Skip to content

Source: examples/fabric-e2e/README.md — this page is rendered live from that file.

Fabric End-to-End — Retail Sales Lakehouse + Direct Lake Semantic Model

Status: Real, deployable end-to-end Microsoft Fabric example. Workspace + Lakehouse + OneLake shortcuts to ADLS gold + Direct Lake semantic model (Power BI Project / TMDL format) + dbt medallion build-out + 4 data contracts + sample CSV data + ARCHITECTURE.md. Deploy in ~25 minutes if you have a Fabric capacity.

A turnkey reference for: how does a real Fabric workload look on disk and in IaC?

What's in this folder

fabric-e2e/
├── README.md                      # this file
├── ARCHITECTURE.md                # mermaid diagrams + design rationale
├── deploy/
│   ├── bicep/
│   │   └── main.bicep             # Fabric capacity (F2/F4/F8 by env)
│   ├── fabric/
│   │   ├── deploy.sh              # az rest calls to provision workspace,
│   │   │                          # lakehouse, shortcut, and import semantic model
│   │   └── workspace.json         # workspace metadata
│   └── README.md                  # deployment walkthrough
├── semantic-model/                # POWER BI PROJECT (PBIP) — TMDL format
│   └── retail-sales.SemanticModel/
│       ├── definition.pbism
│       ├── diagramLayout.json
│       └── definition/
│           ├── model.tmdl
│           ├── relationships.tmdl
│           ├── tables/
│           │   ├── DimCustomer.tmdl
│           │   ├── DimProduct.tmdl
│           │   ├── DimDate.tmdl
│           │   └── FactSales.tmdl
│           └── cultures/
│               └── en-US.tmdl
├── dbt/                            # medallion transformation project
│   ├── dbt_project.yml
│   ├── profiles.yml
│   ├── models/
│   │   ├── bronze/
│   │   │   ├── _sources.yml
│   │   │   └── bronze_*.sql
│   │   ├── silver/
│   │   │   └── silver_*.sql
│   │   └── gold/
│   │       ├── _gold_schema.yml
│   │       ├── dim_customer.sql
│   │       ├── dim_product.sql
│   │       ├── dim_date.sql
│   │       └── fact_sales.sql
│   └── tests/
├── sample_data/                   # tiny synthetic data to bootstrap
│   ├── customers.csv
│   ├── products.csv
│   └── sales.csv
├── contracts/                     # data contracts for gold star schema
│   ├── dim_customer.yaml
│   ├── dim_product.yaml
│   ├── dim_date.yaml
│   └── fact_sales.yaml
└── notebooks/
    └── load_sample_data.ipynb    # Spark notebook: CSV → bronze Delta

What this example shows you

✅ Real Fabric workspace provisioned end-to-end with Bicep (capacity) + REST (workspace/items) ✅ OneLake shortcut from a Fabric Lakehouse to ADLS gold (multi-cloud / hybrid pattern) ✅ Real TMDL semantic model in PBIP format — version-controlled, diffable, edit in VS Code or Power BI Desktop ✅ Direct Lake mode — import-tier perf with no refresh job ✅ Star schema in gold — done correctly (one fact + 3 dims, surrogate keys, role-playing date) ✅ dbt medallion build-out for bronze → silver → gold transformations ✅ Data contracts for every gold table (schema + GE rules + SLA) ✅ Sample synthetic data — 1000 customers, 500 products, 50K sales rows you can bring up in 5 minutes

Prerequisites

  • Azure subscription with a Fabric capacity already created (F2 minimum for dev — see Bicep below to provision one)
  • Service principal with Fabric Administrator role (for REST calls during deploy)
  • ADLS Gen2 storage account with sample data available at abfss://gold@<storage>.dfs.core.windows.net/retail-sales/
  • Azure CLI ≥ 2.65, jq, dbt-fabric ≥ 1.7
  • Power BI Desktop (Apr 2024+) if you want to edit the semantic model graphically

5-minute quickstart (POC)

# 1. Deploy the Fabric capacity (skip if you already have one)
cd deploy/bicep
az group create -n rg-fabric-e2e-dev -l eastus2
az deployment group create -g rg-fabric-e2e-dev \
  --template-file main.bicep \
  --parameters env=dev capacityAdminUpn=alice@contoso.com

# 2. Provision the workspace + lakehouse + shortcut + semantic model
cd ../fabric
./deploy.sh dev <capacity-name> <storage-account> <storage-key>

# 3. Run dbt to build the medallion
cd ../../dbt
dbt deps
dbt seed       # uploads sample CSVs to bronze
dbt run        # bronze → silver → gold
dbt test       # data quality

# 4. Open Fabric portal → workspace `csa-retail-sales-dev` → lakehouse → see gold tables
# 5. Open the semantic model → it's already wired to gold via Direct Lake
# 6. Open the included sample report (or build your own) → instant Direct Lake queries

What it deploys (Azure resources)

Resource SKU Why
Fabric Capacity F2 (dev), F8 (test), F64 (prod) Hosts the workspace and Direct Lake compute
Workspace one per env Logical container for Fabric items
Lakehouse one OneLake-backed Delta storage + SQL endpoint
OneLake Shortcut from lakehouse to ADLS gold Zero-copy access to gold from external storage
Semantic Model (Direct Lake) imported from semantic-model/ PBIP Power BI consumption
(Optional) Report imported from power-bi/reports/ Pre-built sample report

Architecture

See ARCHITECTURE.md for full mermaid diagrams (data flow, identity, deployment topology).

What this example deliberately does NOT include

  • ❌ Real-time streaming (use the streaming/ example or the streaming-cdc pattern)
  • ❌ Cross-region disaster recovery (single region; see DR.md)
  • ❌ Workspace-level RBAC automation (provision identities + assign roles in your platform IaC, then az rest after deploy)
  • ❌ Power BI Apps (publishing a packaged app — manual portal step)
  • ❌ Real Fabric Deployment Pipelines (Dev → Test → Prod — see Power BI Roadmap pattern)

These are deliberately scoped out to keep the example understandable; pointers to where they live in the rest of the repo are above.

Cost estimate (USD, ballpark)

Item Dev (F2) Test (F4) Prod (F64)
Fabric capacity (24h pause off) $263/mo $526/mo $8,410/mo
Fabric capacity (8hr/day, paused otherwise) $88/mo $175/mo $2,803/mo
ADLS storage (50GB gold) ~$1/mo ~$1/mo ~$1/mo
Total dev (paused) ~$89/mo

Cost lever: F-SKUs can be paused. Schedule pause/resume via runbook for non-prod. Prod F64 typically runs 24/7.

Production hardening checklist

Before pointing real users at it:

  • Capacity assignment: prod workspace on its own F-SKU; don't share with dev
  • Workspace identity: configure managed identity for OneLake shortcut auth (vs SAS in this dev sample)
  • Private endpoints on the underlying ADLS (separate Bicep — see shared/modules/storage.bicep)
  • Fabric Trusted Workspace Access for cross-workspace shortcuts (if applicable)
  • Sensitivity labels on the semantic model (Microsoft Purview Information Protection)
  • Row-level security in TMDL — see tables/DimCustomer.tmdl for the example pattern
  • Object-level security — hide sensitive columns from non-privileged roles
  • Backup/restore strategy — currently OneLake doesn't have point-in-time restore; rely on source-of-truth ADLS retention
  • Eval suite for any Copilot-in-Power-BI usage on this model — see LLMOps pattern