merge: T6 知识库实体与图谱（T-P2-01/05）

2026-08-14 20:09:59 +08:00
parent ce429b4011 d9249400af
commit 81117a0104
10 changed files with 1499 additions and 10 deletions
@@ -0,0 +1,3 @@
+# pytest prepend 导入模式按 basename 注册模块：unit 与 integration 下的
+# 同名测试文件（如 test_kb_entity.py，任务契约指定两端同名）会撞模块名。
+# 加包标记后模块名唯一化为 tests.integration.<name>，与 tests/__init__.py 一致。
@@ -0,0 +1,406 @@
+"""kb 实体底座集成用例（真实 PostgreSQL，T-P2-01/05）.
+
+覆盖：实体生命周期（draft→reviewing→published、改 published 被拒、
+draft 改 version+1）、关系写入约束（自环/重复/端点缺失）、递归 CTE
+1~3 跳可解释路径与防环、列表过滤分页、自建色板读写。
+实体/关系录入没有 HTTP 路由（运营后台属 T-P2-09），这些路径直接驱动
+service；读路径走 HTTP，顺带验证路由装配与统一信封。
+"""
+
+from __future__ import annotations
+
+import os
+import uuid
+from collections.abc import AsyncIterator
+from contextlib import asynccontextmanager
+from typing import Any
+
+import pytest
+
+
+def _infra_available() -> bool:
+    return bool(os.getenv("MEIDA_RUN_INTEGRATION"))
+
+
+pytestmark = [
+    pytest.mark.integration,
+    pytest.mark.skipif(
+        not _infra_available(),
+        reason="set MEIDA_RUN_INTEGRATION=1 with PostgreSQL and Redis running",
+    ),
+]
+
+
+@pytest.fixture(autouse=True)
+async def _dispose_engine_after_test() -> AsyncIterator[None]:
+    """用例结束即释放全局引擎。
+
+    引擎是进程级懒加载单例，而 pytest-asyncio 每个用例一个新 event loop：
+    走 HTTP 的用例靠 client 的 lifespan teardown 顺手 dispose；本文件的
+    service 直连用例没有这层兜底，不主动释放会把绑定旧 loop 的连接池
+    泄漏给后续用例（表现为 Event loop is closed / sqlalchemy 报错）。
+    """
+    yield
+    from packages.shared.db import dispose_engine
+
+    await dispose_engine()
+
+
+@pytest.fixture
+async def client() -> Any:
+    import httpx
+
+    from apps.api.main import create_app
+    from packages.shared.config import get_settings
+
+    # kb 链路不碰 LLM，无需替换 provider；lifespan 照常起网关与连接池
+    app = create_app(get_settings())
+    transport = httpx.ASGITransport(app=app)
+    async with (
+        httpx.AsyncClient(transport=transport, base_url="http://test") as http,
+        app.router.lifespan_context(app),
+    ):
+        yield http
+
+
+@pytest.fixture
+async def auth(client: Any) -> dict[str, Any]:
+    resp = await client.post(
+        "/api/v1/auth/dev-token", json={"device_key": f"t6-{uuid.uuid4().hex[:8]}"}
+    )
+    assert resp.json()["code"] == 0
+    data = resp.json()["data"]
+    return {
+        "headers": {"Authorization": f"Bearer {data['access_token']}"},
+        "user_id": data["user_id"],
+    }
+
+
+@asynccontextmanager
+async def service_scope() -> AsyncIterator[Any]:
+    """直连 service 的事务作用域：成功 commit、异常 rollback。"""
+    from packages.kb.service import KbService
+    from packages.shared.db import get_session_factory
+
+    async with get_session_factory()() as session:
+        try:
+            yield KbService(session)
+            await session.commit()
+        except Exception:
+            await session.rollback()
+            raise
+
+
+async def _mk_entity(
+    kb: Any, *, entity_type: str, name: str, attrs: dict[str, Any] | None = None
+) -> Any:
+    from packages.kb.schemas import EntityCreate
+
+    return await kb.create_entity(
+        payload=EntityCreate(entity_type=entity_type, name=name, attrs=attrs or {}),  # type: ignore[arg-type]
+        updated_by="it",
+    )
+
+
+async def test_entity_lifecycle_publish_freezes_and_version_bumps(
+    client: Any, auth: dict[str, Any]
+) -> None:
+    headers = auth["headers"]
+    suffix = uuid.uuid4().hex[:8]
+
+    # 新建 → draft / version=1
+    created = await client.post(
+        "/api/v1/kb/items",
+        json={"entity_type": "item", "name": f"白衬衫-{suffix}", "attrs": {"sleeve": "long"}},
