提交 965bba60 authored 作者: caifuming's avatar caifuming

修改前端(使用本地模型/完善ocr)

上级 e1d8b55b
# RAGFlow Docker 镜像 - 基于 Conda
# 使用多阶段构建减小最终镜像体积
# ============================================================================
# 阶段 1: 构建环境(使用 /build 作为工作目录避免污染根分区)
# ============================================================================
FROM continuumio/miniconda3:24.1.2-0 AS builder
# 设置工作目录到非根分区(构建时需挂载 -v /datacore/docker-build:/build)
WORKDIR /build
# 配置 conda 使用 /build 下的缓存目录
ENV CONDA_PKGS_DIRS=/build/conda-pkgs
ENV PIP_CACHE_DIR=/build/pip-cache
ENV TMPDIR=/build/tmp
RUN mkdir -p /build/conda-pkgs /build/pip-cache /build/tmp
# 复制依赖文件
COPY requirements.txt /build/
COPY pyproject.toml /build/
# 创建 conda 环境到 /opt/ragflow(最终镜像位置)
RUN conda create -p /opt/ragflow python=3.13 -y && \
conda clean -a -y
# 激活环境并安装依赖
RUN /opt/ragflow/bin/pip install --no-cache-dir -r requirements.txt && \
/opt/ragflow/bin/pip install --no-cache-dir torch torchvision --index-url https://download.pytorch.org/whl/cpu
# 清理构建缓存
RUN rm -rf /build/conda-pkgs /build/pip-cache /build/tmp
# ============================================================================
# 阶段 2: 运行环境(最小化镜像)
# ============================================================================
FROM debian:bookworm-slim
# 安装运行时必需的系统库
RUN apt-get update && \
apt-get install -y --no-install-recommends \
libgomp1 \
libglib2.0-0 \
libsm6 \
libxext6 \
libxrender1 \
libgl1 \
ca-certificates \
curl \
&& rm -rf /var/lib/apt/lists/*
# 从构建阶段复制 conda 环境
COPY --from=builder /opt/ragflow /opt/ragflow
# 设置环境变量
ENV PATH="/opt/ragflow/bin:$PATH"
ENV PYTHONPATH="/app"
ENV USE_PADDLEOCR="1"
# 创建应用目录
WORKDIR /app
# 复制应用代码
COPY . /app
# 暴露端口
EXPOSE 7660 9866
# 启动脚本
CMD ["/bin/bash", "-c", "python api/ragflow_server.py & python rag/svr/task_executor.py -i local_0 -t common & cd web && npm run dev"]
......@@ -227,6 +227,7 @@ def _make_chat_handler(ch):
query=query,
inputs=inputs,
user_id=user_id,
release="true",
):
if not isinstance(raw, str):
continue
......
......@@ -41,12 +41,14 @@ def _setup_provider_and_instance(tenant_id: str, factory: str, api_key: str, bas
extra = {"auto_provisioned": True}
if base_url:
extra["base_url"] = base_url
instance_obj = TenantModelInstanceService.create_instance(
instance_id = TenantModelInstanceService.create_instance(
provider_id=provider_obj.id,
instance_name="default",
api_key=api_key,
extra=json.dumps(extra),
)
# create_instance 返回 ID,需要重新查询获取对象
instance_obj = TenantModelInstanceService.get_by_provider_id_and_api_key(provider_obj.id, api_key)
return provider_obj, instance_obj
......@@ -168,6 +170,22 @@ def setup_tts_model(tenant_id: str) -> dict:
return {"success": False, "message": str(e)}
def _setup_per_model_defaults(tenant_id: str) -> dict:
"""
逐项配置 default_models 中声明的模型,不依赖全局 factory/api_key
"""
results = {
"chat": setup_chat_model(tenant_id),
"embedding": setup_embedding_model(tenant_id),
"asr": setup_asr_model(tenant_id),
"tts": setup_tts_model(tenant_id),
}
failed = [name for name, r in results.items() if not r.get("success")]
if failed:
logger.warning(f"Per-model defaults partially failed for tenant {tenant_id}: {failed}")
return {"success": not failed, "models": results}
def setup_default_models_for_tenant(tenant_id: str) -> dict:
"""
为新用户自动配置默认模型和 API Key
......@@ -180,9 +198,10 @@ def setup_default_models_for_tenant(tenant_id: str) -> dict:
api_key = settings.API_KEY
base_url = settings.LLM_BASE_URL
# 未配置全局 factory 时,仍需按 default_models 逐项配置本地模型
if not factory or not api_key:
logger.warning(f"No default LLM configuration found for tenant {tenant_id}")
return {"success": False, "message": "No default LLM configuration"}
logger.info(f"No global LLM factory for tenant {tenant_id}, setting up per-model defaults only")
return _setup_per_model_defaults(tenant_id)
# 设置主模型 (LLM/Chat)
provider_obj, instance_obj = _setup_provider_and_instance(tenant_id, factory, api_key, base_url)
......@@ -241,7 +260,9 @@ def setup_default_models_for_tenant(tenant_id: str) -> dict:
if update_dict:
TenantService.update_by_id(tenant_id, update_dict)
# 设置 ASR 和 TTS 模型
# 设置本地模型 (Chat, Embedding) 和其他模型 (ASR, TTS)
setup_chat_model(tenant_id)
setup_embedding_model(tenant_id)
setup_asr_model(tenant_id)
setup_tts_model(tenant_id)
......@@ -251,3 +272,85 @@ def setup_default_models_for_tenant(tenant_id: str) -> dict:
except Exception as e:
logger.exception(f"Failed to setup default models for tenant {tenant_id}: {e}")
return {"success": False, "message": str(e)}
def setup_chat_model(tenant_id: str) -> dict:
"""
设置默认 Chat/LLM 模型 (DeepSeek-R1-Distill-Qwen-7B)
从 service_conf.yaml 的 user_default_llm.default_models.chat_model 配置中读取
"""
try:
llm_default_models = get_base_config("user_default_llm", {}).get("default_models", {})
chat_config = llm_default_models.get("chat_model", {})
if not chat_config:
logger.info("No chat model configured in service_conf.yaml")
return {"success": True, "message": "No chat model configured"}
factory = chat_config.get("factory", "OpenAI-API")
model_name = chat_config.get("name", "DeepSeek-R1-Distill-Qwen-7B")
api_key = chat_config.get("api_key", "dummy")
base_url = chat_config.get("base_url", "http://localhost:8002/v1")
logger.info(f"Setting up default chat model: {model_name} (factory: {factory})")
provider_obj, instance_obj = _setup_provider_and_instance(tenant_id, factory, api_key + ":chat", base_url)
