Home > Docs > Reference Architectures > Hybrid Cloud

☁️ Hybrid Cloud Reference Architecture¶

Last Updated: 2026-05-05 | Version: 1.0.0

📑 Table of Contents¶

• 🎯 Architecture Overview
• 🏗️ Architecture Diagram
• 📦 Component Table
• 📏 Capacity Sizing Guidance
• 🔒 Network Architecture
• 💰 Cost Estimation Framework
• 🚀 Deploy This Architecture
• ⚖️ Tradeoffs and Limitations
• 📚 References

🎯 Architecture Overview¶

This architecture combines Microsoft Fabric's analytics strengths (lakehouse, Direct Lake BI, data engineering) with Azure services that cover what Fabric doesn't natively provide: Azure SQL Database for transactional OLTP workloads, Azure Kubernetes Service for custom applications and microservices, and Azure Event Hubs for high-volume streaming ingestion exceeding Fabric Eventstream limits. Integration between Azure services and Fabric uses three patterns — Mirroring (real-time replication from Azure SQL to OneLake), Shortcuts (zero-copy access to ADLS Gen2 data), and Dataflow Gen2 (ETL from diverse sources into lakehouses). This hybrid approach lets organizations keep workloads on the best-fit platform while maintaining a unified analytics layer in Fabric.

🏗️ Architecture Diagram¶

graph TB
    subgraph AzureServices["Azure Platform Services"]
        SQLDB[Azure SQL Database<br/>OLTP Workloads]
        AKS[Azure Kubernetes Service<br/>Custom Applications]
        EH[Azure Event Hubs<br/>High-Volume Streaming]
        ADLS[ADLS Gen2<br/>Data Lake Storage]
        KV[Azure Key Vault<br/>Secrets Management]
        APIM[Azure API Management<br/>API Gateway]
    end

    subgraph Integration["Integration Layer"]
        MIRROR[Mirroring<br/>Azure SQL → OneLake]
        SHORTCUT[OneLake Shortcuts<br/>ADLS Gen2 → Lakehouse]
        DF2[Dataflow Gen2<br/>Multi-Source ETL]
        ES[Eventstream<br/>Event Hubs Consumer]
    end

    subgraph FabricPlatform["Fabric Platform (Capacity F64–F128)"]
        subgraph DataEng["Data Engineering Workspace"]
            BRONZE[Bronze Lakehouse<br/>Raw Ingestion]
            SILVER[Silver Lakehouse<br/>Cleansed Data]
            GOLD[Gold Lakehouse<br/>Business KPIs]
            NB[Spark Notebooks<br/>Transformations]
            PIPE[Data Pipelines<br/>Orchestration]
        end

        subgraph RTI["Real-Time Intelligence Workspace"]
            EVTH[Eventhouse<br/>KQL Analytics]
            RTDASH[Real-Time Dashboard]
        end

        subgraph BI["BI Workspace"]
            SEM[Semantic Model<br/>Direct Lake]
            PBI[Power BI Reports]
        end
    end

    subgraph Consumers["Consumers"]
        APP[Custom Apps via AKS]
        USERS[Business Users]
        OPS[Operations Team]
    end

    SQLDB -->|Mirroring| MIRROR
    MIRROR --> BRONZE
    ADLS -->|Shortcuts| SHORTCUT
    SHORTCUT --> BRONZE
    AKS -->|REST API writes| ADLS
    AKS -->|Events| EH
    EH --> ES
    ES --> EVTH
    DF2 --> BRONZE
    PIPE --> NB
    NB --> BRONZE
    BRONZE --> NB
    NB --> SILVER
    SILVER --> NB
    NB --> GOLD
    GOLD --> SEM
    SEM --> PBI
    EVTH --> RTDASH
    PBI --> USERS
    RTDASH --> OPS
    APIM --> AKS
    KV --> AKS
    KV --> PIPE
    APP --> APIM

📦 Component Table¶

📏 Capacity Sizing Guidance¶

Fabric Capacity¶

Azure Services (Complementary)¶

🔒 Network Architecture¶

Hybrid deployments require unified networking across Azure services and Fabric:

┌─────────────────────────────────────────────┐
│                Hub VNet                      │
│  ┌──────────┐  ┌──────────┐  ┌───────────┐ │
│  │ Firewall │  │ DNS Zone │  │ VNet GW   │ │
│  └──────────┘  └──────────┘  └───────────┘ │
└────────┬──────────┬──────────────┬──────────┘
         │          │              │
   ┌─────┴───┐ ┌───┴────┐  ┌─────┴──────┐
   │AKS Spoke│ │SQL Spoke│  │Fabric Spoke│
   │  + EH   │ │  + KV   │  │   + ADLS   │
   └─────────┘ └────────┘  └────────────┘

💰 Cost Estimation Framework¶

Cost optimization strategies:

1. Mirroring over ETL — Eliminates pipeline CU cost for SQL-to-Fabric data movement
2. Shortcuts over copy — Zero-copy reads from ADLS avoid data duplication costs
3. Event Hubs auto-inflate — Scale throughput units dynamically with demand
4. AKS spot nodes — Use spot instances for non-critical batch processing (save 60–90%)
5. Reserved instances — Apply to SQL DB and AKS for predictable workloads (save 30–40%)
6. Pause Fabric capacity — During hours when only Azure services are active

🚀 Deploy This Architecture¶

Infrastructure as Code¶

Step-by-Step Tutorials¶

Key Feature Documentation¶

⚖️ Tradeoffs and Limitations¶

📚 References¶