Skip to content
CSA Loom — the Microsoft Fabric experience for Azure tenants where Fabric isn't yet available: lakehouses, warehouses, notebooks, semantic models, Activator rules, Data Agents, across Commercial, GCC, GCC-High, and DoD IL5

Direct Lake-Replacement Pattern

Comparative positioning note

This document is written from the perspective of Microsoft Azure, Cloud Scale Analytics, and CSA Loom. Any description of third-party or competing products, services, pricing, or capabilities is derived from publicly available documentation and sources believed accurate at the time of writing, and is provided for general comparison only. We do not claim expertise in, or authority over, any non-Microsoft product or service; the respective vendor's official documentation is the authoritative source for their offerings, which may change over time. Nothing here is intended to disparage any vendor — where a competing product has genuine advantages, we aim to note them honestly. Verify all third-party details against the vendor's current official documentation before making decisions.

Customers migrating from a third-party BI platform or on-prem Power BI Report Server use the CSA Loom Direct-Lake-Shim warm-cache materializer pattern to get Power BI Premium semantic models with framed-like freshness (5-30 s) — without waiting for Fabric Gov GA.

When this pattern fits

  • Customer has aging on-prem BI server (a third-party BI server / Power BI Report Server)
  • Customer wants to modernize to Power BI cloud + lakehouse
  • Customer's audit boundary blocks Fabric / Direct Lake
  • Customer has Power BI Premium F-SKU (GCC-H / IL5) or P-SKU (GCC; no Direct Lake parity available)
  • Workload characteristics: aggregate analytics + interactive dashboards; not sub-second-required

Architecture

flowchart LR
    classDef source fill:#5C2D91,stroke:#fff,color:#fff,stroke-width:2px
    classDef bronze fill:#cd7f32,stroke:#fff,color:#fff,stroke-width:2px
    classDef silver fill:#A8A9AD,stroke:#fff,color:#fff,stroke-width:2px
    classDef gold fill:#D4AF37,stroke:#fff,color:#fff,stroke-width:2px
    classDef shim fill:#D83B01,stroke:#fff,color:#fff,stroke-width:2px
    classDef bi fill:#0078D4,stroke:#fff,color:#fff,stroke-width:2px

    subgraph Sources["Operational sources (DB / files / streams)"]
        direction TB
        DB["DB"]:::source
        Files["Files"]:::source
        Streams["Streams"]:::source
    end

    subgraph Lake["Loom lakehouse (ADLS Gen2)"]
        direction TB
        Bronze["Bronze table"]:::bronze
        Silver["Silver table"]:::silver
        Gold["Gold table"]:::gold
        Bronze --> Silver --> Gold
    end

    subgraph Shim["Direct-Lake-Shim refresh pipeline"]
        direction TB
        EventGrid["Storage Event Grid"]:::shim
        ShimSvc["Loom Direct-Lake-Shim"]:::shim
        TOM["TOM partition refresh<br/>on Power BI Premium model"]:::shim
        EventGrid --> ShimSvc --> TOM
    end

    subgraph BI["BI surface (Power BI Premium)"]
        direction TB
        Model["Semantic Model<br/>(Premium Import)<br/>Authored in TMDL<br/>Deployed via TOM"]:::bi
        Reports["Power BI Reports<br/>(refresh on every<br/>model refresh)"]:::bi
        Model --> Reports
    end

    Sources -- "Loom Mirroring" --> Bronze
    Gold -- "commit event" --> EventGrid
    Gold -- "refresh" --> Model
    TOM --> Model

Migration playbook

Step 1 — Inventory the source BI estate

  • Catalogue all third-party BI workbooks / apps / on-prem PBI reports
  • Identify primary data sources (databases, files, cubes)
  • Map dashboards to underlying data models
  • Prioritize: top 20% of dashboards cover 80% of usage

Step 2 — Stand up Loom

Per Quick Start.

Step 3 — Migrate sources to Bronze

For each source: - Database sources → Loom Mirroring Engine (Debezium / Cosmos connector) - File sources → Azure Data Factory copy → ADLS Gen2 Bronze - Stream sources → Azure Stream Analytics → ADX (cross-engine to Bronze if needed)

Step 4 — Build Silver + Gold via Databricks notebooks

  • Cleanse + conform in Silver
  • Business semantics + star schemas in Gold
  • Partition Gold tables by date (or appropriate dimension) so partition-refresh works

Step 5 — Re-author semantic models in Power BI Desktop

For each priority dashboard: - Open original (third-party BI tool / PBI Report Server) - Re-author the semantic model in Power BI Desktop targeting the Gold tables - Use TMDL format (Power BI Project / .pbip) - Commit .pbip to Git

Step 6 — Configure Direct-Lake-Shim refresh policy

Per Tutorial 03:

loom-dl-shim configure \
  --semantic-model <model-id> \
  --table <gold-table-name> \
  --refresh-policy partition \
  --partition-column date

Step 7 — Build Power BI reports

  • Re-create dashboards in Power BI Desktop against the new semantic models
  • Publish to Power BI Premium workspace
  • Verify visual parity with original

Step 8 — Cutover

  • Open new Power BI report URLs to a pilot user group
  • Validate report performance + freshness (5-30 s should be acceptable for most analytical workloads)
  • Migrate users in cohorts
  • Decommission original BI server after parallel-run period

Expected outcomes

  • Self-service Power BI authoring (vs a centralized third-party BI admin model)
  • 5-30 s freshness on partition-refresh (vs daily / hourly on legacy BI servers)
  • Modern data lake foundation (Loom lakehouse) supports additional workloads beyond BI
  • When Fabric Gov GA arrives: re-author semantic models for native Direct Lake on OneLake → sub-second freshness without rewriting the underlying analytics

Cost comparison (typical)

Item Legacy on-prem BI server Loom + Power BI Premium
BI server licenses $$$$ annual $0 (Power BI Premium consumption)
Server hardware + maintenance $$$ annual $0 (cloud-managed)
Storage $$ annual $$$ (Azure ADLS Gen2 + Power BI workspace storage)
Compute (ETL) $$$ annual $$ (Databricks DBU usage-based)
Refresh latency hours to nightly 5-30 seconds
Self-service authoring Limited Yes (Power BI Desktop)

Typical net savings: 40-60% TCO + step-function improvement in freshness + self-service.

Honest gap

This pattern delivers 5-30 second freshness. Fabric's native Direct Lake gives sub-second. For workloads requiring sub-second: - Wait for Fabric Gov GA - Or use Databricks SQL Warehouse + DirectQuery (Commercial only; always live but slower DAX engine) - Or use Power BI DirectQuery against Synapse Serverless (slower per-query)

For most analytical workloads (monthly / quarterly / yearly aggregates + daily refresh), 5-30 s is better than what the legacy BI server delivered and acceptable for the user's experience.