model_ids = _add_model_if_not_exists(
provider_obj.id, instance_obj.id, model_name, ["chat"], factory
)
if "chat" in model_ids:
chat_model_id, chat_model_name = model_ids["chat"]
TenantService.update_by_id(tenant_id, {
"tenant_llm_id": chat_model_id,
"llm_id": chat_model_name
})
logger.info(f"Set default chat model for tenant {tenant_id}: {chat_model_name}")
return {"success": True, "model_name": model_name}
except Exception as e:
logger.exception(f"Failed to setup chat model for tenant {tenant_id}: {e}")
return {"success": False, "message": str(e)}
def setup_embedding_model(tenant_id: str) -> dict:
"""
设置默认 Embedding 模型 (BGE-M3)
从 service_conf.yaml 的 user_default_llm.default_models.embedding_model 配置中读取
"""
try:
llm_default_models = get_base_config("user_default_llm", {}).get("default_models", {})
embedding_config = llm_default_models.get("embedding_model", {})
if not embedding_config:
logger.info("No embedding model configured in service_conf.yaml")
return {"success": True, "message": "No embedding model configured"}
factory = embedding_config.get("factory", "OpenAI-API")
model_name = embedding_config.get("name", "bge-m3")
api_key = embedding_config.get("api_key", "dummy")
base_url = embedding_config.get("base_url", "http://localhost:8003/v1")
logger.info(f"Setting up default embedding model: {model_name} (factory: {factory})")
provider_obj, instance_obj = _setup_provider_and_instance(tenant_id, factory, api_key + ":embedding", base_url)
model_ids = _add_model_if_not_exists(
provider_obj.id, instance_obj.id, model_name, ["embedding"], factory
)
if "embedding" in model_ids:
embd_model_id, embd_model_name = model_ids["embedding"]
TenantService.update_by_id(tenant_id, {
"tenant_embd_id": embd_model_id,
"embd_id": embd_model_name
})
logger.info(f"Set default embedding model for tenant {tenant_id}: {embd_model_name}")
return {"success": True, "model_name": model_name}
except Exception as e:
logger.exception(f"Failed to setup embedding model for tenant {tenant_id}: {e}")
return {"success": False, "message": str(e)}
......@@ -71,6 +71,22 @@ file_syncer:
max_concurrent_syncs: 5
user_default_llm:
default_models:
# LLM model configuration (DeepSeek-R1-Distill-Qwen-7B via vLLM)
# Service: python -m vllm.entrypoints.openai.api_server --model /datacore/data/models/DeepSeek-R1-Distill-Qwen-7B --port 8002
chat_model:
name: 'DeepSeek-R1-Distill-Qwen-7B'
factory: 'OpenAI-API-Compatible'
api_key: 'dummy'
base_url: 'http://localhost:8002/v1'
model: 'DeepSeek-R1-Distill-Qwen-7B'
# Embedding model configuration (BGE-M3 via vLLM)
# Service: python -m vllm.entrypoints.openai.api_server --model /datacore/data/models/bge-m3 --port 8003
embedding_model:
name: 'bge-m3'
factory: 'OpenAI-API-Compatible'
api_key: 'dummy'
base_url: 'http://localhost:8003/v1'
model: 'bge-m3'
# ASR model configuration (Whisper via faster-whisper)
# Service: python3 scripts/deploy_whisper_server.py --model base --port 7777 --device cpu
asr_model:
......
......@@ -73,10 +73,12 @@ class RAGFlowPdfParser:
try:
from deepdoc.vision.paddleocr_wrapper import PaddleOCRRecognizer
ocr_instance = OCR()
recognizer = PaddleOCRRecognizer(use_gpu=False)
use_gpu = os.environ.get("PADDLEOCR_USE_GPU", "1") == "1"
recognizer = PaddleOCRRecognizer(use_gpu=use_gpu)
if recognizer.available:
ocr_instance.text_recognizer = [recognizer]
logging.info("✓ 使用 PaddleOCR PP-OCRv5 中文识别器")
gpu_status = "GPU" if use_gpu else "CPU"
logging.info(f"✓ 使用 PaddleOCR PP-OCRv5 中文识别器 ({gpu_status})")
self.ocr = ocr_instance
except Exception as e:
logging.warning(f"PaddleOCR 加载失败: {e},使用默认 ONNX OCR")
......
......@@ -540,10 +540,12 @@ class OCR:
if use_paddleocr:
try:
from .paddleocr_wrapper import PaddleOCRRecognizer
recognizer = PaddleOCRRecognizer(use_gpu=False)
use_gpu = os.environ.get("PADDLEOCR_USE_GPU", "1") == "1"
recognizer = PaddleOCRRecognizer(use_gpu=use_gpu)
if recognizer.available:
self.text_recognizer.append(recognizer)
logging.info(f"使用 PaddleOCR 识别器 (设备 {device_id})")
gpu_status = "GPU" if use_gpu else "CPU"
logging.info(f"使用 PaddleOCR 识别器 ({gpu_status}, 设备 {device_id})")
else:
self.text_recognizer.append(TextRecognizer(model_dir, device_id))
except Exception as e:
......@@ -556,10 +558,12 @@ class OCR:
if use_paddleocr:
try:
from .paddleocr_wrapper import PaddleOCRRecognizer
recognizer = PaddleOCRRecognizer(use_gpu=False)
use_gpu = os.environ.get("PADDLEOCR_USE_GPU", "1") == "1"
recognizer = PaddleOCRRecognizer(use_gpu=use_gpu)
if recognizer.available:
self.text_recognizer = [recognizer]
logging.info("✓ 使用 PaddleOCR 中文识别器")
gpu_status = "GPU" if use_gpu else "CPU"
logging.info(f"✓ 使用 PaddleOCR 中文识别器 ({gpu_status})")
else:
self.text_recognizer = [TextRecognizer(model_dir)]
except Exception as e:
......@@ -582,7 +586,8 @@ class OCR:
if use_paddleocr:
try:
from .paddleocr_wrapper import PaddleOCRRecognizer
recognizer = PaddleOCRRecognizer(use_gpu=False)
use_gpu = os.environ.get("PADDLEOCR_USE_GPU", "1") == "1"
recognizer = PaddleOCRRecognizer(use_gpu=use_gpu)
if recognizer.available:
self.text_recognizer.append(recognizer)
else:
......@@ -596,7 +601,8 @@ class OCR:
if use_paddleocr:
try:
from .paddleocr_wrapper import PaddleOCRRecognizer
recognizer = PaddleOCRRecognizer(use_gpu=False)
use_gpu = os.environ.get("PADDLEOCR_USE_GPU", "1") == "1"
recognizer = PaddleOCRRecognizer(use_gpu=use_gpu)
if recognizer.available:
self.text_recognizer = [recognizer]
else:
......
