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

Data Activator parity (Reflex)

What Fabric does

Reflex (Data Activator) is a declarative rules engine over streaming + tabular events. Built on top of Eventhouse (KQL/ADX) for the query work + per-object state tracking for andStays / noPresence semantics. Per Microsoft engineering blogs (see research/03-fabric- only-internals.md §3), Reflex is not its own stream-processing runtime — it's a thin layer that schedules KQL queries against Eventhouse on intervals and maintains object state.

Entity model: Data sources → Events (filters / projections) → Objects (stateful entities keyed by splitColumn) → Attributes (computed properties) → Rules → Actions.

Rule primitives: increasesAbove, decreasesBelow, is above, is below, changesTo, andStays(duration), noPresenceOfData(seconds), everyNthTime(n, seconds).

Action surface: Teams, Email, Power Automate, Pipeline run, Notebook run, Spark job, Dataflow Gen2 refresh, User Data Function call.

End-to-end latency: 5-30 s.

CSA Loom parity design — apps/fiab-activator-engine

Full design in ADR fiab-0005. Summary:

Component Implementation
Query engine Azure Data Explorer (same Kusto engine Fabric Eventhouse uses)
Rule evaluator NRules (.NET production Rete rules engine)
Per-object state Azure Cache for Redis Premium (TTL'd keys)
Rule scheduler Cron orchestrator reading from Cosmos DB definitions
Action dispatcher Azure Functions with output bindings

All 8 Fabric Reflex primitives implemented. Latency target: 5-30 s end-to-end (matches Fabric).

Rule definition

JSON schema documented in apps/fiab-activator-engine/schemas/. Authored via Loom Console "Activator" pane (visual designer with KQL backing store) or directly in JSON for Git-based authoring.

Example (CPU high sustained):

{
  "id": "rule-cpu-high-andstays",
  "workspaceId": "ws-001",
  "name": "VM CPU sustained high",
  "dataSource": {
    "type": "adx-kql",
    "cluster": "adx-loom-eastus2",
    "database": "telemetry",
    "query": "VmMetrics | where TimestampUtc > ago(15m) | summarize avg(CpuPercent) by VmId, bin(TimestampUtc, 1m)",
    "splitColumn": "VmId",
    "cadenceMinutes": 1
  },
  "rules": [
    {
      "name": "Sustained CPU > 85% for 5 min",
      "expression": {
        "operator": "andStays",
        "left": {
          "operator": "isAbove",
          "attribute": "avg_CpuPercent",
          "threshold": 85
        },
        "durationMinutes": 5
      },
      "actions": [
        {
          "type": "teams-message",
          "channel": "#ops-alerts",
          "template": "VM {VmId} CPU is at {avg_CpuPercent}% for >5 min."
        },
        {
          "type": "databricks-job",
          "workspace": "dbx-ws-001",
          "jobId": "12345"
        }
      ]
    }
  ],
  "enabled": true
}

Action surface

Status as of 2026-06-06 (✅ shipped / 🟡 editor-only / ⛔ roadmap, not built):

Action type Binding Status
Teams message Graph API / webhook ✅ engine + editor
Email Azure Communication Services ✅ engine + editor
Logic App HTTP webhook ✅ engine
Generic webhook HTTPS POST ✅ engine + editor
ADF Pipeline ADF REST 🟡 editor configures it
Notebook run notebook id 🟡 editor configures it
Power Automate flow HTTP trigger 🟡 editor configures it
Databricks Job Databricks Jobs API ⛔ not implemented
User Data Function Function call ⛔ not implemented

The C# engine (ActionDispatcher.cs) dispatches the four ✅ rows; the Console activator editor (phase3-editors.tsx) additionally configures the three 🟡 rows. The ⛔ rows are roadmap.

Per-boundary behavior

Boundary ADX NRules / Redis / Functions
Commercial ✅ ✅ Container Apps
GCC ✅ ✅ Container Apps
GCC-High / IL4 ✅ ✅ AKS workload (Container Apps not at IL4+)
IL5 (v1.1) ✅ ✅ AKS workload

Honest gaps

  • Per-object state in Redis is functionally equivalent but Fabric's internal state store may have stricter exactly-once guarantees (Microsoft hasn't published these — flagged in research/03-fabric-only-internals.md §3 as thin info). Loom's semantics documented openly.
  • Visual rule designer is functional in v1 but won't match Fabric's drag-drop UX polish; v1.1 invests in UX parity.
  • Cron-based 1-min minimum cadence — Fabric's internal scheduler may run more often. For sub-second latency, use Azure Stream Analytics directly (out of scope).

Forward migration

Rules export to JSON via apps/fiab-activator-engine/export → Reflex definition import via Fabric REST API. State doesn't migrate (by design — rules re-arm in Fabric).