+        headers=headers,
+    )
+    assert created.status_code == 200
+    body = created.json()
+    assert body["code"] == 0
+    entity = body["data"]
+    entity_id = entity["entity_id"]
+    assert entity["status"] == "draft"
+    assert entity["version"] == 1
+    assert entity["attrs"] == {"sleeve": "long"}
+
+    # 详情可读
+    detail = (await client.get(f"/api/v1/kb/items/{entity_id}", headers=headers)).json()
+    assert detail["code"] == 0
+    assert detail["data"]["name"] == f"白衬衫-{suffix}"
+
+    # 非法跃迁：draft 不能直接 published
+    shortcut = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"status": "published"}, headers=headers
+    )
+    assert shortcut.status_code == 409
+    assert shortcut.json()["code"] == 40900
+
+    # draft → reviewing（纯跃迁，version 不变）→ published
+    reviewing = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"status": "reviewing"}, headers=headers
+    )
+    assert reviewing.json()["data"]["status"] == "reviewing"
+    assert reviewing.json()["data"]["version"] == 1
+    published = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"status": "published"}, headers=headers
+    )
+    assert published.json()["data"]["status"] == "published"
+
+    # 改 published 被拒（内容与回 draft 都不行）
+    frozen = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"attrs": {"sleeve": "short"}}, headers=headers
+    )
+    assert frozen.status_code == 409
+    assert frozen.json()["code"] == 40900
+    backdoor = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"status": "draft"}, headers=headers
+    )
+    assert backdoor.status_code == 409
+    assert backdoor.json()["code"] == 40900
+
+    # published → archived 收尾合法；archived 终态
+    archived = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"status": "archived"}, headers=headers
+    )
+    assert archived.json()["data"]["status"] == "archived"
+    resurrect = await client.patch(
+        f"/api/v1/kb/items/{entity_id}", json={"status": "reviewing"}, headers=headers
+    )
+    assert resurrect.status_code == 409
+
+    # draft 实体改内容 → version+1；同名同类目拒绝
+    draft2 = await client.post(
+        "/api/v1/kb/items",
+        json={"entity_type": "item", "name": f"牛仔裤-{suffix}"},
+        headers=headers,
+    )
+    draft2_id = draft2.json()["data"]["entity_id"]
+    edited = await client.patch(
+        f"/api/v1/kb/items/{draft2_id}",
+        json={"attrs": {"fit": "slim"}, "source": "搭配手册"},
+        headers=headers,
+    )
+    assert edited.json()["data"]["version"] == 2
+    assert edited.json()["data"]["attrs"] == {"fit": "slim"}
+    duplicate = await client.post(
+        "/api/v1/kb/items",
+        json={"entity_type": "item", "name": f"牛仔裤-{suffix}"},
+        headers=headers,
+    )
+    assert duplicate.status_code == 400
+    assert duplicate.json()["code"] == 40000
+
+    # 非法 entity_type / 超界 max_hops 由入口校验拦成 422
+    bad_type = await client.post(
+        "/api/v1/kb/items", json={"entity_type": "user", "name": "x"}, headers=headers
+    )
+    assert bad_type.status_code == 422
+    assert bad_type.json()["code"] == 40001
+    bad_hops = await client.get(f"/api/v1/kb/items/{draft2_id}/graph?max_hops=4", headers=headers)
+    assert bad_hops.status_code == 422
+
+
+async def test_relation_rules_self_loop_duplicate_and_missing() -> None:
+    from packages.kb.schemas import RelationCreate
+    from packages.shared.errors import AppError, ErrorCode
+
+    suffix = uuid.uuid4().hex[:8]
+    async with service_scope() as kb:
+        a = await _mk_entity(kb, entity_type="item", name=f"selfloop-{suffix}")
+        b = await _mk_entity(kb, entity_type="style", name=f"duppair-{suffix}")