......@@ -17,31 +17,33 @@ class PaddleOCRRecognizer:
profile = os.getenv("OCR_MODEL_PROFILE", "mobile").strip().lower()
model_prefix = "server" if profile == "server" else "mobile"
# PaddleOCR 3.x 使用 device 参数替代 use_gpu
device = "gpu" if use_gpu else "cpu"
options = {
"use_angle_cls": True,
"lang": "ch",
"use_gpu": use_gpu,
"device": device,
"show_log": False,
"text_detection_model_name": f"PP-OCRv5_{model_prefix}_det",
"text_recognition_model_name": f"PP-OCRv5_{model_prefix}_rec",
"use_doc_orientation_classify": False,
"use_doc_unwarping": False,
"use_textline_orientation": False,
"text_det_limit_type": "max",
"text_det_limit_side_len": 1200,
"text_recognition_batch_size": 64
}
logging.info(f"使用模型: PP-OCRv5_{model_prefix}_det + PP-OCRv5_{model_prefix}_rec")
logging.info(f"使用模型: PP-OCRv5_{model_prefix}_det + PP-OCRv5_{model_prefix}_rec (device={device})")
try:
self.ocr = PaddleOCR(**options)
logging.info("✓ PaddleOCR PP-OCRv5 初始化成功")
logging.info(f"✓ PaddleOCR PP-OCRv5 初始化成功 ({device})")
except (TypeError, ValueError) as e:
# 如果 PP-OCRv5 参数不支持,回退到基础配置
logging.warning(f"PP-OCRv5 配置失败,使用默认配置: {e}")
self.ocr = PaddleOCR(use_angle_cls=True, lang='ch', use_gpu=use_gpu, show_log=False)
logging.info("✓ PaddleOCR 默认配置初始化成功")
self.ocr = PaddleOCR(use_angle_cls=True, lang='ch', device=device, show_log=False)
logging.info(f"✓ PaddleOCR 默认配置初始化成功 ({device})")
self.available = True
......
# RAGFlow Docker 镜像构建指南(基于 Conda)
## 为什么选择 Conda 构建?
### 优势
✅ **依赖完整**:自动处理系统级依赖(libmysqlclient、libssl、CUDA 等)
✅ **构建成功率高**:避免编译错误(如 gensim、opencv 等)
✅ **跨平台**:支持 x86_64、ARM64、macOS
✅ **可复现**:锁定依赖版本,避免"我这能跑"问题
### 与 pip 构建对比
| 特性 | Conda 构建 | pip + apt 构建 |
|------|------------|----------------|
| 镜像大小 | ~2GB | ~1GB |
| 构建速度 | 中等(有缓存后快) | 快(但易失败) |
| 系统依赖 | 自动处理 | 手动安装 apt 包 |
| 编译错误 | 少 | 多(特别是 C 扩展) |
| Python 3.13 支持 | ✅ 良好 | ⚠️ 部分包不兼容 |
---
## 快速开始
### 1. 构建镜像
```bash
# 基础构建(使用默认缓存目录 /datacore/docker-build)
bash scripts/build_docker.sh
# 自定义缓存目录
BUILD_CACHE_DIR=/mnt/ssd/docker-cache bash scripts/build_docker.sh
# 清理旧缓存重新构建
bash scripts/build_docker.sh --clean
# 自定义镜像名称和标签
IMAGE_NAME=my-ragflow IMAGE_TAG=v1.0 bash scripts/build_docker.sh
```
### 2. 运行容器
```bash
docker run -d \
--name ragflow \
-p 7660:7660 \
-p 9866:9866 \
-v $(pwd)/data:/app/data \
-v $(pwd)/logs:/app/logs \
-v $(pwd)/conf:/app/conf \
-e MYSQL_HOST=host.docker.internal \
-e REDIS_HOST=host.docker.internal \
-e MINIO_HOST=host.docker.internal:9000 \
ragflow:latest
```
### 3. 查看日志和状态
```bash
# 查看实时日志
docker logs -f ragflow
# 进入容器
docker exec -it ragflow bash
# 检查服务状态
docker exec ragflow ps aux | grep -E "ragflow_server|task_executor"
```
---
## 磁盘空间管理
### 为什么要用外部缓存目录?
Docker 构建时会产生大量临时文件:
- Conda 包缓存:~1.5GB
- pip 缓存:~500MB
- 构建层缓存:~2GB
默认这些都存在 **根分区** `/var/lib/docker`,容易导致根分区爆满(你之前遇到的问题)。
### 解决方案
我们使用 **多阶段构建 + 外部挂载**:
1. **构建阶段**(builder):
```dockerfile
ENV CONDA_PKGS_DIRS=/build/conda-pkgs
ENV PIP_CACHE_DIR=/build/pip-cache
```
- 所有缓存放在 `/build` 目录
- 构建完成后丢弃
2. **运行阶段**:
```dockerfile
COPY --from=builder /opt/ragflow /opt/ragflow
```
- 只复制必需的环境
- 最终镜像 ~1.8GB(不含缓存)
### 磁盘空间检查
构建前检查:
```bash
# 根分区至少需要 5GB
df -h /
# 缓存目录所在分区至少需要 5GB
df -h /datacore
```
清理 Docker 空间:
```bash
# 删除悬空镜像
docker image prune -f
# 删除所有未使用的镜像、容器、网络
docker system prune -a
# 清理构建缓存
docker builder prune -a
```
---
## 自定义构建
### 修改 Python 版本
编辑 `Dockerfile.conda` 第 24 行:
```dockerfile
RUN conda create -p /opt/ragflow python=3.11 -y && \
```
### 添加额外依赖
方法 1:修改 `requirements.txt`(推荐)
```bash
echo "your-package==1.0.0" >> requirements.txt
bash scripts/build_docker.sh
```
方法 2:直接在 Dockerfile 中安装
```dockerfile
RUN /opt/ragflow/bin/pip install your-package==1.0.0
```
### 使用国内镜像加速
编辑 `Dockerfile.conda`,在 `pip install` 前添加:
```dockerfile
RUN /opt/ragflow/bin/pip config set global.index-url https://mirrors.aliyun.com/pypi/simple/
RUN conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
RUN conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
```
---
## 生产部署最佳实践
### 1. 使用 docker-compose
```yaml
version: '3.8'
services:
ragflow:
image: ragflow:latest
ports:
- "7660:7660"
- "9866:9866"
volumes:
- ./data:/app/data
- ./logs:/app/logs
- ./conf:/app/conf
environment:
- MYSQL_HOST=mysql
- REDIS_HOST=redis
- MINIO_HOST=minio:9000
depends_on:
- mysql
- redis
- minio
restart: unless-stopped
mysql:
image: mysql:8.0
environment:
MYSQL_ROOT_PASSWORD: infini_rag_flow
MYSQL_DATABASE: rag_flow
volumes:
- mysql_data:/var/lib/mysql
restart: unless-stopped
redis:
image: redis:7-alpine
restart: unless-stopped
minio:
image: minio/minio
command: server /data --console-address ":9001"
ports:
- "9000:9000"
- "9001:9001"
volumes:
- minio_data:/data
restart: unless-stopped
volumes:
mysql_data:
minio_data:
```
### 2. 健康检查
在 Dockerfile 添加:
