Jerry Yan
39338df808
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 ✅
DataMate All-in-One Data Work Platform
DataMate is an enterprise-level data processing platform for model fine-tuning and RAG retrieval, supporting core functions such as data collection, data management, operator marketplace, data cleaning, data synthesis, data annotation, data evaluation, and knowledge generation.
If you like this project, please give it a Star⭐️!
🌟 Core Features
- Core Modules: Data Collection, Data Management, Operator Marketplace, Data Cleaning, Data Synthesis, Data Annotation, Data Evaluation, Knowledge Generation.
- Visual Orchestration: Drag-and-drop data processing workflow design.
- Operator Ecosystem: Rich built-in operators and support for custom operators.
🚀 Quick Start
Prerequisites
- Git (for pulling source code)
- Make (for building and installing)
- Docker (for building images and deploying services)
- Docker-Compose (for service deployment - Docker method)
- Kubernetes (for service deployment - k8s method)
- Helm (for service deployment - k8s method)
This project supports deployment via two methods: docker-compose and helm. After executing the command, please enter the corresponding number for the deployment method. The command echo is as follows:
Choose a deployment method:
1. Docker/Docker-Compose
2. Kubernetes/Helm
Enter choice:
Clone the Code
git clone git@github.com:ModelEngine-Group/DataMate.git
cd DataMate
Deploy the basic services
make install
If the machine you are using does not have make installed, please run the following command to deploy it:
# Windows
set REGISTRY=ghcr.io/modelengine-group/
docker compose -f ./deployment/docker/datamate/docker-compose.yml up -d
docker compose -f ./deployment/docker/milvus/docker-compose.yml up -d
# Linux/Mac
export REGISTRY=ghcr.io/modelengine-group/
docker compose -f ./deployment/docker/datamate/docker-compose.yml up -d
docker compose -f ./deployment/docker/milvus/docker-compose.yml up -d
Once the container is running, access http://localhost:30000 in a browser to view the front-end interface.
To list all available Make targets, flags and help text, run:
make help
Build and deploy Mineru Enhanced PDF Processing
make build-mineru
make install-mineru
Deploy the DeerFlow service
make install-deer-flow
Local Development and Deployment
After modifying the local code, please execute the following commands to build the image and deploy using the local image.
make build
make install dev=true
Uninstall
make uninstall
When running make uninstall, the installer will prompt once whether to delete volumes; that single choice is applied to all components. The uninstall order is: milvus -> label-studio -> datamate, which ensures the datamate network is removed cleanly after services that use it have stopped.
🤝 Contribution Guidelines
Thank you for your interest in this project! We warmly welcome contributions from the community. Whether it's submitting bug reports, suggesting new features, or directly participating in code development, all forms of help make the project better.
• 📮 GitHub Issues: Submit bugs or feature suggestions.
• 🔧 GitHub Pull Requests: Contribute code improvements.
📄 License
DataMate is open source under the MIT license. You are free to use, modify, and distribute the code of this project in compliance with the license terms.