Files
DataMate/runtime/datamate-python/app/module/kg_graphrag/kg_client.py
Jerry Yan 39338df808 feat(kg): 实现 Phase 2 GraphRAG 融合功能
核心功能:
- 三层检索策略:向量检索(Milvus)+ 图检索(KG 服务)+ 融合排序
- LLM 生成:支持同步和流式(SSE)响应
- 知识库访问控制:knowledge_base_id 归属校验 + collection_name 绑定验证

新增模块(9个文件):
- models.py: 请求/响应模型(GraphRAGQueryRequest, RetrievalStrategy, GraphContext 等)
- milvus_client.py: Milvus 向量检索客户端(OpenAI Embeddings + asyncio.to_thread)
- kg_client.py: KG 服务 REST 客户端(全文检索 + 子图导出,fail-open)
- context_builder.py: 三元组文本化(10 种关系模板)+ 上下文构建
- generator.py: LLM 生成(ChatOpenAI,支持同步和流式)
- retriever.py: 检索编排(并行检索 + 融合排序)
- kb_access.py: 知识库访问校验(归属验证 + collection 绑定,fail-close)
- interface.py: FastAPI 端点(/query, /retrieve, /query/stream)
- __init__.py: 模块入口

修改文件(3个):
- app/core/config.py: 添加 13 个 graphrag_* 配置项
- app/module/__init__.py: 注册 kg_graphrag_router
- pyproject.toml: 添加 pymilvus 依赖

测试覆盖(79 tests):
- test_context_builder.py: 13 tests(三元组文本化 + 上下文构建)
- test_kg_client.py: 14 tests(KG 响应解析 + PagedResponse + 边字段映射)
- test_milvus_client.py: 8 tests(向量检索 + asyncio.to_thread)
- test_retriever.py: 11 tests(并行检索 + 融合排序 + fail-open)
- test_kb_access.py: 18 tests(归属校验 + collection 绑定 + 跨用户负例)
- test_interface.py: 15 tests(端点级回归 + 403 short-circuit)

关键设计:
- Fail-open: Milvus/KG 服务失败不阻塞管道,返回空结果
- Fail-close: 访问控制失败拒绝请求,防止授权绕过
- 并行检索: asyncio.gather() 并发运行向量和图检索
- 融合排序: Min-max 归一化 + 加权融合(vector_weight/graph_weight)
- 延迟初始化: 所有客户端在首次请求时初始化
- 配置回退: graphrag_llm_* 为空时回退到 kg_llm_*

安全修复:
- P1-1: KG 响应解析(PagedResponse.content)
- P1-2: 子图边字段映射(sourceEntityId/targetEntityId)
- P1-3: collection_name 越权风险(归属校验 + 绑定验证)
- P1-4: 同步 Milvus I/O(asyncio.to_thread)
- P1-5: 测试覆盖(79 tests,包括安全负例)