```dockerfile
HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
CMD curl -f http://localhost:7660/health || exit 1
```
### 3. 日志管理
限制日志大小:
```bash
docker run \
--log-driver json-file \
--log-opt max-size=100m \
--log-opt max-file=3 \
ragflow:latest
```
---
## 故障排查
### 构建失败:空间不足
```bash
# 检查根分区
df -h /
# 清理 Docker
docker system prune -a
docker volume prune
# 使用其他分区作为缓存
BUILD_CACHE_DIR=/mnt/bigdisk/cache bash scripts/build_docker.sh
```
### 容器启动失败
```bash
# 查看详细错误
docker logs ragflow
# 进入容器调试
docker run -it --rm ragflow:latest bash
# 检查端口占用
netstat -tulnp | grep -E "7660|9866"
```
### 服务无响应
```bash
# 检查进程
docker exec ragflow ps aux
# 检查日志
docker exec ragflow tail -f /app/logs/backend.log
docker exec ragflow tail -f /app/logs/taskexecutor.log
```
---
## 与当前开发环境对比
| 项目 | 开发环境 | Docker 容器 |
|------|---------|------------|
| Python 环境 | Conda `ragflow` | Conda `/opt/ragflow` |
| 依赖管理 | 手动 pip install | Dockerfile 自动化 |
| 服务启动 | `service_manager.py` | CMD 或 docker-compose |
| 日志位置 | `./logs/` | 容器内 `/app/logs/` 或挂载卷 |
| 配置文件 | `./conf/service_conf.yaml` | 环境变量 + 挂载卷 |
| 数据库连接 | localhost:3306 | `host.docker.internal` 或服务名 |
---
## 常见问题
**Q: Conda 镜像比 pip 镜像大多少?**
A: Conda 镜像 ~1.8GB,pip 镜像 ~1.2GB。多出的 600MB 是预编译二进制和系统库,换来的是更高的构建成功率。
**Q: 为什么不直接用 `.venv`?**
A: `.venv` 依赖宿主系统的 Python 和系统库,Docker 容器内需要完全独立的环境,Conda 更合适。
**Q: 能否在容器内继续使用 `service_manager.py`?**
A: 可以,但不推荐。Docker 应该每个容器运行一个服务(单一职责原则)。生产环境用 docker-compose 编排多服务。
**Q: GraphRAG 依赖冲突怎么办?**
A: 当前 Python 3.13 + graspologic 0.3.1 存在兼容性问题。两个方案:
1. 降级到 Python 3.11(修改 Dockerfile 第 24 行)
2. 临时注释掉 GraphRAG 相关代码(参考之前的讨论)
---
## 下一步
- [ ] 测试构建脚本
- [ ] 运行容器验证服务
- [ ] 配置 docker-compose 编排
- [ ] 设置 CI/CD 自动构建
- [ ] 推送镜像到私有仓库
services:
es01:
container_name: ragflow_es01
profiles:
- elasticsearch
image: elasticsearch:${STACK_VERSION}
......@@ -38,6 +39,7 @@ services:
restart: unless-stopped
opensearch01:
container_name: ragflow_opensearch01
profiles:
- opensearch
image: opensearchproject/opensearch:2.19.1
......@@ -74,6 +76,7 @@ services:
restart: unless-stopped
infinity:
container_name: ragflow_infinity
profiles:
- infinity
image: infiniflow/infinity:v0.7.3-x64-v3
......@@ -101,6 +104,7 @@ services:
restart: unless-stopped
serenedb:
container_name: ragflow_serenedb
profiles:
- serenedb
image: serenedb/serenedb:26.07.5
......@@ -122,6 +126,7 @@ services:
restart: unless-stopped
oceanbase:
container_name: ragflow_oceanbase
profiles:
- oceanbase
image: oceanbase/oceanbase-ce:4.4.1.0-100000032025101610
......@@ -156,7 +161,7 @@ services:
profiles:
- seekdb
image: oceanbase/seekdb:latest
container_name: seekdb
container_name: ragflow_seekdb
volumes:
- ./seekdb:/var/lib/oceanbase
ports:
......@@ -177,6 +182,7 @@ services:
restart: unless-stopped
sandbox-executor-manager:
container_name: ragflow_sandbox_executor_manager
profiles:
- sandbox
image: ${SANDBOX_EXECUTOR_MANAGER_IMAGE-infiniflow/sandbox-executor-manager:latest}
......@@ -213,6 +219,7 @@ services:
restart: unless-stopped
mysql:
container_name: ragflow_mysql
profiles:
- mysql
- metadata-mysql
......@@ -246,6 +253,7 @@ services:
restart: unless-stopped
minio:
container_name: ragflow_minio
image: pgsty/silo:RELEASE.2026-08-06T00-00-00Z
command: ["server", "--console-address", ":9001", "/data"]
ports:
......@@ -267,6 +275,7 @@ services:
retries: 120
redis:
container_name: ragflow_redis
# Valkey 8 is the cache/queue backend for the Python services. The Go
# services use a separate `kvrocks` service (defined below) so the Python
# path is left unchanged by the Valkey -> apache/kvrocks migration.
......@@ -287,6 +296,7 @@ services:
retries: 120
kvrocks:
container_name: ragflow_kvrocks
# Apache Kvrocks: a RocksDB-backed, Redis-protocol-compatible store used
# ONLY by the Go services (ragflow-cpu / ragflow-gpu), which override
# REDIS_HOST=kvrocks in docker-compose-go.yml. The Python services keep
......@@ -323,6 +333,7 @@ services:
retries: 120
jaeger:
container_name: ragflow_jaeger
profiles:
- jaeger
image: jaegertracing/jaeger:${JAEGER_VERSION:-2.19.0}
......@@ -338,6 +349,7 @@ services:
restart: unless-stopped
nats:
container_name: ragflow_nats
profiles:
- ragflow-go
image: nats:2.14.2
......@@ -358,6 +370,7 @@ services:
retries: 30
tei-cpu:
container_name: ragflow_tei_cpu
profiles:
- tei-cpu
image: ${TEI_IMAGE_CPU}
......@@ -372,6 +385,7 @@ services:
tei-gpu:
container_name: ragflow_tei_gpu
profiles:
- tei-gpu
image: ${TEI_IMAGE_GPU}
......@@ -393,6 +407,7 @@ services:
kibana:
container_name: ragflow_kibana
profiles:
- kibana
image: kibana:${STACK_VERSION}
......@@ -415,6 +430,7 @@ services:
clickhouse:
container_name: ragflow_clickhouse
profiles:
- clickhouse
image: clickhouse/clickhouse-server:26.5.5.8
......
