Platform foundations¶
Start with a platform model that makes ownership, workload boundaries, capacity, and lifecycle decisions visible. Fabric unifies analytics experiences, but an enterprise design still needs deliberate workspace, capacity, data, and operating boundaries.
flowchart TB
T[Fabric tenant and governance model] --> C[Capacity plan]
C --> W[Workspaces and domains]
W --> D[OneLake data products and workload items]
D --> U[Consumers, operations, and stewardship]
Foundation decisions¶
| Decision | Questions to resolve |
|---|---|
| Capacity | Which workloads share capacity, which need isolation, and how will peak demand and throttling be managed? |
| Workspace model | How are development, test, production, team, and domain responsibilities separated? |
| Data architecture | Which Lakehouse, Warehouse, Real-Time Intelligence, semantic-model, or BI patterns suit each workload? |
| Lifecycle | Who owns an item, how is it changed, and when is it retired or archived? |
| Connectivity | Which approved cloud connections, gateways, identities, and network controls are required? |
Workload-aware design¶
The repository includes practices for Data Factory, Data Engineering, Data Warehouse, Data Science, Real-Time Intelligence, and Power BI. Use the workload's data volume, latency, transformation, consumption, and operating requirements to select patterns.
Establish a repeatable baseline¶
- Define platform owners, data-product owners, workspace administrators, and support escalation paths.
- Map the target environments, capacities, workspaces, and approved connections before teams begin building.
- Use infrastructure-as-code where supported to make prerequisite resources and policies reviewable and repeatable.
- Document naming, tagging, ownership, data classification, lifecycle, and recovery expectations.
- Pilot a representative workload before standardizing the design broadly.
Tip
A Medallion architecture is a pattern, not a mandate. Apply it when it helps make ingestion, refinement, quality, and serving responsibilities clear; adapt it to the actual workload and data-product contract.