feat: route background ai workloads by model

ai_knowledge_base_v2/apps/backend/tests/test_ai_cost_tracking.py

@@ -9,6 +9,7 @@ from sqlalchemy.pool import StaticPool
 from app.models import Base
 from app.models.ai_config import ModelConfig
 from app.models.logs import AiRequestLog
+from app.services.admin_service import AdminDashboardService
 from app.services.ai_request_log_service import AiRequestLogService
 
 
@@ -58,3 +59,52 @@ def test_ai_request_log_estimates_cost_from_model_price():
         assert log.question_type == "knowledge_grounded"
         assert "命中知识库" in (log.route_reason or "")
 
+
+def test_dashboard_breaks_cost_down_by_actual_scene_and_model():
+    with _db() as db:
+        model = ModelConfig(
+            id=1,
+            provider="test",
+            api_type="openai_compatible",
+            model_name="report-model",
+            api_url="https://example.com",
+            api_key="secret",
+            input_price_per_1k=Decimal("0.002"),
+            output_price_per_1k=Decimal("0.006"),
+            currency="CNY",
+            timeout_second=30,
+        )
+        db.add(model)
+        db.commit()
+        AiRequestLogService.write_success(
+            db,
+            session_id=None,
+            message_id=None,
+            user_id=3,
+            model_id=1,
+            model_name="report-model",
+            prompt="weekly report",
+            knowledge_ids="",
+            retrieve_count=0,
+            input_token=1000,
+            output_token=500,
+            cost_ms=120,
+            question_type="background_report",
+            route_reason="场景分流：周期报告",
+        )
+        db.commit()
+
+        stats = AdminDashboardService.stats(db)
+
+        assert stats["costBreakdown"] == [
+            {
+                "scene": "background_report",
+                "modelName": "report-model",
+                "requestCount": 1,
+                "inputToken": 1000,
+                "outputToken": 500,
+                "totalToken": 1500,
+                "estimatedCost": 0.005,
+                "currency": "CNY",
+            }
+        ]

