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

Scale-by-SKU admin page

Surface: /admin/scaling BFF: apps/fiab-console/app/api/admin/scaling/*/route.ts Bicep dependency: every backing service module under platform/fiab/bicep/modules/**

The Scale-by-SKU admin page lets a Loom administrator change the service tier of every backing Azure resource that powers CSA Loom — without leaving the console and without opening the Azure portal.

Per .claude/rules/no-vaporware.md, every dropdown on this page maps to a real Azure REST call. If a service is not provisioned the card shows an honest MessageBar with the precise env var + bicep module the admin needs to fix.

Services covered

Service Scale axis Backend
Power BI / Fabric Capacity F-SKU (F2 → F2048) / P-SKU (P1/P2/P3) ARM PATCH Microsoft.Fabric/capacities/{n} or Microsoft.PowerBIDedicated/capacities/{n}
Synapse Dedicated SQL Pool DWU (DW100c → DW30000c) ARM PATCH bigDataPools/sqlPools/{n}
ADX Cluster vCore tier (Dev → E2/E4/E8/E16/E64) + capacity ARM PATCH Microsoft.Kusto/clusters/{n}
Databricks SQL Warehouse cluster_size (2X-Small → 4X-Large) Databricks REST POST /api/2.0/sql/warehouses/{id}/edit
Databricks Cluster node_type_id + num_workers + autoscale Databricks REST POST /api/2.0/clusters/edit
AI Search SKU + replica count + partition count ARM PATCH Microsoft.Search/searchServices/{n}
APIM SKU + capacity ARM PATCH Microsoft.ApiManagement/service/{n}
Cosmos DB RU/s per container OR autoscale max RU/s Container.readOffer + OfferDefinition.replace (data plane)
Container Apps Workload profile (Consumption / D-/E-series) + replicas ARM PATCH Microsoft.App/containerApps/{n}
AI Foundry compute vmSize + min/max nodes for AmlCompute targets ARM PATCH Microsoft.MachineLearningServices/workspaces/{n}/computes/{c}

Authentication & RBAC

All cards run as the Console UAMI (uami-loom-console-{region}). The UAMI needs:

  • Capacity Contributor on each Fabric / Power BI Premium capacity (otherwise the GET returns 401/403 and the card shows the remediation text verbatim).
  • Synapse Administrator on the Synapse workspace.
  • Azure Kusto Contributor (or Contributor) on the ADX cluster.
  • Databricks Workspace Admin entitlement (granted via SCIM bootstrap).
  • Search Service Contributor on AI Search.
  • API Management Service Contributor on APIM.
  • Cosmos DB Built-in Data Contributor on the Cosmos account.
  • Container Apps Contributor on the ACA RG.
  • Contributor on the AI Foundry workspace.

How the UI surfaces honest state

Situation What the card shows
Service provisioned + UAMI authorized Current SKU + dropdown + cost preview + Apply button
Service not provisioned (env var missing) MessageBar intent="warning" with the exact env var + bicep module path
UAMI not authorized Verbatim 401/403 from the upstream API + hint with the role to grant
Async scale in progress provisioningState: Updating returned to the UI
SKU transition blocked by Azure (e.g. Developer → Premium APIM) Verbatim 400 from ARM surfaced in the card's error MessageBar
Cosmos serverless account Card shows "Serverless — no RU/s dial (billed per request)" — no fake control
ComputeInstance scale change Card shows 409 "delete + recreate to change vmSize" — no fake PATCH

Cost preview disclaimers

The cost-preview component (lib/components/admin-scaling/cost-preview.tsx) uses a hardcoded lookup table of East US 2 list prices in USD. It excludes:

  • Reserved-instance and savings-plan discounts
  • Regional differential (Gov pricing differs)
  • SLA surcharges (Premium APIM, ADX Optimized AutoScale, etc.)
  • Storage + bandwidth charges that go alongside compute

For exact billing, use the Cost Management → Cost Analysis blade in the Azure portal. The admin page deliberately does not call the Cost Management API today — that's tracked for v3.5.

Utilization metrics deferred (honest gap)

The cards do not show current utilization (DBU, CPU, request rate, RU consumption) because that requires Azure Monitor metrics integration — a separate piece of work. Per the no-vaporware rule we surface this gap in a MessageBar at the bottom of the page rather than show AI-hallucinated numbers.

Env vars consumed

Env var Used by Default
LOOM_SUBSCRIPTION_ID every card —
LOOM_DLZ_RG Synapse, Cosmos —
LOOM_ADMIN_RG APIM, AI Search, Foundry, Container Apps rg-csa-loom-admin-eastus2
LOOM_SYNAPSE_WORKSPACE Synapse DWU —
LOOM_KUSTO_CLUSTER_NAME ADX —
LOOM_DATABRICKS_HOSTNAME Databricks warehouse + cluster —
LOOM_AI_SEARCH_SERVICE AI Search —
LOOM_APIM_NAME APIM apim-csa-loom-eastus2
LOOM_COSMOS_ENDPOINT Cosmos —
LOOM_ACA_RG Container Apps falls back to LOOM_ADMIN_RG
LOOM_FOUNDRY_NAME AI Foundry compute —

Async scale gotchas

Most scale operations are asynchronous:

  • Fabric capacity SKU change — 30-90s for propagation; the workspace remains usable throughout but new artifact creation may briefly fail with "capacity is provisioning".
  • Synapse DWU change — pool transitions to Scaling, then back to Online after ~2-5 minutes. The pool is paused-effectively during the transition.
  • ADX SKU change — instances are added/removed in the background; the cluster remains queryable but ingestion may briefly stutter.
  • APIM SKU change — Premium → Standard transition can take 15-45 min; the gateway stays up.
  • Container Apps profile change — triggers a new revision deploy; the ingress stays up if revisionsMode: multiple.

The UI surfaces provisioningState: Updating so the admin knows it's in flight; they need to refresh the card after a few minutes to see the final state.

Bicep sync requirement

Per the no-vaporware rule, every scale axis must be provisioned by bicep before it can be exercised:

Service Bicep module
Fabric capacity platform/fiab/bicep/modules/fabric/capacity.bicep
Synapse pools platform/fiab/bicep/modules/data-platform/synapse.bicep
ADX platform/fiab/bicep/modules/real-time-intelligence/adx.bicep
Databricks platform/fiab/bicep/modules/data-platform/databricks.bicep
AI Search platform/fiab/bicep/modules/ai/ai-search.bicep
APIM platform/fiab/bicep/modules/integration/apim.bicep
Cosmos platform/fiab/bicep/modules/data-platform/cosmos.bicep
Container Apps platform/fiab/bicep/modules/compute/container-apps.bicep
AI Foundry platform/fiab/bicep/modules/ai/foundry.bicep

If a customer's deployment skipped a module, the scale card surfaces an explicit MessageBar that points at the missing bicep — never a fake "working" dropdown.

Testing

  • Unit (Vitest): apps/fiab-console/lib/azure/__tests__/scaling-clients.test.ts covers every new client method, and scaling-routes.test.ts covers the 10 BFF routes (auth, validation, happy path).
  • E2E (Playwright UAT): apps/fiab-console/e2e/admin-scaling.uat.ts walks /admin/scaling, asserts all 10 cards render, and verifies each BFF GET returns either ok=true or a 503 + hint (no opaque 5xx).

Run locally:

cd apps/fiab-console
pnpm exec vitest run lib/azure/__tests__/scaling-clients.test.ts lib/azure/__tests__/scaling-routes.test.ts
SESSION_SECRET=<from-KV> pnpm exec playwright test e2e/admin-scaling.uat.ts --project=uat