+
+        # 自环拒绝
+        with pytest.raises(AppError) as loop_err:
+            await kb.add_relation(
+                payload=RelationCreate(
+                    from_entity_id=a.entity_id, to_entity_id=a.entity_id, rel_type="suits"
+                )
+            )
+        assert loop_err.value.code == ErrorCode.BAD_REQUEST
+
+        # 端点缺失 → 40400
+        with pytest.raises(AppError) as missing_err:
+            await kb.add_relation(
+                payload=RelationCreate(from_entity_id=a.entity_id, to_entity_id=1, rel_type="suits")
+            )
+        assert missing_err.value.code == ErrorCode.NOT_FOUND
+
+        # 三元组重复拒绝
+        await kb.add_relation(
+            payload=RelationCreate(
+                from_entity_id=a.entity_id, to_entity_id=b.entity_id, rel_type="suits"
+            )
+        )
+        with pytest.raises(AppError) as dup_err:
+            await kb.add_relation(
+                payload=RelationCreate(
+                    from_entity_id=a.entity_id, to_entity_id=b.entity_id, rel_type="suits"
+                )
+            )
+        assert dup_err.value.code == ErrorCode.BAD_REQUEST
+
+
+async def test_graph_two_hops_explainable_paths_and_cycle_guard(
+    client: Any, auth: dict[str, Any]
+) -> None:
+    headers = auth["headers"]
+    suffix = uuid.uuid4().hex[:8]
+    async with service_scope() as kb:
+        from packages.kb.schemas import RelationCreate
+
+        shirt = await _mk_entity(kb, entity_type="item", name=f"衬衫-{suffix}")
+        commut = await _mk_entity(kb, entity_type="style", name=f"通勤风-{suffix}")
+        interview = await _mk_entity(kb, entity_type="occasion", name=f"面试-{suffix}")
+        for from_e, to_e, rel in (
+            (shirt, commut, "suits"),
+            (commut, interview, "fits"),
+            (interview, shirt, "matches"),  # 成环：面试 → 衬衫 → 通勤风 → …
+        ):
+            await kb.add_relation(
+                payload=RelationCreate(
+                    from_entity_id=from_e.entity_id, to_entity_id=to_e.entity_id, rel_type=rel
+                )
+            )
+
+    graph = (
+        await client.get(f"/api/v1/kb/items/{shirt.entity_id}/graph?max_hops=2", headers=headers)
+    ).json()
+    assert graph["code"] == 0
+    data = graph["data"]
+    assert data["from_entity_id"] == str(shirt.entity_id)
+    assert data["max_hops"] == 2
+
+    def _signature(path: dict[str, Any]) -> tuple[tuple[str, ...], tuple[str, ...]]:
+        return (
+            tuple(node["entity_id"] for node in path["nodes"]),
+            tuple(hop["rel_type"] for hop in path["hops"]),
+        )
+
+    signatures = {_signature(p) for p in data["paths"]}
+    assert (
+        (str(shirt.entity_id), str(commut.entity_id)),
+        ("suits",),
+    ) in signatures, "1 跳路径应包含每跳 rel_type"
+    assert (
+        (str(shirt.entity_id), str(commut.entity_id), str(interview.entity_id)),
+        ("suits", "fits"),
+    ) in signatures, "2 跳路径应可解释：节点序列 + 逐跳 rel_type"
+    # 节点摘要可读（不是光秃秃的 id）
+    first = next(p for p in data["paths"] if len(p["hops"]) == 1)
+    assert first["nodes"][0]["name"] == f"衬衫-{suffix}"
+    assert first["nodes"][0]["entity_type"] == "item"
+
+    # 防环：3 跳也不会绕回起点或无限展开
+    deep = (
+        await client.get(f"/api/v1/kb/items/{shirt.entity_id}/graph?max_hops=3", headers=headers)
+    ).json()["data"]
+    for path in deep["paths"]:
+        ids = [node["entity_id"] for node in path["nodes"]]
+        assert len(ids) == len(set(ids)), "路径内不得重复出现同一实体"
+        assert ids[0] == str(shirt.entity_id)
