5.5 Microsoft Fabric
Key Takeaways
- Microsoft Fabric went GA on November 15, 2023, and unifies Synapse, Data Factory, Power BI, and Azure Data Explorer into one SaaS platform billed by Fabric capacity (F-SKUs).
- OneLake is the single tenant-wide data lake; every Fabric item stores data as Delta-Parquet, and Shortcuts query ADLS Gen2, S3, GCS, or Dataverse in place without copying.
- Fabric exposes seven experiences: Data Factory, Data Engineering, Data Warehouse, Data Science, Real-Time Intelligence, Power BI, and Industry Solutions.
- Lakehouse is Spark-first with a read-only SQL endpoint; Warehouse is SQL-first with full multi-table T-SQL transactions. Both write Delta-Parquet to OneLake.
- Direct Lake is a Fabric-only Power BI storage mode that loads OneLake Delta files into VertiPaq on demand, combining DirectQuery freshness with Import speed.
Microsoft Fabric
Microsoft Fabric is Microsoft's unified, SaaS analytics platform. It went generally available on November 15, 2023, and is now the strategic direction for every analytics workload that previously required stitching together Synapse, Data Factory, Azure Data Explorer, and Power BI. For DP-900 you need to recognize what Fabric is, how OneLake works, and which experience matches each scenario.
Quick Answer: Fabric is a single SaaS product that ships seven analytics experiences sharing one tenant-wide data lake (OneLake), one open storage format (Delta-Parquet), and one billing model (Fabric capacity, F-SKUs). Power BI Premium customers get Fabric included; everyone else can buy a Fabric capacity directly.
How Fabric Differs From Azure-Resource Analytics
Azure Synapse Analytics, ADF, and Azure Data Explorer are Azure resources — you create them inside an Azure subscription and resource group. Fabric is a SaaS service in the Microsoft 365 / Power BI tenant: it lives at app.fabric.microsoft.com, and you carve up a Fabric capacity into workspaces instead of resource groups.
This shift has real consequences:
- No infrastructure management. No clusters to size, no networking to wire, no patching.
- One copy of data. Every experience reads and writes the same OneLake files.
- One identity and governance. Microsoft Entra ID, sensitivity labels, and Purview integration apply across all experiences.
- One bill. Compute is unified under capacity units (CUs) measured against the F-SKU you purchase, with the ability to pause the capacity.
OneLake: The Heart of Fabric
OneLake is automatically provisioned with every Fabric tenant. It is conceptually similar to OneDrive for analytics data — one logical lake per tenant, organized into workspaces and items.
- All Fabric items store data as Delta-Parquet by default — Delta Lake on top of Parquet files.
- Shortcuts let a OneLake folder transparently reference data in another OneLake workspace, Azure Data Lake Storage Gen2, Amazon S3, Google Cloud Storage, or Dataverse. Data is queried in place, not copied.
- Mirroring continuously replicates data from operational sources (Azure SQL Database, Cosmos DB, Snowflake) into OneLake in near real-time, with no ETL.
The Seven Fabric Experiences
| Experience | Persona | What it does |
|---|---|---|
| Data Factory | Data engineer | Ingest and orchestrate data with pipelines and dataflows |
| Data Engineering | Data engineer | Build Lakehouses, run Spark notebooks, manage Delta tables |
| Data Warehouse | SQL developer | Author T-SQL Warehouses with multi-table transactions |
| Data Science | Data scientist | Train and manage ML models with notebooks and AutoML |
| Real-Time Intelligence | Analyst / SRE | Ingest streams, store in Eventhouse / KQL Database, query with KQL |
| Power BI | Business analyst | Build datasets, reports, dashboards, paginated reports |
| Industry Solutions | Vertical teams | Pre-built data solutions for healthcare, sustainability, retail |
A workspace can contain items from any experience, all backed by OneLake.
Lakehouse vs Warehouse Inside Fabric
Fabric ships two SQL-capable items that often confuse beginners:
- Lakehouse — Spark-first. Created in the Data Engineering experience. Stores Delta tables, supports notebook-driven ELT, and exposes a SQL analytics endpoint that is read-only for T-SQL queries.
- Warehouse — SQL-first. Created in the Data Warehouse experience. Stores Delta tables too, but supports full multi-table T-SQL transactions (INSERT, UPDATE, DELETE, MERGE) and is the right home for traditional star-schema dimensional models.
Both write Delta-Parquet to OneLake, so a single Power BI semantic model can join data from a lakehouse and a warehouse without copying.
Direct Lake: Power BI Without Import
Direct Lake is a Power BI storage mode that is unique to Fabric. It loads Delta-Parquet files from OneLake directly into the VertiPaq engine on demand, without an Import refresh and without the round-trip latency of DirectQuery.
| Storage mode | Where data lives during query | Available in |
|---|---|---|
| Import | In-memory copy in the Power BI service | Power BI Pro / Premium / Fabric |
| DirectQuery | Queried live in the source system | Power BI Pro / Premium / Fabric |
| Composite | Mix of Import and DirectQuery tables | Power BI Pro / Premium / Fabric |
| Direct Lake | Read directly from OneLake Delta files | Fabric only |
Direct Lake gives Power BI dashboards the freshness of DirectQuery with the speed of Import, because OneLake is already optimized columnar storage.
Medallion Architecture in Fabric
The bronze-silver-gold medallion pattern fits Fabric naturally:
- Bronze Lakehouse — Land raw files from Data Factory pipelines, Eventstreams, or mirroring.
- Silver Lakehouse — Cleansed and conformed Delta tables, built with Spark notebooks.
- Gold Warehouse (or Gold Lakehouse) — Business-curated star schema. A Power BI semantic model in Direct Lake mode then reads gold tables for sub-second BI.
Capacities and SKUs
Compute is purchased as a Fabric capacity identified by an F-SKU such as F2, F8, F64, F2048. Capacity is measured in Capacity Units (CUs) and is shared across every experience in workspaces assigned to that capacity. F64 and above include free Power BI Pro licensing for all users in the tenant. Capacities can be paused to halt billing.
Governance and Security
Fabric inherits identity from Microsoft Entra ID. Workspace roles (Admin, Member, Contributor, Viewer) gate access at the container level, and item-level permissions narrow it further. Sensitivity labels, audit logs, and Microsoft Purview integration provide enterprise governance. Row-level security (RLS) and object-level security (OLS) work on warehouses, lakehouses, and Power BI semantic models alike.
An analyst needs to build a Power BI report that always reflects the latest data in OneLake without scheduled refreshes, but with near-Import query performance. Which Power BI storage mode should they use?
A team needs to point Microsoft Fabric at an existing data lake in Amazon S3 so Fabric notebooks can query the files without copying them into OneLake. Which Fabric feature meets this requirement?