Prepare Data (46% of Exam DP-600)
This domain covers the full data preparation lifecycle inside Microsoft Fabric: connecting to sources, selecting the right data store, shaping and cleaning data, and querying it with SQL, KQL, DAX, or visual tools. At 46% of the exam, it is the heaviest single domain. Expect scenario-based questions where you must choose the correct Fabric artifact, transformation approach, or query syntax given a specific business requirement and set of constraints.
Data Connections; OneLake Catalog and Real-Time Hub Discovery
Data Connections
Fabric provides data connectivity through Dataflows Gen2, Data Pipelines, Direct Lake mode, shortcuts, and mirroring. Understanding which mechanism applies to which scenario is critical.
- Shortcuts create virtual links in OneLake to external storage (ADLS Gen2, AWS S3, Google Cloud Storage, Dataverse, S3-compatible) without physically copying data. They appear as folders in the lakehouse Files or Tables section.
- Mirroring continuously replicates an entire external database (Azure SQL DB, Cosmos DB, Snowflake, Azure SQL Managed Instance) into OneLake as Delta tables, with near-real-time latency.
- Dataflows Gen2 use Power Query Online for low-code transformation; output is staged to a lakehouse or warehouse.
- Data Pipelines orchestrate movement and transformation at scale (Copy activity, Notebook activity, Dataflow activity).
OneLake Catalog
The OneLake catalog (formerly OneLake data hub) is the central discovery surface inside Fabric. It lists every Fabric item — lakehouses, warehouses, eventhouses, semantic models, reports — across all workspaces the user has access to. Key behaviors:
- Users can endorse items as Promoted (workspace admins and members) or Certified (tenant-level certification admins).
- The catalog exposes lineage, sensitivity labels, and item metadata.
- Filtering by item type, workspace, or endorsement status is built-in.
Real-Time Hub
The Real-Time hub is the discovery and connection surface specifically for streaming and real-time data sources. It centralizes:
- Eventstreams already created in the tenant.
- Microsoft sources (Fabric system events, Azure Event Hubs, IoT Hub).
- Fabric events (OneLake file events, workspace item events).
- External streaming sources connectable via Eventstream.
Common traps
- Confusing OneLake catalog (all Fabric items, batch and streaming) with Real-Time hub (streaming sources only).
- Assuming shortcuts copy data — they do not; they reference external storage in place.
- Mirroring vs. shortcuts: mirroring maintains a replicated, always-updated Delta copy; shortcuts leave data in the origin location.
- Dataflows Gen2 require a staging lakehouse by default when loading to a warehouse destination.
| Mechanism | Copies Data? | Latency | Typical Use |
|---|---|---|---|
| Shortcut | No | Instant (virtual) | Reuse existing ADLS/S3 data in Fabric |
| Mirroring | Yes (to OneLake) | Near-real-time | Keep Delta copy of operational DB |
| Dataflow Gen2 | Yes | Scheduled/on-demand | ETL with Power Query transforms |
| Data Pipeline Copy | Yes | On-demand/scheduled | Large-scale bulk movement |
| Eventstream | Yes (streaming) | Sub-second | Real-time ingestion to KQL/lakehouse |
Choosing Between Data Stores: Lakehouse, Warehouse, Eventhouse
This is one of the most heavily tested decision points on the exam.
Decision criteria at a glance:
| Dimension | Lakehouse | Warehouse | Eventhouse |
|---|---|---|---|
| Primary workload | Analytics on files + Delta tables | Enterprise SQL analytics | Real-time / time-series / streaming |
| Query language | Spark (PySpark/SparkSQL), SQL endpoint | T-SQL only | KQL (primary), some T-SQL |
| Data format | Delta Parquet (Tables), any file (Files) | Delta Parquet (managed by Warehouse) | Parquet columns in KQL Database |
| Schema enforcement | Optional (schema-on-read for Files) | Enforced (DDL required) | Semi-structured; dynamic schema |
| Transactions | Delta ACID on Tables | Full T-SQL ACID | Append-optimized, limited updates |
| Best for | Data engineering, ML, mixed workloads | BI reporting, governed SQL access | IoT, logs, telemetry, time-series |
| OneLake exposure | Native — Tables folder is Delta | Exposes tables via OneLake shortcut | Optional OneLake availability (toggle) |
| Compute | Spark pools (auto-scale) | Serverless SQL (DWUs auto-managed) | KQL engine (auto-scale) |
Worked scenario: A team receives IoT sensor data at 50,000 events/second and must query it with time-series operators like summarize ... by bin(Timestamp, 1m) and sliding-window aggregations. → Choose Eventhouse (KQL database). bin() and time-series operators are KQL constructs; the Eventhouse is purpose-built for this pattern.
Common traps
- Choosing Warehouse when the requirement mentions Spark notebooks or PySpark — Warehouse has no Spark engine.
- Choosing Lakehouse SQL endpoint for strict T-SQL DDL operations (CREATE TABLE with constraints) — the SQL endpoint is read-only; DDL runs only in the Warehouse or via Spark in the lakehouse.
- Eventhouse does not replace a warehouse for complex multi-table star-schema reporting.
- "Low-latency streaming analytics" always signals Eventhouse over Lakehouse or Warehouse.
Decision tree for store selection:
Diagram 1 — Choosing the right Fabric data store
Unlock with a paid plan
9 more sections (about 2,579 words) continue below. Every paid plan unlocks all DP-600 study guides, the full question bank and mock exams, from $4.97/month.