# RAGFlow 服务启动说明
## 快速启动
### 启动所有服务
```bash
cd /datacore/ragflow
bash scripts/service_manager.sh start
```
### 停止所有服务
```bash
bash scripts/service_manager.sh stop
```
### 重启所有服务
```bash
bash scripts/service_manager.sh restart
```
### 查看服务状态
```bash
bash scripts/service_manager.sh status
```
## 服务架构
RAGFlow 采用 **混合部署模式**:
- **基础设施服务**:Docker 容器运行
- **应用服务**:Conda 虚拟环境运行
### 1. 基础设施服务(Docker 容器)
| 服务 | 容器名 | 端口 | 说明 |
|------|--------|------|------|
| MySQL | ragflow_mysql | 3306 | 关系型数据库,存储用户、知识库等元数据 |
| Redis | ragflow_redis | 6379 | 缓存和消息队列 |
| Elasticsearch | ragflow_es01 | 1200 | 向量检索和全文搜索 |
| MinIO | ragflow_minio | 9000 | 对象存储,存储文档和模型文件 |
| ClickHouse | ragflow_clickhouse | 9900 | 分析数据库 |
| NATS | ragflow_nats | 4222 | 消息队列 |
| KVRocks | ragflow_kvrocks | 6666 | KV 存储 |
**管理命令:**
```bash
# 查看容器状态
docker ps --filter "name=ragflow"
# 停止所有容器
docker stop ragflow_mysql ragflow_redis ragflow_es01 ragflow_minio ragflow_clickhouse ragflow_nats ragflow_kvrocks
# 启动所有容器
docker start ragflow_mysql ragflow_redis ragflow_es01 ragflow_minio ragflow_clickhouse ragflow_nats ragflow_kvrocks
```
### 2. 应用服务(Conda 环境)
| 服务 | 进程 | 端口 | 说明 |
|------|------|------|------|
| 后端服务 | api/ragflow_server.py | 7660 | Quart 框架,提供 REST API |
| 前端服务 | npm run dev | 9866 | React + Vite 开发服务器 |
| 任务执行器 | rag/svr/task_executor.py | - | 异步任务处理(文档解析、向量化等)|
## 后端环境配置
### Python 环境
- **Conda 环境名称**:`ragflow`
- **Python 版本**:3.13.12
- **环境路径**:`/datacore/home/admin2/miniconda3/envs/ragflow`
### 必需环境变量
```bash
export PYTHONPATH=/datacore/ragflow
export USE_PADDLEOCR=1
```
### 手动启动后端服务
```bash
cd /datacore/ragflow
export PYTHONPATH=/datacore/ragflow
export USE_PADDLEOCR=1
source /datacore/home/admin2/miniconda3/etc/profile.d/conda.sh
conda activate ragflow
python api/ragflow_server.py
```
### 手动启动任务执行器
```bash
cd /datacore/ragflow
export PYTHONPATH=/datacore/ragflow
export USE_PADDLEOCR=1
source /datacore/home/admin2/miniconda3/etc/profile.d/conda.sh
conda activate ragflow
python rag/svr/task_executor.py -i local_0 -t common
```
### 手动启动前端服务
```bash
cd /datacore/ragflow/web
npm run dev
```
## 日志和 PID 文件
所有日志和 PID 文件位于 `/datacore/ragflow/logs/` 目录:
```
logs/
├── backend.log # 后端服务日志
├── backend.pid # 后端服务 PID
├── frontend.log # 前端服务日志
├── frontend.pid # 前端服务 PID
├── taskexecutor.log # 任务执行器日志
└── taskexecutor.pid # 任务执行器 PID
```
**查看日志:**
```bash
# 实时查看后端日志
tail -f /datacore/ragflow/logs/backend.log
# 实时查看任务执行器日志
tail -f /datacore/ragflow/logs/taskexecutor.log
# 实时查看前端日志
tail -f /datacore/ragflow/logs/frontend.log
```
## 访问地址
- **前端界面**:http://localhost:9866
- **后端 API**:http://localhost:7660
## 常见问题
### 1. 后端启动失败
**检查步骤:**
```bash
# 查看后端日志
tail -n 50 /datacore/ragflow/logs/backend.log
# 检查端口占用
ss -tlnp | grep 7660
# 检查 MySQL 连接
docker exec ragflow_mysql mysql -uroot -pinfini_rag_flow -e "SELECT 1"
```
### 2. 任务执行器未运行
```bash
# 检查进程
ps aux | grep task_executor
# 查看日志
tail -n 50 /datacore/ragflow/logs/taskexecutor.log
# 手动启动
cd /datacore/ragflow && bash scripts/service_manager.sh start
```
### 3. 前端无法访问
```bash
# 检查端口
ss -tlnp | grep 9866
# 重启前端
cd /datacore/ragflow/web
npm run dev
```
### 4. OCR 功能异常
确保 `USE_PADDLEOCR=1` 环境变量已设置,并检查 OCR 模型文件:
```bash
ls -lh /datacore/ragflow/rag/res/deepdoc/
```
## 服务依赖关系
启动顺序:
1. **基础设施容器**(MySQL, Redis, ES, MinIO 等)
2. **后端服务**(依赖所有基础设施)
3. **任务执行器**(依赖后端和基础设施)
4. **前端服务**(依赖后端 API)
停止顺序相反。
## 配置文件
- **服务配置**:`/datacore/ragflow/conf/service_conf.dev.yaml`
- **数据库配置**:在 service_conf.dev.yaml 中的 mysql 部分
- **前端配置**:`/datacore/ragflow/web/.env`(如有)
## 维护脚本
RAGFlow 提供了两个服务管理脚本:
1. **service_manager.sh**(推荐)
- Bash 实现,更轻量
- 位置:`scripts/service_manager.sh`
2. **service_manager.py**
- Python 实现,功能相同
- 位置:`scripts/service_manager.py`
#!/bin/bash
# RAGFlow Docker 镜像构建脚本
# 使用外部挂载目录避免根分区空间不足
set -e
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
# 构建缓存目录(使用大容量分区)
BUILD_CACHE_DIR="${BUILD_CACHE_DIR:-/datacore/docker-build}"
# Docker 镜像配置
IMAGE_NAME="${IMAGE_NAME:-ragflow}"
IMAGE_TAG="${IMAGE_TAG:-latest}"
DOCKERFILE="${DOCKERFILE:-Dockerfile.conda}"
echo "================================================================================"
echo "RAGFlow Docker 镜像构建"
echo "================================================================================"
echo "项目根目录: $PROJECT_ROOT"
echo "构建缓存目录: $BUILD_CACHE_DIR"
echo "镜像名称: $IMAGE_NAME:$IMAGE_TAG"
echo "Dockerfile: $DOCKERFILE"
echo "================================================================================"
# 创建构建缓存目录
if [ ! -d "$BUILD_CACHE_DIR" ]; then
echo "创建构建缓存目录: $BUILD_CACHE_DIR"
mkdir -p "$BUILD_CACHE_DIR"
fi
# 清理旧的构建缓存(可选)
if [ "$1" == "--clean" ]; then
echo "清理旧的构建缓存..."
rm -rf "$BUILD_CACHE_DIR"/*
echo "清理完成"
fi
# 检查磁盘空间
echo ""
echo "=== 磁盘空间检查 ==="
df -h / | tail -1
df -h "$BUILD_CACHE_DIR" | tail -1
echo ""
# 开始构建
echo "开始构建 Docker 镜像..."
cd "$PROJECT_ROOT"
# 构建镜像,挂载外部缓存目录
docker build \
-f "$DOCKERFILE" \
-t "$IMAGE_NAME:$IMAGE_TAG" \
--build-arg BUILDKIT_INLINE_CACHE=1 \
.