测试结果:79 tests pass 
2026-02-20 09:41:55 +08:00

198 lines
6.6 KiB
Python

"""KG 服务 REST 客户端。
通过 httpx 调用 Java 侧 knowledge-graph-service 的查询 API,
包括全文检索和子图导出。
失败策略:fail-open —— KG 服务不可用时返回空结果 + 日志告警。
"""
from __future__ import annotations
import httpx
from app.core.logging import get_logger
from app.module.kg_graphrag.models import EntitySummary, RelationSummary
logger = get_logger(__name__)
class KGServiceClient:
"""Java KG 服务 REST 客户端。"""
def __init__(
self,
*,
base_url: str = "http://datamate-kg:8080",
internal_token: str = "",
timeout: float = 30.0,
) -> None:
self._base_url = base_url.rstrip("/")
self._internal_token = internal_token
self._timeout = timeout
self._client: httpx.AsyncClient | None = None
@classmethod
def from_settings(cls) -> KGServiceClient:
from app.core.config import settings
return cls(
base_url=settings.graphrag_kg_service_url,
internal_token=settings.graphrag_kg_internal_token,
timeout=30.0,
)
def _get_client(self) -> httpx.AsyncClient:
if self._client is None:
self._client = httpx.AsyncClient(
base_url=self._base_url,
timeout=self._timeout,
)
return self._client
def _headers(self, user_id: str = "") -> dict[str, str]:
headers: dict[str, str] = {}
if self._internal_token:
headers["X-Internal-Token"] = self._internal_token
if user_id:
headers["X-User-Id"] = user_id
return headers
async def fulltext_search(
self,
graph_id: str,
query: str,
size: int = 10,
user_id: str = "",
) -> list[EntitySummary]:
"""调用 KG 服务全文检索,返回匹配的实体列表。
Fail-open: KG 服务不可用时返回空列表。
"""
try:
return await self._fulltext_search_impl(graph_id, query, size, user_id)
except Exception:
logger.exception(
"KG fulltext search failed for graph_id=%s (fail-open, returning empty)",
graph_id,
)
return []
async def _fulltext_search_impl(
self,
graph_id: str,
query: str,
size: int,
user_id: str,
) -> list[EntitySummary]:
client = self._get_client()
resp = await client.get(
f"/knowledge-graph/{graph_id}/query/search",
params={"q": query, "size": size},
headers=self._headers(user_id),
)
resp.raise_for_status()
body = resp.json()
# Java 返回 PagedResponse<SearchHitVO>:
# 可能被全局包装为 {"code": 200, "data": PagedResponse}
# 也可能直接返回 PagedResponse {"page": 0, "content": [...]}
data = body.get("data", body)
# PagedResponse 将实体列表放在 content 字段中
items: list[dict] = (
data.get("content", []) if isinstance(data, dict) else data if isinstance(data, list) else []
)
entities: list[EntitySummary] = []
for item in items:
entities.append(
EntitySummary(
id=str(item.get("id", "")),
name=item.get("name", ""),
type=item.get("type", ""),
description=item.get("description", ""),
)
)
return entities
async def get_subgraph(
self,
graph_id: str,
entity_ids: list[str],
depth: int = 1,
user_id: str = "",
) -> tuple[list[EntitySummary], list[RelationSummary]]:
"""获取种子实体的 N-hop 子图。
Fail-open: KG 服务不可用时返回空子图。
"""
try:
return await self._get_subgraph_impl(graph_id, entity_ids, depth, user_id)
except Exception:
logger.exception(
"KG subgraph export failed for graph_id=%s (fail-open, returning empty)",
graph_id,
)
return [], []
async def _get_subgraph_impl(
self,
graph_id: str,
entity_ids: list[str],
depth: int,
user_id: str,
) -> tuple[list[EntitySummary], list[RelationSummary]]:
client = self._get_client()
resp = await client.post(
f"/knowledge-graph/{graph_id}/query/subgraph/export",
params={"depth": depth},
json={"entityIds": entity_ids},
headers=self._headers(user_id),
)
resp.raise_for_status()
body = resp.json()
# Java 返回 SubgraphExportVO:
# 可能被全局包装为 {"code": 200, "data": SubgraphExportVO}
# 也可能直接返回 SubgraphExportVO {"nodes": [...], "edges": [...]}
data = body.get("data", body) if isinstance(body.get("data"), dict) else body
nodes_raw = data.get("nodes", [])
edges_raw = data.get("edges", [])
# ExportNodeVO: id, name, type, description, properties (Map)
entities: list[EntitySummary] = []
for node in nodes_raw:
entities.append(
EntitySummary(
id=str(node.get("id", "")),
name=node.get("name", ""),
type=node.get("type", ""),
description=node.get("description", ""),
)
)
relations: list[RelationSummary] = []
# 构建 id -> entity 的映射用于查找 source/target 名称和类型
entity_map = {e.id: e for e in entities}
# ExportEdgeVO: sourceEntityId, targetEntityId, relationType
# 注意:sourceId 是数据来源 ID,不是源实体 ID
for edge in edges_raw:
source_id = str(edge.get("sourceEntityId", ""))
target_id = str(edge.get("targetEntityId", ""))
source_entity = entity_map.get(source_id)
target_entity = entity_map.get(target_id)
relations.append(
RelationSummary(
source_name=source_entity.name if source_entity else source_id,
source_type=source_entity.type if source_entity else "",
target_name=target_entity.name if target_entity else target_id,
target_type=target_entity.type if target_entity else "",
relation_type=edge.get("relationType", ""),
)
)
return entities, relations
async def close(self) -> None:
if self._client is not None:
await self._client.aclose()
self._client = None