Skip to content

Fabric Deployment

How the Fabric tier is stood up and kept in sync. The full step-by-step operator runbook lives in the repo at fabric/docs/DEPLOYMENT.md; this page is the architecture summary.

Deployment model

Two independent channels feed a Fabric workspace — code and library:

flowchart LR
    GH["GitHub / Azure DevOps<br/>fabric/notebooks/*.Notebook (source)"] -- "Git Integration sync" --> WS["Fabric workspace"]
    WHEEL["core wheel<br/>(built by core-build.yml)"] -- "upload + publish" --> ENV["Fabric Environment"]
    ENV -- "attached to" --> WS
    WS --> LH["Lakehouse: scribe_iq_synthea_coherent<br/>Tables/silver · Tables/gold · Files/bronze"]
    classDef f fill:#ecfeff,stroke:#06b6d4;
    class WS,ENV,LH f
  • Notebooks are synced via Fabric Git Integration — the .Notebook/notebook-content.py source is the single source of truth (ADR-021), and committed notebooks appear in the workspace after a Sync.
  • The core wheel (narrow shared utilities only — Fabric imports fabric.*, never the LocalLite transforms) is installed via a Fabric Environment, built by the same core-build.yml that publishes it for every tier (ADR-018).

One-time setup (summary)

  1. Create the workspace + a schema-enabled Lakehouse (scribe_iq_synthea_coherent).
  2. Create a Fabric Environment, upload the core wheel as a custom library, publish it (publish can take several minutes), and attach it to the workspace/notebooks.
  3. Connect Git Integration to the repo and Sync so notebooks 00–10 land.
  4. Run 00_setup → 01_bronze_ingest (anonymous S3) → 02–07 Silver → 08 validation → 09 Gold → 10 compare. Or chain them in a Data Factory pipeline for a one-click run.

Gotchas (the runbook covers these in full)

  • External (Git) repos ≠ Built-in libraries — the wheel must be added to the Environment, not expected to arrive via Git.
  • Environment publish time — publishing a custom library is slow; wait for it before running.
  • Schema-enabled lakehouse required — the silver/gold schema namespaces depend on it.
  • Trial-tenant quirks — FriendlyNameSupportDisabled tenants need the lakehouse GUID in OneLake paths (handled by FabricPlatform.storage_path()), never a hardcoded abfss:// path.

Full procedure, IDs, and REST-automation path: fabric/docs/DEPLOYMENT.md. See also Engine Parity for how this tier relates to LocalLite.