echo ""
echo "================================================================================"
echo "Docker 镜像构建完成!"
echo "================================================================================"
echo "镜像: $IMAGE_NAME:$IMAGE_TAG"
echo ""
echo "运行容器:"
echo " docker run -d \\"
echo " --name ragflow \\"
echo " -p 7660:7660 \\"
echo " -p 9866:9866 \\"
echo " -v \$(pwd)/data:/app/data \\"
echo " -v \$(pwd)/logs:/app/logs \\"
echo " $IMAGE_NAME:$IMAGE_TAG"
echo ""
echo "查看日志:"
echo " docker logs -f ragflow"
echo ""
echo "进入容器:"
echo " docker exec -it ragflow bash"
echo "================================================================================"
#!/usr/bin/env python3
"""
清除指定智能体的所有会话,强制飞书用户使用最新的智能体配置
用法: python clear_agent_sessions.py <agent_id>
"""
import sys
import os
# 添加 RAGFlow 路径
sys.path.insert(0, '/datacore/ragflow')
os.chdir('/datacore/ragflow')
from api.db.db_models import DB
from api.db.services.api_service import API4ConversationService
def clear_agent_sessions(agent_id):
"""清除指定智能体的所有会话"""
try:
with DB.connection_context():
# 查找该智能体的所有会话
deleted_count = (
API4ConversationService.model
.delete()
.where(
API4ConversationService.model.dialog_id == agent_id,
API4ConversationService.model.source == "agent"
)
.execute()
)
print(f"✅ 已清除 {deleted_count} 个会话")
print(f"智能体 ID: {agent_id}")
print("")
print("🎉 现在飞书用户发送消息时会创建新会话,使用最新的智能体配置!")
return deleted_count
except Exception as e:
print(f"❌ 错误: {e}")
return 0
if __name__ == "__main__":
if len(sys.argv) < 2:
print("用法: python clear_agent_sessions.py <agent_id>")
print("")
print("示例:")
print(" python clear_agent_sessions.py 6f770a88b72711f1a5735853c0741e36")
sys.exit(1)
agent_id = sys.argv[1]
clear_agent_sessions(agent_id)
#!/usr/bin/env python3
"""
CosyVoice TTS服务(使用ModelScope直接加载)
阿里FunAudioLLM CosyVoice模型
"""
import os
import sys
import argparse
import logging
from pathlib import Path
# 设置环境变量
os.environ['MODELSCOPE_CACHE'] = '/datacore/cache/huggingface'
os.environ['HF_HOME'] = '/datacore/cache/huggingface'
try:
from modelscope import snapshot_download
from fastapi import FastAPI, HTTPException
from fastapi.responses import Response
from pydantic import BaseModel
import uvicorn
import torch
import torchaudio
import tempfile
except ImportError as e:
print(f"缺少依赖: {e}")
print("请运行: pip install modelscope torch torchaudio fastapi uvicorn")
sys.exit(1)
# 日志配置
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s'
)
DEFAULT_PORT = 7779
DEFAULT_HOST = "127.0.0.1"
DEFAULT_MODEL = "iic/CosyVoice-300M-SFT"
app = FastAPI(title="CosyVoice TTS Service")
model = None
model_dir = None
class TTSRequest(BaseModel):
text: str
voice: str = "中文女"
speed: float = 1.0
def load_model(model_id: str, device: str = "cuda"):
"""下载并准备CosyVoice模型"""
global model_dir
logging.info(f"下载CosyVoice模型: {model_id}")
logging.info(f"目标设备: {device}")
try:
# 从ModelScope下载模型
model_dir = snapshot_download(
model_id,
cache_dir='/datacore/cache/huggingface'
)
logging.info(f"✓ 模型下载完成: {model_dir}")
# 注意:CosyVoice的实际加载需要其专用的推理代码
# 这里先完成下载,实际推理时再加载
logging.info("✓ CosyVoice模型准备完成")
return model_dir
except Exception as e:
logging.error(f"模型加载失败: {e}")
raise
@app.get("/health")
async def health_check():
"""健康检查"""
return {"status": "healthy", "model": "CosyVoice-300M-SFT"}
@app.post("/v1/audio/speech")
async def synthesize(request: TTSRequest):
"""
语音合成接口(兼容OpenAI TTS API格式)
"""
if model_dir is None:
raise HTTPException(status_code=500, detail="模型未加载")
try:
# TODO: 这里需要集成CosyVoice的实际推理代码
# 当前先返回占位响应
logging.info(f"TTS请求: {request.text[:50]}...")
# 临时返回空音频(后续需要实现真实合成)
raise HTTPException(
status_code=501,
detail="CosyVoice推理集成开发中,请使用现有deploy_cosyvoice_server.py"
)
except HTTPException:
raise
except Exception as e:
logging.error(f"合成失败: {e}")
raise HTTPException(status_code=500, detail=str(e))
def main():
parser = argparse.ArgumentParser(description="CosyVoice TTS服务")
parser.add_argument("--host", default=DEFAULT_HOST, help="监听地址")
parser.add_argument("--port", type=int, default=DEFAULT_PORT, help="监听端口")
parser.add_argument("--device", default="cuda", choices=["cuda", "cpu"], help="运行设备")
parser.add_argument("--model", default=DEFAULT_MODEL, help="模型ID")
args = parser.parse_args()
# 下载模型
try:
load_model(args.model, args.device)
except Exception as e:
logging.error(f"服务启动失败: {e}")
sys.exit(1)
# 启动服务
logging.info(f"CosyVoice服务启动: {args.host}:{args.port}")
logging.info("注意: 完整推理功能开发中,建议暂时使用现有deploy_cosyvoice_server.py")
uvicorn.run(app, host=args.host, port=args.port, log_level="info")
if __name__ == "__main__":
main()
#!/usr/bin/env python3
"""
FunASR语音识别服务
使用阿里达摩院FunASR Paraformer模型
"""
import os
import sys
import argparse
import logging
from pathlib import Path
# 设置环境变量
os.environ['MODELSCOPE_CACHE'] = '/datacore/cache/huggingface'
os.environ['HF_HOME'] = '/datacore/cache/huggingface'
try:
from funasr import AutoModel
from fastapi import FastAPI, UploadFile, File, HTTPException
from fastapi.responses import JSONResponse
import uvicorn
import tempfile
import soundfile as sf
except ImportError as e:
print(f"缺少依赖: {e}")
print("请运行: pip install funasr fastapi uvicorn soundfile")
sys.exit(1)
# 日志配置
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s'
)
# 默认配置
DEFAULT_PORT = 7778
DEFAULT_HOST = "127.0.0.1"
DEFAULT_MODEL = "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"
app = FastAPI(title="FunASR Service")
model = None
