and types.
- Tested as
- Classification questions.
- Trap
- Calling CSV unstructured.
-
Semi-structured data
Self-describing data with tags or keys.
- Remember
- JSON, XML and YAML; records can vary.
- Tested as
- Classification and Cosmos DB questions.
- Trap
- Thinking it needs a predefined schema.
-
Unstructured data
Data without an internal queryable model.
- Remember
- Images, audio, video and documents.
- Tested as
- Storage choice questions.
- Trap
- Assuming it cannot be stored in Azure at all.
-
Parquet
Open columnar file format.
- Remember
- Strong compression and fast reads of selected columns.
- Tested as
- File format questions.
- Trap
- Labeling it row-based.
-
Avro
Row-based binary format with schema stored in the file.
- Remember
- Good for serialization and write-heavy ingestion.
- Tested as
- File format questions.
- Trap
- Choosing it for column-scan analytics.
-
ACID
Atomicity, consistency, isolation, durability.
- Remember
- Defines reliable transactions in OLTP systems.
- Tested as
- Transactional workload questions.
- Trap
- Thinking only relational engines can offer it.
-
OLTP
Online transaction processing.
- Remember
- Many small, concurrent reads and writes on current data.
- Tested as
- Workload classification.
- Trap
- Using it for heavy historical reporting.
-
Analytical workload
Read-heavy processing of large historical datasets.
- Remember
- Often denormalized into fact and dimension tables.
- Tested as
- Workload classification.
- Trap
- Assuming it is always streaming.
-
ETL and ELT
Extract, transform, load versus extract, load, transform.
- Remember
- ELT transforms after loading into the target store.
- Tested as
- Ingestion questions.
- Trap
- Treating the order of steps as irrelevant.
-
Database administrator
Owner of database operations.
- Remember
- Backups, security, availability and performance.
- Tested as
- Role questions.
- Trap
- Giving them report design.
-
Data engineer
Owner of data pipelines and stores.
- Remember
- Ingestion, transformation and lake or warehouse design.
- Tested as
- Role questions.
- Trap
- Giving them dashboard creation.
-
Data analyst
Owner of insight and reporting.
- Remember
- Models, visualizations and dashboards, often in Power BI.
- Tested as
- Role questions.
- Trap
- Giving them server patching.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Data shapes
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Structured | Fixed schema of rows and columns | Schema enforced before data is written | Assuming CSV is unstructured |
| Semi-structured | Tags or keys describe each record | Fields can vary between records | Expecting a fixed table schema |
| Unstructured | No internal data model | Requires extraction to query content | Calling a JSON file unstructured |
Common file formats
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| CSV | Delimited plain text | Readable and simple, no types or compression | Using it for large analytical scans |
| JSON | Hierarchical text documents | Supports nesting and varied fields | Assuming it is columnar |
| Avro | Row-based binary with schema | Efficient record writes and schema evolution | Choosing it for column-pruning reads |
| Parquet | Columnar binary | Compression and fast selective column reads | Treating it as human-readable |
Workload types
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Transactional | Records live business events | Small, frequent, ACID writes on normalized tables | Running big trend reports against it |
| Analytical | Answers questions over history | Large reads over denormalized or columnar data | Expecting row-by-row updates to be fast |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Text means unstructured
- What candidates pick
- A CSV export is unstructured data.
- Why it looks right
- It is just a text file.
- Why it's wrong
- Consistent delimited columns give it a tabular schema, so it is structured.
Clue to watch for The file has a header row and the same fields in every line.
-
JSON as structured
- What candidates pick
- JSON documents are structured because they have field names.
- Why it looks right
- Field names look like column names.
- Why it's wrong
- JSON records can differ and nest, so the category is semi-structured.
Clue to watch for Records described as having different or optional fields.
-
Avro for analytics
- What candidates pick
- Store data as Avro to speed queries that read two columns.
- Why it looks right
- Avro is a compact binary format.
