Home > Docs > Reference Architectures > Real-Time Analytics

⚡ Real-Time Analytics 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 is purpose-built for workloads that require sub-second data ingestion, real-time analytics, and automated alerting — such as IoT sensor monitoring, casino floor telemetry, financial tick data, and operational dashboards. Events flow from producers (devices, applications, brokers) through Fabric Eventstreams into an Eventhouse powered by KQL (Kusto Query Language) for high-performance time-series analytics. Real-Time Dashboards provide live visualizations with auto-refresh, while Data Activator monitors data streams for conditions that trigger automated alerts (email, Teams, Power Automate). A parallel lakehouse path archives streaming data into the medallion architecture for historical analytics and batch BI.

🏗️ Architecture Diagram¶

graph TB
    subgraph Producers["Event Producers"]
        IOT[IoT Devices<br/>MQTT / HTTP]
        APP[Application Events<br/>REST / SDK]
        KAFKA[Apache Kafka<br/>Broker]
        EH_EXT[Azure Event Hubs<br/>External]
    end

    subgraph FabricRTI["Fabric Real-Time Intelligence (Capacity F64)"]
        subgraph Ingestion["Ingestion Layer"]
            ES1[Eventstream<br/>Primary Ingest]
            ES2[Eventstream<br/>Secondary Ingest]
        end

        subgraph Analytics["Analytics Engine"]
            EVTH[Eventhouse<br/>KQL Database]
            KQLQS[KQL Queryset<br/>Saved Queries]
        end

        subgraph Visualization["Real-Time Visualization"]
            RTDASH[Real-Time Dashboard<br/>Auto-Refresh]
            MAPS[Maps Visual<br/>Geospatial]
        end

        subgraph Alerting["Automated Response"]
            DA[Data Activator<br/>Trigger Engine]
            EMAIL[Email Alerts]
            TEAMS[Teams Notifications]
            PA[Power Automate<br/>Workflows]
        end
    end

    subgraph HistoricalPath["Historical Analytics Path"]
        subgraph LHArch["Lakehouse Archive"]
            BRONZE_LH[Bronze Lakehouse<br/>Raw Event Archive]
            SILVER_LH[Silver Lakehouse<br/>Enriched Events]
            GOLD_LH[Gold Lakehouse<br/>Aggregated KPIs]
        end
        SEM[Semantic Model<br/>Direct Lake]
        PBI[Power BI Reports<br/>Historical BI]
    end

    subgraph Consumers["Consumers"]
        OPS[Operations Center]
        ANALYST[Data Analysts]
        AUTO[Automated Systems]
    end

    IOT --> ES1
    APP --> ES1
    KAFKA --> ES2
    EH_EXT --> ES2

    ES1 --> EVTH
    ES2 --> EVTH
    ES1 -->|Archive| BRONZE_LH

    EVTH --> KQLQS
    KQLQS --> RTDASH
    EVTH --> RTDASH
    EVTH --> MAPS
    EVTH --> DA

    DA --> EMAIL
    DA --> TEAMS
    DA --> PA

    BRONZE_LH --> SILVER_LH
    SILVER_LH --> GOLD_LH
    GOLD_LH --> SEM
    SEM --> PBI

    RTDASH --> OPS
    PBI --> ANALYST
    PA --> AUTO

📦 Component Table¶

📏 Capacity Sizing Guidance¶

By Ingestion Volume¶

By Use Case¶

Key sizing considerations:

• Hot cache determines query performance — more cache = faster queries on recent data
• Retention controls total storage — archive to lakehouse for data beyond retention period
• Eventhouse scales within capacity; if CU pressure is high, split workloads across capacities
• For volumes exceeding 1M events/sec, use Azure Event Hubs as a buffer (see Hybrid Cloud)

🔒 Network Architecture¶

💰 Cost Estimation Framework¶

Cost optimization strategies:

1. Right-size hot cache — Reduce cache duration for data queried infrequently (saves Eventhouse CUs)
2. Archive to lakehouse — Move data beyond retention window to cheaper OneLake storage
3. Materialized views — Pre-aggregate common queries to reduce CU per query
4. Pause capacity — If streaming is only active during business hours, pause overnight
5. Batch hybrid — Use Eventhouse for real-time window (last 7 days); use lakehouse for historical
6. Event batching — Batch small events at the producer to reduce per-event overhead

🚀 Deploy This Architecture¶

Infrastructure as Code¶

Step-by-Step Tutorials¶

Key Feature Documentation¶

Casino Floor Example¶

This reference architecture directly maps to the casino floor monitoring scenario in this POC:

See notebooks: - notebooks/bronze/01_bronze_slot_telemetry.py — Real-time slot ingestion - notebooks/gold/01_gold_slot_performance.py — Slot KPI aggregations

⚖️ Tradeoffs and Limitations¶

📚 References¶