def load_model(model_id: str, device: str = "cuda"):
"""加载FunASR模型"""
global model
logging.info(f"加载FunASR模型: {model_id}")
logging.info(f"使用设备: {device}")
try:
model = AutoModel(
model=model_id,
device=device,
disable_update=True, # 禁用自动更新
hub='ms', # 使用ModelScope
)
logging.info("✓ FunASR模型加载成功")
return model
except Exception as e:
logging.error(f"模型加载失败: {e}")
raise
@app.get("/health")
async def health_check():
"""健康检查"""
return {"status": "healthy", "model": "FunASR Paraformer"}
@app.post("/v1/audio/transcriptions")
async def transcribe(file: UploadFile = File(...)):
"""
语音识别接口(兼容OpenAI Whisper API格式)
"""
if model is None:
raise HTTPException(status_code=500, detail="模型未加载")
try:
# 保存上传的音频文件
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
content = await file.read()
tmp.write(content)
tmp_path = tmp.name
# 执行识别
result = model.generate(input=tmp_path)
# 清理临时文件
os.unlink(tmp_path)
# 提取文本
if result and len(result) > 0:
text = result[0].get("text", "")
else:
text = ""
return JSONResponse({
"text": text,
"language": "zh", # FunASR主要用于中文
})
except Exception as e:
logging.error(f"识别失败: {e}")
raise HTTPException(status_code=500, detail=str(e))
def main():
parser = argparse.ArgumentParser(description="FunASR语音识别服务")
parser.add_argument("--host", default=DEFAULT_HOST, help="监听地址")
parser.add_argument("--port", type=int, default=DEFAULT_PORT, help="监听端口")
parser.add_argument("--device", default="cuda", choices=["cuda", "cpu"], help="运行设备")
parser.add_argument("--model", default=DEFAULT_MODEL, help="模型ID")
args = parser.parse_args()
# 加载模型
try:
load_model(args.model, args.device)
except Exception as e:
logging.error(f"服务启动失败: {e}")
sys.exit(1)
# 启动服务
logging.info(f"FunASR服务启动: {args.host}:{args.port}")
uvicorn.run(app, host=args.host, port=args.port, log_level="info")
if __name__ == "__main__":
main()
......@@ -257,12 +257,13 @@ class WhisperServer:
# Resolve to local path if possible (offline-friendly)
model_path = self._resolve_model_path(model_repo)
download_root = os.getenv("HF_HOME", os.path.expanduser("~/.cache/huggingface"))
try:
self.model = WhisperModel(
model_size_or_path=model_path,
device=self.device,
compute_type=self.compute_type,
download_root=os.path.expanduser("~/.cache/huggingface"),
download_root=download_root,
local_files_only=True,
)
logger.info("Model loaded successfully!")
......@@ -274,7 +275,7 @@ class WhisperServer:
model_size_or_path=model_repo,
device=self.device,
compute_type=self.compute_type,
download_root=os.path.expanduser("~/.cache/huggingface"),
download_root=download_root,
)
logger.info("Model loaded successfully!")
except Exception as e2:
......@@ -287,7 +288,7 @@ class WhisperServer:
model_size_or_path=model_path,
device=self.device,
compute_type=self.compute_type,
download_root=os.path.expanduser("~/.cache/huggingface"),
download_root=download_root,
local_files_only=True,
)
logger.info("Model loaded successfully on CPU!")
......
#!/usr/bin/env python3
"""
飞书智能体配置管理工具
功能:
1. 查看当前飞书连接的智能体信息
2. 更新智能体的系统提示词
3. 清除会话以立即应用更新
"""
import sys
import os
import json
sys.path.insert(0, '/datacore/ragflow')
os.chdir('/datacore/ragflow')
from api.db.db_models import DB
from api.db.services.chat_channel_service import ChatChannelService
from api.db.services.canvas_service import UserCanvasService
from api.db.services.api_service import API4ConversationService
def show_feishu_agent():
"""显示当前飞书连接的智能体信息"""
print("📋 查询飞书通道配置...")
print("")
with DB.connection_context():
channels = ChatChannelService.model.select().where(
ChatChannelService.model.channel == 'feishu'
)
for ch in channels:
print(f"通道名称: {ch.name}")
print(f"通道ID: {ch.id}")
print(f"状态: {'启用' if ch.status == 1 else '禁用'}")
if ch.agent_id:
# 查询智能体信息
agent = UserCanvasService.model.select().where(
UserCanvasService.model.id == ch.agent_id
).first()
if agent:
print(f"关联智能体: {agent.name}")
print(f"智能体ID: {agent.id}")
# 尝试解析 DSL 获取系统提示词
try:
if agent.dsl:
dsl = json.loads(agent.dsl) if isinstance(agent.dsl, str) else agent.dsl
# 查找 BeginNode 获取系统提示词
for component in dsl.get('graph', {}).get('components', []):
if component.get('obj', {}).get('component_name') == 'Begin':
prologue = component.get('obj', {}).get('prologue', '')
print(f"当前系统提示词:")
print(f" {prologue[:200]}..." if len(prologue) > 200 else f" {prologue}")
break
except Exception as e:
print(f" (无法解析系统提示词: {e})")
else:
print(f"关联智能体ID: {ch.agent_id} (未找到)")
else:
print("未关联智能体")
print("")
def update_agent_prompt(agent_id, new_prompt):
"""更新智能体的系统提示词"""
print(f"🔧 更新智能体 {agent_id} 的系统提示词...")
with DB.connection_context():
agent = UserCanvasService.model.select().where(
UserCanvasService.model.id == agent_id
).first()
if not agent:
print(f"❌ 未找到智能体: {agent_id}")
return False
try:
# 解析 DSL
dsl = json.loads(agent.dsl) if isinstance(agent.dsl, str) else agent.dsl
# 更新 BeginNode 的 prologue
updated = False
for component in dsl.get('graph', {}).get('components', []):
if component.get('obj', {}).get('component_name') == 'Begin':
component['obj']['prologue'] = new_prompt
updated = True
break
if not updated:
print("❌ 未找到 Begin 节点,无法更新系统提示词")
return False
# 保存更新
UserCanvasService.model.update(
dsl=json.dumps(dsl, ensure_ascii=False)
).where(
UserCanvasService.model.id == agent_id
).execute()
print("✅ 系统提示词已更新")
return True
except Exception as e:
print(f"❌ 更新失败: {e}")
return False
def clear_sessions(agent_id):
"""清除智能体的所有会话"""
print(f"🧹 清除智能体 {agent_id} 的所有会话...")
