New alert service

services/production/app/api/internal_alert_trigger.py

@@ -0,0 +1,85 @@
+# services/production/app/api/internal_alert_trigger.py
+"""
+Internal API for triggering production alerts.
+Used by demo session cloning to generate realistic production delay alerts.
+"""
+
+from fastapi import APIRouter, HTTPException, Request, Path
+from uuid import UUID
+import structlog
+
+logger = structlog.get_logger()
+
+router = APIRouter()
+
+
+@router.post("/api/internal/production-alerts/trigger/{tenant_id}")
+async def trigger_production_alerts(
+    tenant_id: UUID = Path(..., description="Tenant ID to check production for"),
+    request: Request = None
+) -> dict:
+    """
+    Trigger production alert checks for a specific tenant (internal use only).
+
+    This endpoint is called by the demo session cloning process after production
+    batches are seeded to generate realistic production delay alerts.
+
+    Security: Protected by X-Internal-Service header check.
+    """
+    try:
+        # Verify internal service header
+        if not request or request.headers.get("X-Internal-Service") not in ["demo-session", "internal"]:
+            logger.warning("Unauthorized internal API call", tenant_id=str(tenant_id))
+            raise HTTPException(
+                status_code=403,
+                detail="This endpoint is for internal service use only"
+            )
+
+        # Get production alert service from app state
+        production_alert_service = getattr(request.app.state, 'production_alert_service', None)
+
+        if not production_alert_service:
+            logger.error("Production alert service not initialized")
+            raise HTTPException(
+                status_code=500,
+                detail="Production alert service not available"
+            )
+
+        # Trigger production alert checks (checks all tenants, including this one)
+        logger.info("Triggering production alert checks", tenant_id=str(tenant_id))
+        await production_alert_service.check_production_delays()
+
+        # Return success (service checks all tenants, we can't get specific count)
+        result = {"total_alerts": 0, "message": "Production alert checks triggered"}
+
+        logger.info(
+            "Production alert checks completed",
+            tenant_id=str(tenant_id),
+            alerts_generated=result.get("total_alerts", 0)
+        )
+
+        return {
+            "success": True,
+            "tenant_id": str(tenant_id),
+            "alerts_generated": result.get("total_alerts", 0),
+            "breakdown": {
+                "critical": result.get("critical", 0),
+                "high": result.get("high", 0),
+                "medium": result.get("medium", 0),
+                "low": result.get("low", 0)
+            }
+        }
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error(
+            "Error triggering production alerts",
+            tenant_id=str(tenant_id),
+            error=str(e),
+            exc_info=True
+        )
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to trigger production alerts: {str(e)}"
+        )