ai_knowledge_base_v2/apps/backend/tests/test_model_routing.py

@@ -0,0 +1,154 @@
+from __future__ import annotations
+
+from decimal import Decimal
+
+import pytest
+from fastapi import HTTPException
+from sqlalchemy import create_engine
+from sqlalchemy.orm import Session
+from sqlalchemy.pool import StaticPool
+
+from app.models import Base
+from app.api.admin_settings import enable_model, set_default_model
+from app.models.admin import Admin
+from app.models.ai_config import ModelConfig, SystemConfig
+from app.models.logs import AiRequestLog
+from app.schemas.admin import DefaultModelRequest, EnableModelRequest
+from app.services.model_routing_service import ModelRoutingService
+from app.services.model_service import ModelClientService
+from app.services.tracked_generation_service import TrackedGenerationService
+
+
+def _db() -> Session:
+    engine = create_engine(
+        "sqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    Base.metadata.create_all(engine)
+    return Session(engine)
+
+
+def _model(
+    model_id: int,
+    name: str,
+    *,
+    is_default: int,
+    allow_report: int,
+    allow_summary: int,
+) -> ModelConfig:
+    return ModelConfig(
+        id=model_id,
+        provider="test",
+        api_type="openai_compatible",
+        model_name=name,
+        api_url="https://example.com",
+        api_key="secret",
+        enabled=1,
+        is_default=is_default,
+        allow_report=allow_report,
+        allow_summary=allow_summary,
+        allow_fixed_info=1,
+        allow_deep_chat=1,
+        input_price_per_1k=Decimal("0.002"),
+        output_price_per_1k=Decimal("0.006"),
+        currency="CNY",
+        timeout_second=30,
+    )
+
+
+def test_background_scenario_routes_without_changing_live_default_model():
+    with _db() as db:
+        main = _model(1, "main-model", is_default=1, allow_report=0, allow_summary=1)
+        report = _model(2, "report-model", is_default=0, allow_report=1, allow_summary=0)
+        db.add_all([main, report])
+        db.commit()
+
+        report_route = ModelRoutingService.resolve(db, "report")
+        summary_route = ModelRoutingService.resolve(db, "summary")
+
+        assert report_route.model is report
+        assert report_route.fallback_used is False
+        assert summary_route.model is main
+        assert ModelClientService._get_enabled_model(db) is main
+
+
+def test_background_scenario_falls_back_to_default_model():
+    with _db() as db:
+        main = _model(1, "main-model", is_default=1, allow_report=0, allow_summary=0)
+        db.add(main)
+        db.commit()
+
+        route = ModelRoutingService.resolve(db, "report")
+
+        assert route.model is main
+        assert route.fallback_used is True
+        assert "回退默认主模型" in route.reason
+
+
+def test_tracked_background_generation_records_actual_route_tokens_and_cost():
+    with _db() as db:
+        db.add(SystemConfig(config_key="mock_model_enabled", config_value="true"))
+        report = _model(2, "report-model", is_default=0, allow_report=1, allow_summary=0)
+        db.add(report)
+        db.commit()
+
+        completion = TrackedGenerationService.generate(
+            db,
+            prompt="生成本周报告",
+            scenario="report",
+            user_id=7,
+        )
+        db.commit()
+
+        log = db.query(AiRequestLog).one()
+        assert completion.model_name == "report-model"
+        assert log.model_id == report.id
+        assert log.user_id == 7
+        assert log.question_type == "background_report"
+        assert log.total_token == completion.input_token + completion.output_token
+        assert log.estimated_cost is not None
+        assert "周期报告" in (log.route_reason or "")
+
+
+def test_model_pool_keeps_one_default_and_rejects_disabling_last_default():
+    with _db() as db:
+        admin = Admin(id=1, username="admin", password="hash", name="管理员", status=1)
+        main = _model(1, "main-model", is_default=1, allow_report=1, allow_summary=1)
+        alternate = _model(2, "alternate-model", is_default=0, allow_report=1, allow_summary=1)
+        alternate.enabled = 0
+        db.add_all([admin, main, alternate])
+        db.commit()
+
+        enable_model(
+            EnableModelRequest(modelId=alternate.id, enabled=1),
+            db=db,
+            current_admin=admin,
+        )
+        db.refresh(main)
+        db.refresh(alternate)
+        assert main.is_default == 1
+        assert alternate.enabled == 1
+
+        set_default_model(
+            DefaultModelRequest(modelId=alternate.id),
+            db=db,
+            current_admin=admin,
+        )
+        db.refresh(main)
+        db.refresh(alternate)
+        assert main.is_default == 0
+        assert alternate.is_default == 1
+
+        enable_model(
+            EnableModelRequest(modelId=main.id, enabled=0),
+            db=db,
+            current_admin=admin,
+        )
+        with pytest.raises(HTTPException) as exc_info:
+            enable_model(
+                EnableModelRequest(modelId=alternate.id, enabled=0),
+                db=db,
+                current_admin=admin,
+            )
+        assert exc_info.value.status_code == 400

ai_knowledge_base_v2/apps/backend/tests/test_periodic_reports.py

@@ -12,9 +12,9 @@ from app.models.chat import ChatSession, TopicSession
 from app.models.entitlement import EntitlementPlan, UserEntitlement
 from app.models.growth import PeriodicReport, TopicSummary, UserGrowthProfile
 from app.models.user import User
-from app.services.model_service import ModelClientService
 from app.services.periodic_report_service import PeriodicReportService, periodic_report_dict
 from app.services.periodic_report_worker import PeriodicReportWorker, scheduled_period
+from app.services.tracked_generation_service import TrackedGenerationService
 
 
 def _db() -> Session:
@@ -150,9 +150,9 @@ def test_failed_async_report_is_retried(monkeypatch):
         )
         db.commit()
         monkeypatch.setattr(
-            ModelClientService,
-            "generate_text_or_raise",
-            staticmethod(lambda _db, _prompt: (_ for _ in ()).throw(RuntimeError("模型暂时不可用"))),
+            TrackedGenerationService,
+            "generate",
+            staticmethod(lambda *_args, **_kwargs: (_ for _ in ()).throw(RuntimeError("模型暂时不可用"))),
         )
 
         report_id = PeriodicReportWorker.claim_next(db, worker_id="retry-worker", now=_now() + timedelta(seconds=1))