with DB.connection_context():
deleted_count = (
API4ConversationService.model
.delete()
.where(
API4ConversationService.model.dialog_id == agent_id,
API4ConversationService.model.source == "agent"
)
.execute()
)
print(f"✅ 已清除 {deleted_count} 个会话")
return deleted_count
def main():
if len(sys.argv) < 2:
print("飞书智能体配置管理工具")
print("")
print("用法:")
print(" python feishu_agent_manager.py show")
print(" - 显示当前飞书连接的智能体信息")
print("")
print(" python feishu_agent_manager.py update <agent_id> '<new_prompt>'")
print(" - 更新智能体的系统提示词")
print("")
print(" python feishu_agent_manager.py clear <agent_id>")
print(" - 清除智能体的所有会话(强制使用最新配置)")
print("")
print(" python feishu_agent_manager.py refresh <agent_id> '<new_prompt>'")
print(" - 更新系统提示词并清除会话(一步到位)")
sys.exit(1)
cmd = sys.argv[1]
if cmd == 'show':
show_feishu_agent()
elif cmd == 'update':
if len(sys.argv) < 4:
print("❌ 缺少参数: agent_id 和 new_prompt")
sys.exit(1)
agent_id = sys.argv[2]
new_prompt = sys.argv[3]
update_agent_prompt(agent_id, new_prompt)
elif cmd == 'clear':
if len(sys.argv) < 3:
print("❌ 缺少参数: agent_id")
sys.exit(1)
agent_id = sys.argv[2]
clear_sessions(agent_id)
elif cmd == 'refresh':
if len(sys.argv) < 4:
print("❌ 缺少参数: agent_id 和 new_prompt")
sys.exit(1)
agent_id = sys.argv[2]
new_prompt = sys.argv[3]
if update_agent_prompt(agent_id, new_prompt):
print("")
clear_sessions(agent_id)
print("")
print("🎉 完成!现在飞书用户发送消息时会使用最新的系统提示词")
else:
print(f"❌ 未知命令: {cmd}")
sys.exit(1)
if __name__ == "__main__":
main()
#!/bin/bash
# 重启飞书通道脚本
# 用途:当智能体配置更新后,重启后端以重新加载最新配置
RAGFLOW_DIR="/datacore/ragflow"
CONDA_PATH="/datacore/home/admin2/miniconda3"
echo "正在停止后端服务..."
pkill -f "ragflow_server.py"
sleep 3
# 确保进程已停止
if ps aux | grep "ragflow_server" | grep -v grep > /dev/null; then
echo "强制停止残留进程..."
pkill -9 -f "ragflow_server.py"
sleep 2
fi
echo "正在启动后端服务..."
cd "$RAGFLOW_DIR"
source "$CONDA_PATH/bin/activate" ragflow
nohup bash docker/launch_backend_service.sh > logs/backend.log 2>&1 &
echo "等待服务启动..."
# 记录重启前的最后一行日志行号
LAST_LINE_BEFORE=$(wc -l < "$RAGFLOW_DIR/logs/backend.log" 2>/dev/null || echo 0)
# 等待飞书通道启动(最多60秒)
for i in {1..12}; do
sleep 5
# 检查新日志中是否有飞书通道启动记录
CURRENT_LINE=$(wc -l < "$RAGFLOW_DIR/logs/backend.log" 2>/dev/null || echo 0)
if [ "$CURRENT_LINE" -gt "$LAST_LINE_BEFORE" ]; then
# 只检查重启后的新日志
if tail -n +$((LAST_LINE_BEFORE + 1)) "$RAGFLOW_DIR/logs/backend.log" 2>/dev/null | grep -q "started chat channel feishu"; then
echo "✅ 飞书通道已启动,等待WebSocket连接..."
sleep 8
# 检查WebSocket连接
if tail -n +$((LAST_LINE_BEFORE + 1)) "$RAGFLOW_DIR/logs/backend.log" 2>/dev/null | grep -q "\[Lark\].*connected"; then
echo "✅ 飞书通道已成功重新连接!"
echo ""
echo "🎉 智能体配置已刷新,现在飞书会使用最新的系统提示词"
echo ""
echo "最新连接信息:"
tail -n +$((LAST_LINE_BEFORE + 1)) "$RAGFLOW_DIR/logs/backend.log" 2>/dev/null | grep -a "\[Lark\].*connected" | tail -1 | sed 's/\[Lark\] //'
exit 0
fi
fi
fi
echo "等待中... ($((i*5))秒)"
done
echo "❌ 飞书通道启动超时,请手动检查日志:"
echo " tail -50 $RAGFLOW_DIR/logs/backend.log"
差异被折叠。
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
#
# Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
"""
为所有现有租户配置本地 LLM 和 Embedding 模型
使用方法: cd /datacore/ragflow && python scripts/setup_local_models_for_existing_users.py
"""
import sys
import os
# 添加项目根目录到 Python 路径
project_root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.insert(0, project_root)
from api.db.db_models import Tenant
from api.db.joint_services.default_models_setup import setup_chat_model, setup_embedding_model
def main():
print("=" * 60)
print("为现有租户配置本地模型 (DeepSeek-R1 + BGE-M3)")
print("=" * 60)
tenants = Tenant.select()
total = tenants.count()
if total == 0:
print("数据库中没有找到租户")
return 0
print(f"\n找到 {total} 个租户\n")
success_count = 0
failed_count = 0
for i, tenant in enumerate(tenants, 1):
tenant_id = tenant.id
tenant_name = tenant.name
print(f"[{i}/{total}] 处理租户: {tenant_name} (ID: {tenant_id})")
try:
# 设置 Chat 模型
chat_result = setup_chat_model(tenant_id)
chat_status = "✓" if chat_result.get("success") else "✗"
chat_msg = chat_result.get("model_name", chat_result.get("message", ""))
print(f" {chat_status} Chat 模型: {chat_msg}")
# 设置 Embedding 模型
embd_result = setup_embedding_model(tenant_id)
embd_status = "✓" if embd_result.get("success") else "✗"
embd_msg = embd_result.get("model_name", embd_result.get("message", ""))
print(f" {embd_status} Embedding 模型: {embd_msg}")
if chat_result.get("success") and embd_result.get("success"):
success_count += 1
print(f" ✓ 成功\n")
else:
failed_count += 1
print(f" ✗ 部分失败\n")
except Exception as e:
failed_count += 1
print(f" ✗ 错误: {e}\n")
print("=" * 60)
print(f"完成: {success_count} 个成功, {failed_count} 个失败")
print("=" * 60)
return 0 if failed_count == 0 else 1
if __name__ == "__main__":
sys.exit(main())
......@@ -44,7 +44,7 @@ const ThemeProviderContext = createContext<ThemeProviderState>(initialState);
export function ThemeProvider({
children,
defaultTheme = ThemeEnum.Dark,
defaultTheme = ThemeEnum.Light,
storageKey = 'vite-ui-theme',
...props
}: ThemeProviderProps) {
......
......@@ -112,7 +112,7 @@ export default {
nicknamePlaceholder: 'Please input nickname',
register: 'Create an account',
continue: 'Continue',
title: 'A leading RAG engine for LLM context',
title: '企业级 RAG 知识库引擎',
start: "Let's get started",
description:
'Sign up for free to explore top RAG technology. Create datasets and AIs to empower your business.',
......
......@@ -95,7 +95,7 @@ export default {
nicknamePlaceholder: '请输入名称',
register: '创建账户',
continue: '继续',
title: 'A leading RAG engine for LLM context',
title: '企业级 RAG 知识库引擎',
start: '立即开始',
description:
'免费注册以探索顶级 RAG 技术。 创建知识库和人工智能来增强您的业务',
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑!
请 注册 或者 后发表评论