Read this the night before or on exam morning. It assumes you’ve studied the notes. It’s here to refresh, not to teach.
Exam day
- Microsoft publishes no fixed question count. Plan on about 45 minutes of exam time; check your appointment for the exact allowance. Scaled score out of 1,000; you need 700.
- No penalty for wrong answers, so answer everything. Some question types (drag and drop, hot area, case-style) don’t allow going back, so read carefully.
- Domain weights: core data concepts 25-30%, relational data 20-25%, non-relational data 15-20%, analytics workloads 25-30%.
- It is a fundamentals exam: definitions, service selection and scenario matching. No coding.
How to read a DP-900 question
| The question says | Think |
|---|---|
| Fixed rows and columns | Structured, relational |
| JSON, flexible fields | Semi-structured, document store |
| Images, video, PDFs | Unstructured, Blob |
| Short transactions, all or nothing | OLTP, ACID |
| Aggregate years of data | OLAP, warehouse |
| Least administration | PaaS (Azure SQL Database) |
| Migrate SQL Server with minimal changes | SQL Managed Instance |
| Control over the OS | SQL Server on a VM |
| Global, low latency, schema-free | Cosmos DB |
| Seconds of latency or less | Streaming |
Core facts
| Topic | Facts |
|---|---|
| ACID | Atomicity, Consistency, Isolation, Durability |
| Columnar formats | Parquet, ORC (Delta = Parquet plus transaction log) |
| Row formats | CSV, Avro, JSON |
| Roles | DBA = run and secure; data engineer = pipelines; analyst = insight |
| Fact vs dimension | Fact = measurements; dimension = descriptive context |
SQL one-liners
- DDL: CREATE, ALTER, DROP. DML: SELECT, INSERT, UPDATE, DELETE. DCL: GRANT, REVOKE, DENY.
- WHERE filters rows; HAVING filters groups. INNER JOIN = matches only; LEFT JOIN = all left rows.
- View = saved query. Index = faster lookups. Stored procedure = reusable logic. Primary key = unique id; foreign key = link.
- Azure SQL speaks T-SQL.
Azure SQL options
| SQL Database | Managed Instance | SQL Server on VM | |
|---|---|---|---|
| Model | PaaS | PaaS | IaaS |
| Compatibility | High | Near 100% | 100% |
| OS access | No | No | Yes |
| Best for | New cloud apps | Lift and shift | Full control |
Extras: elastic pool (shared resources), serverless (auto-pause), Hyperscale (huge databases), vCore vs DTU purchasing. Open source: Azure Database for PostgreSQL and MySQL.
Storage and Cosmos DB
| Item | Key facts |
|---|---|
| Blob tiers | Hot, Cool (30 days), Cold (90 days), Archive (180 days, offline) |
| Blob types | Block (default), append (logs), page (VM disks) |
| ADLS Gen2 | Blob + hierarchical namespace |
| Azure Files | SMB/NFS shares, File Sync |
| Table storage | PartitionKey + RowKey, schemaless, cheap |
| Redundancy | LRS (1 datacenter), ZRS (zones), GRS (region pair), GZRS (both), RA- = readable secondary |
| Cosmos DB | RU/s, partition key, automatic indexing, multi-region writes |
| Consistency | Strong, Bounded staleness, Session (default), Consistent prefix, Eventual |
Cosmos APIs: NoSQL (JSON documents), MongoDB, Cassandra (wide column), Gremlin (graph), Table (key-value). Chosen at creation, cannot be changed.
Analytics
| Need | Service |
|---|---|
| Unified SaaS analytics | Microsoft Fabric (OneLake, Lakehouse, Warehouse) |
| Spark and ML | Azure Databricks |
| Orchestration, ETL/ELT | Data Factory / Fabric pipelines |
| Older unified platform | Azure Synapse Analytics |
| Event ingestion | Event Hubs (IoT Hub for devices) |
| Stream queries | Stream Analytics, Fabric Eventstreams |
| Telemetry store, KQL | Eventhouse / Azure Data Explorer |
- ETL transforms before loading; ELT loads then transforms. Medallion: bronze, silver, gold.
- Batch = large sets, higher latency; streaming = continuous, low latency.
Power BI
- Desktop (author) → service (publish, share, refresh). Report = pages of visuals; dashboard = pinned tiles.
- Model: tables, relationships, star schema, DAX measures, hierarchies, row-level security.
- Visuals: bar/column compare, line trend, pie parts of whole, scatter relationship, card single number, map geography, matrix detail, slicer filter.
Last checks
- Read the last sentence first, then constraint words.
- Remove answers that don’t fit the data model (for example, Gremlin for key-value).
- Don’t overthink: fundamentals questions test the most standard choice.