安全加固 + 数据管线修复 + 质量改进(第2批)

安全加固: - /api/v1/predict 限流改用 get_client_ip 防 X-Forwarded-For 伪造 - 新增 TRUST_PROXY_HEADERS/REQUIRE_ADMIN_AUTH 配置,默认 fail-closed - 生产环境管理接口未配置鉴权时拒绝(503),不再放行数据管线修复: - injuries IntegrityError 改用 begin_nested(SAVEPOINT)隔离批次 - injuries 空名单 vs 未配置语义区分(has_data 精确标记) - bzzoiro 统计字段映射(shots/possession/cards) + 入库条件放宽 - available_at 语义收紧(回测防泄漏) - team_names NFKD 去变音双重查找 + 补齐变体键预测系统改进: - multi-agent 全专家失败时 status=degraded 跳过终裁 - agent_weights 独立持久化到 predictions 表新增迁移: - 0014_predictions_agent_weights.py 新增测试(8个文件): - test_ip_spoofing.py: 限流防伪造 - test_require_admin_fail_closed.py: 生产 fail-closed - test_injuries_integrity_rollback.py: SAVEPOINT 隔离 - test_injuries_slice_semantic.py: 空名单 vs 未配置 - test_available_at.py: 回测防泄漏 - test_bzzoirot_stats.py: 统计字段映射 - test_team_names_normalize.py: NFKD 变体 - test_multi_agent_degraded.py: 全失败 degraded - test_agent_weights_persist.py: 权重持久化
2026-09-19 08:02:57 +00:00
parent bee330f31f
commit 6c24672e87
21 changed files with 1580 additions and 87 deletions
@@ -0,0 +1,141 @@
+"""回归测试: predictions 表 agent_weights 独立持久化。
+
+验证:
+  1. ORM 模型有 agent_weights 列(JSONB, 可空)
+  2. 迁移文件存在且可逆
+  3. orchestrator 写入 agent_weights
+"""
+from __future__ import annotations
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from src.db.models import Prediction
+
+
+class TestAgentWeightsColumn:
+    """验证 predictions 表有 agent_weights 列。"""
+
+    def test_orm_has_agent_weights_column(self):
+        """ORM 模型应包含 agent_weights 列。"""
+        cols = {c.name: c for c in Prediction.__table__.columns}
+        assert "agent_weights" in cols, "predictions 表应有 agent_weights 列"
+
+    def test_agent_weights_is_jsonb(self):
+        """agent_weights 应为 JSONB 类型。"""
+        col = Prediction.__table__.columns["agent_weights"]
+        # JSONB 类型检查
+        assert "JSON" in str(col.type).upper() or "JSONB" in str(col.type).upper()
+
+    def test_agent_weights_nullable(self):
+        """agent_weights 应可空(旧行保持 NULL)。"""
+        col = Prediction.__table__.columns["agent_weights"]
+        assert col.nullable is True, "agent_weights 应可空"
+
+
+class TestMigration:
+    """验证迁移文件存在且内容正确。"""
+
+    def test_migration_exists(self):
+        import os
+
+        path = "/.octop/workspaces/CA7PFH/Profeto/alembic/versions/0014_predictions_agent_weights.py"
+        assert os.path.exists(path)
+
+    def test_migration_content(self):
+        path = "/.octop/workspaces/CA7PFH/Profeto/alembic/versions/0014_predictions_agent_weights.py"
+        content = open(path).read()
+
+        assert "agent_weights" in content
+        assert "upgrade" in content
+        assert "downgrade" in content
+        assert "downgrade" in content and "drop_column" in content
+        assert "nullable=True" in content
+
+
+class TestOrchestratorWritesAgentWeights:
+    """验证 orchestrator 写入 agent_weights。"""
+
+    @pytest.mark.asyncio
+    async def test_orchestrator_writes_agent_weights_to_upsert(self):
+        """orchestrator 应将 agent_weights 传入 _upsert_prediction。"""
+        from src.llm.agents import orchestrator as orch_mod
+        from src.llm.agents.base import AgentReport
+        from src.llm.context_builder import MatchHeader
+
+        header = MatchHeader(
+            match_id=999, home_name="A", away_name="B",
+            league_name="X", season=None, match_date="?",
+            match_dt=None, stage=None,
+            home_team_id=1, away_team_id=2, league_id=1,
+        )
+
+        # 构造有效专家报告(至少 1 个 ok)
+        reports = [
+            AgentReport(agent="form", status="ok", analysis="good"),
+            AgentReport(agent="stats", status="error", analysis="failed"),
+            AgentReport(agent="home_away", status="ok", analysis="good"),
+            AgentReport(agent="injuries", status="no_data", analysis="无数据"),
+            AgentReport(agent="h2h", status="error", analysis="failed"),
+        ]
+
+        captured_values = {}
+
+        async def mock_specialists(h, *, version, before):
+            return reports
+
+        async def mock_provider(aid, **kw):
+            return MagicMock(model="test-model")
+
+        async def mock_header(mid, db=None):
+            return header
+
+        async def mock_aggregator(header, reports, *, provider, version):
+            return {
+                "pred_home_goals": 2,
+                "pred_away_goals": 1,
+                "pred_1x2": "1",
+                "subjective_confidence": 0.7,
+                "reasoning": "test",
+                "agent_weights": {"form": 0.3, "home_away": 0.5, "stats": 0.2},
+            }, 100, 50
+
+        async def mock_upsert(session, **kw):
+            captured_values.update(kw.get("values", {}))
+            p = MagicMock()
+            p.id = 1
+            p.provider = "test"
+            p.model = "test"
+            p.prompt_version = "v1"
+            p.pred_home_goals = 2
+            p.pred_away_goals = 1
+            p.pred_1x2 = "1"
+            p.subjective_confidence = 0.7
+            p.reasoning = "test"
+            p.agent_outputs = []
+            p.agent_weights = kw["values"].get("agent_weights")
+            return p
+
+        class FakeUow:
+            async def __aenter__(self):
+                return self
+            async def __aexit__(self, *a):
+                pass
+            async def get(self, cls, id):
+                return MagicMock()
+
+        with patch.object(orch_mod, "run_specialists", mock_specialists), \
+             patch.object(orch_mod, "_agent_provider", mock_provider), \
+             patch.object(orch_mod, "load_match_header", mock_header), \
+             patch.object(orch_mod, "_upsert_prediction", mock_upsert), \
+             patch.object(orch_mod, "run_aggregator", mock_aggregator), \
+             patch.object(orch_mod, "get_uow", FakeUow):
+
+            result = await orch_mod.predict_match_multi(999)
+
+        # 断言 agent_weights 被写入
+        assert "agent_weights" in captured_values, "agent_weights 应传入 _upsert_prediction"
+        assert captured_values["agent_weights"] is not None, "agent_weights 不应为 None"
+        assert "form" in captured_values["agent_weights"], "agent_weights 应包含专家权重"
+        print(f"PASS: agent_weights = {captured_values['agent_weights']}")