+    # 环上只有 2 条不回头路径：衬衫→通勤风、衬衫→通勤风→面试（→衬衫 被防环剪掉）
+    assert len(deep["paths"]) == 2
+
+    # 起点 404 语义
+    missing = await client.get("/api/v1/kb/items/1/graph", headers=headers)
+    assert missing.status_code == 404
+    assert missing.json()["code"] == 40400
+
+
+async def test_list_items_filter_and_pagination(client: Any, auth: dict[str, Any]) -> None:
+    headers = auth["headers"]
+    suffix = uuid.uuid4().hex[:8]
+    for i in range(3):
+        resp = await client.post(
+            "/api/v1/kb/items",
+            json={"entity_type": "style", "name": f"pag-{suffix}-{i}"},
+            headers=headers,
+        )
+        assert resp.json()["code"] == 0
+    await client.post(
+        "/api/v1/kb/items",
+        json={"entity_type": "material", "name": f"pag-{suffix}-denim"},
+        headers=headers,
+    )
+
+    # entity_type 过滤 + 名称模糊（把别的历史数据隔离掉）
+    base = "/api/v1/kb/items"
+    first = (
+        await client.get(
+            f"{base}?entity_type=style&name=pag-{suffix}&limit=2&offset=0", headers=headers
+        )
+    ).json()["data"]
+    assert first["total"] == 3
+    assert len(first["items"]) == 2
+    second = (
+        await client.get(
+            f"{base}?entity_type=style&name=pag-{suffix}&limit=2&offset=2", headers=headers
+        )
+    ).json()["data"]
+    assert len(second["items"]) == 1
+    ids_first = {item["entity_id"] for item in first["items"]}
+    ids_second = {item["entity_id"] for item in second["items"]}
+    assert ids_first | ids_second and not ids_first & ids_second
+
+    # status 过滤：归档一条后，draft 视图少一行
+    archived_id = first["items"][0]["entity_id"]
+    await client.patch(
+        f"/api/v1/kb/items/{archived_id}", json={"status": "archived"}, headers=headers
+    )
+    drafts = (
+        await client.get(
+            f"{base}?entity_type=style&name=pag-{suffix}&status=draft", headers=headers
+        )
+    ).json()["data"]
+    assert drafts["total"] == 2
+
+    # 非法 status 枚举 → 422
+    bad_status = await client.get(f"{base}?status=frozen", headers=headers)
+    assert bad_status.status_code == 422
+
+
+async def test_color_palette_roundtrip() -> None:
+    from packages.shared.errors import AppError, ErrorCode
+
+    # season_code 是 varchar(16)：uuid 后缀只能取 8 位防超长
+    season = f"sw{uuid.uuid4().hex[:8]}"
+    async with service_scope() as kb:
+        color = await kb.add_color(
+            season_code=season,
+            role="base",
+            lab_l=82.35,
+            lab_a=1.2,
+            lab_b=9.87,
+            hex_code="#F2D8C2",
+            name_cn="杏色",
+        )
+        assert color.role == "base"
+        assert color.hex == "#F2D8C2"
+        assert color.lab_l == pytest.approx(82.35)
+
+        with pytest.raises(AppError) as hex_err:
+            await kb.add_color(
+                season_code=season,
+                role="accent",
+                lab_l=50,
+                lab_a=10,
+                lab_b=-10,
+                hex_code="F2D8C2",  # 缺 #
+                name_cn="坏色",
+            )
+        assert hex_err.value.code == ErrorCode.BAD_REQUEST
+
+        with pytest.raises(AppError) as role_err:
+            await kb.add_color(
+                season_code=season,
+                role="primary",  # 不在 base/accent/avoid
+                lab_l=50,
+                lab_a=10,
+                lab_b=-10,
+                hex_code="#112233",
+                name_cn="坏角色",
+            )
+        assert role_err.value.code == ErrorCode.BAD_REQUEST
+
+        rows = await kb.list_colors(season_code=season)
+        assert [row.hex for row in rows] == ["#F2D8C2"]