- Why it's wrong
- Avro stores whole rows together; columnar Parquet lets engines skip unneeded columns.
Clue to watch for The requirement mentions reading a few columns across many rows.
-
Analyst builds pipelines
- What candidates pick
- The data analyst ingests and cleans raw data from source systems.
- Why it looks right
- Analysts do some data preparation in Power BI.
- Why it's wrong
- Building and operating ingestion pipelines is the data engineer's primary responsibility.
Clue to watch for Words like pipeline, ingestion or data lake design.
-
OLTP for reporting
- What candidates pick
- Run monthly trend reports directly on the order processing database.
- Why it looks right
- The data is already there.
- Why it's wrong
- Heavy analytical scans compete with transactions; a separate analytical store is the expected design.
Clue to watch for Large aggregations over years of history.
-
Normalization everywhere
- What candidates pick
- Analytical models should be fully normalized.
- Why it looks right
- Normalization sounds like good practice.
- Why it's wrong
- Analytical stores usually denormalize into star schemas to cut joins and speed aggregation.
Clue to watch for A reporting or data warehouse scenario.
Answer framework
Use these steps for any Domain 1 item that asks you to classify data, a format, a workload or a role.
- Look at how the data is organized: a fixed set of columns, self-describing keys or tags, or no internal model at all.
- For file formats, decide whether the need is human readability, row-by-row writes or fast reads of a few columns, then match CSV or JSON, Avro, or Parquet.
- For workloads, check whether operations are many small current-state changes or large reads over accumulated history.
- For stores, pair relational needs with Azure SQL or open-source databases, flexible documents with Cosmos DB, and files with Azure Storage.
- For roles, ask whether the task keeps a database running, moves and shapes data, or turns prepared data into reports.
Practice question
Want more like this? The DP-900 practice test has questions for every domain.
Checklist
0 of 13
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Start with definitions, because this domain rewards precision. Write one line for each data shape, each file format and each workload type, then test yourself with random examples: a tweet, a PDF invoice, a Parquet file, a SQL table, a JSON order. Say the category aloud and the reason. If you hesitate, the definition is not yet sharp enough.
Next, open the Microsoft Learn path for core data concepts and complete its knowledge checks, paying attention to how Microsoft words each feature. Create a small JSON file and a CSV with the same records to see why one is tabular and the other can nest. For roles, make a three-column table and sort about twenty tasks, such as configuring backups, building a pipeline, defining a measure or granting permissions, into the right owner. Finish by linking each concept forward: semi-structured data leads to Domain 3, OLTP to Domain 2, and analytical workloads to Domain 4. Two short sessions are usually enough here.
FAQ
How much of DP-900 is Domain 1, Describe core data concepts?
Microsoft weights Describe core data concepts at 25–30% of the exam, the joint largest share with the analytics domain, across four functional groups.
Is DP-900 being retired or replaced?
As of October 31, 2026 Microsoft shows no retirement date or successor for DP-900. The skills outline was refreshed on July 21, 2026 with minor wording changes only.
Is a CSV file structured or unstructured?
Structured. Its rows share the same delimited columns, so it follows a fixed tabular schema even though it is plain text.
Why is Parquet preferred for analytics?
It stores values by column and compresses them, so engines can read only the columns a query needs and skip the rest.
What is the difference between OLTP and analytical workloads?
OLTP records many small, concurrent business transactions with ACID guarantees. Analytical workloads read large amounts of historical data to find trends and totals.
Which role is responsible for building data pipelines?
The data engineer. Database administrators run and protect databases, and data analysts build models and reports from prepared data.
Do I need hands-on Azure skills for this domain?
Not much. It tests concepts and classification, so clear definitions and examples matter more than portal configuration.
Other Data & Analytics certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 1 knowledge
You have reviewed data shapes, file formats, data stores, workload types and data roles. Check whether you can classify each one quickly and correctly.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the DP-900 AI tutor Get any concept explained your way