How gt-analytics/fuam-basic works
Project type and purpose
• FUAM (Fabric Unified Admin Monitoring) is a solution-accelerator built entirely on Microsoft Fabric. It orchestrates data extraction via Pipelines and Notebooks, persists it in a Lakehouse (Delta Parquet), and exposes it through Semantic models and Power BI reports for unified monitoring of tenant usage, capacities, activities, and settings.
File structure
• Top-level files:
• LICENSE, README.md – project metadata
• files/deployment_file.json – deployment parameters (likely used by the Deploy notebook)
• start/Deploy_FUAM_Basic_Gen2.ipynb – primary deployment notebook
• media/… – images and screenshots for documentation
• No traditional “src/” or “backend/” folders because Fabric handles compute and storage via the service surface.
README for architectural insights
• Core components: Pipelines, Notebooks, Lakehouses, Semantic models, Power BI reports
• Modular structure: ingestion modules for Activities, Workspaces, Capacities, Tenant Settings, etc.
• Data flow: initial + incremental load via parameterized orchestration → raw data stored → transformation into Delta Parquet → semantic layer & Direct Lake consumption.
Main components and relationships
a. Components
1. Orchestration Pipelines (Fabric Pipelines)
2. Compute Units running Notebooks (extraction & transformation code)
3. Lakehouse (raw zone + curated zone with Delta Parquet tables)
4. Semantic Model (Lakehouse SQL endpoint / Direct Lake)
5. Power BI Reports (FUAM_Basic_Overview_Report + drill-downs)
6. Deployment Notebook (start/Deploy_FUAM_Basic_Gen2.ipynb)
7. Deployment Config (files/deployment_file.json)
8. External Fabric APIs (Tenant Scanner API, Git Connections, etc.)
b. Data flows
• Pipelines → trigger Notebooks → call Fabric REST APIs → write raw JSON to Lakehouse raw zone
• Notebooks (transform) → read raw JSON → write Delta Parquet tables to curated zone
• Semantic Model → points to Delta Parquet via Direct Lake → serves as dataset to Power BI
• Power BI → consumes dataset for dashboards
c. Architectural patterns
• Modular ingestion: each data domain (Activities, Capacities, Workspaces) is its own Notebook/module
• Parameterized orchestration: single main pipeline controls full vs. incremental loads
• Layered data zones: raw (persist API payloads) vs. curated (schema-on-write Delta Parquet)
• Direct Lake semantic layer for “live” reporting
d. Key technologies
• Fabric Pipelines (orchestration)
• Fabric Notebooks (Python/Power Query)
• Fabric Lakehouse (Delta Parquet storage)
• Fabric SQL Endpoint (Direct Lake)
• Power BI Reports
• Deployment via Notebook using deployment_file.json