master
97a77d4745
1. 删除 ConfigAgent (与 MQTT 配置更新重叠且未启用) 2. OTA 完整安装流程: 解压tar.gz -> 校验ELF -> 替换二进制 -> systemctl重启 3. StreamManager stopAll 添加 Wait() 防止僵尸进程 4. InferClient socket 读写添加 timeout 防止永久阻塞 5. Heartbeat 集成 NPU 监控 (npu-smi + fallback 脚本) 6. 配置更新时动态重启 ffmpeg 以应用新 fps
289 lines
7.4 KiB
Markdown
289 lines
7.4 KiB
Markdown
# AI-tianyan 边缘部署操作指南
|
||
|
||
## 系统环境
|
||
|
||
| 项目 | 要求 |
|
||
|---|---|
|
||
| 硬件 | Atlas 200I A2 (Ascend 310B4) 或同系列设备 |
|
||
| OS | Ubuntu 22.04 aarch64 |
|
||
| Python | 3.9.x (Miniconda 或系统自带) |
|
||
| CANN | 6.2.RC2 (Ascend Toolkit V100R003C11) |
|
||
| Go | 1.18+ (仅编译时需要) |
|
||
| 云端 | 101.36.73.102 (API/MQTT/InfluxDB/OTA/ZLMediaKit) |
|
||
|
||
> 如果目标设备环境完全一致（同镜像版本），可直接复用已编译的 Go 二进制和 .om 模型，无需重新编译。
|
||
|
||
---
|
||
|
||
## 部署方式
|
||
|
||
### 方式一：离线包部署（推荐，最快）
|
||
|
||
适用于环境一致的设备，所有构建产物已包含在内。
|
||
|
||
#### 1. 在源设备打包
|
||
|
||
```bash
|
||
cd /root
|
||
tar czf ai-tianyan.tar.gz --exclude='.git' --exclude='__pycache__' --exclude='*.pyc' AI-tianyan/
|
||
```
|
||
|
||
#### 2. 传输到目标设备
|
||
|
||
```bash
|
||
# 网络传输
|
||
scp ai-tianyan.tar.gz root@目标IP:/root/
|
||
|
||
# 或 U盘拷贝
|
||
```
|
||
|
||
#### 3. 目标设备上安装
|
||
|
||
```bash
|
||
# 确保 pip 存在
|
||
python3 -m pip --version || apt update && apt install -y python3-pip
|
||
|
||
# 解压
|
||
cd /root
|
||
tar xzf ai-tianyan.tar.gz
|
||
cd AI-tianyan
|
||
|
||
# 一键安装（部署文件 + Python依赖 + systemd服务 + 自动启动）
|
||
bash scripts/install.sh
|
||
```
|
||
|
||
#### 4. 修改配置
|
||
|
||
```bash
|
||
# 生成新的设备 UUID
|
||
NEW_UUID=$(cat /proc/sys/kernel/random/uuid | tr -d '-')
|
||
|
||
# 更新配置
|
||
sudo sed -i "s/device_uuid:.*/device_uuid: $NEW_UUID/" /opt/tianyan-edge/config/edge.yaml
|
||
sudo sed -i "s/edge_id:.*/edge_id: edge-demo-002/" /opt/tianyan-edge/config/edge.yaml
|
||
|
||
# 如有不同的摄像头流地址
|
||
sudo vi /opt/tianyan-edge/config/edge.yaml
|
||
```
|
||
|
||
#### 5. 验证
|
||
|
||
```bash
|
||
sudo systemctl status edge-agent edge-infer
|
||
journalctl -u edge-agent -n 50 --no-pager
|
||
```
|
||
|
||
---
|
||
|
||
### 方式二：Git 源码部署
|
||
|
||
适用于需要自定义代码或环境有差异的设备。
|
||
|
||
#### 1. 拉取代码
|
||
|
||
```bash
|
||
git clone http://101.36.73.102:3112/fyah/AI-tianyan.git
|
||
cd AI-tianyan
|
||
```
|
||
|
||
#### 2. 编译 Go 边缘代理
|
||
|
||
```bash
|
||
export GOPROXY=https://goproxy.cn,direct
|
||
bash scripts/build.sh
|
||
# 输出: build/edge-agent
|
||
```
|
||
|
||
#### 3. 转换模型（仅 ATC 版本不同时需要）
|
||
|
||
```bash
|
||
source /usr/local/Ascend/ascend-toolkit/set_env.sh
|
||
bash scripts/atc_convert.sh
|
||
# 或手动:
|
||
atc --model=yolov8n.onnx \
|
||
--framework=5 \
|
||
--output=model/model.om \
|
||
--soc_version=Ascend310B4 \
|
||
--input_format=NCHW \
|
||
--input_shape="images:1,3,640,640" \
|
||
--output_type=FP16
|
||
```
|
||
|
||
#### 4. 安装部署
|
||
|
||
```bash
|
||
# 确保 pip 存在
|
||
python3 -m pip --version || apt update && apt install -y python3-pip
|
||
|
||
# 安装 Python 依赖
|
||
python3 -m pip install opencv-python-headless numpy
|
||
|
||
# 一键安装
|
||
bash scripts/install.sh
|
||
```
|
||
|
||
---
|
||
|
||
## 安装脚本说明
|
||
|
||
`scripts/install.sh` 会自动执行以下操作：
|
||
|
||
```
|
||
1. 创建目录: /opt/tianyan-edge/{bin,python,config,systemd,staging,model}
|
||
2. 复制 edge-agent -> /opt/tianyan-edge/bin/edge-agent
|
||
3. 复制 infer_server.py -> /opt/tianyan-edge/python/
|
||
4. 复制 edge.yaml.template -> /opt/tianyan-edge/config/edge.yaml (仅在不存在时)
|
||
5. 安装 Python 依赖 (opencv-python-headless, numpy)
|
||
6. 安装 systemd 服务文件
|
||
7. daemon-reload + enable + restart 服务
|
||
```
|
||
|
||
---
|
||
|
||
## 配置文件说明
|
||
|
||
安装后配置位于 `/opt/tianyan-edge/config/edge.yaml`：
|
||
|
||
```yaml
|
||
device_uuid: 8541db9f77826e39605ef2c032f8fb93 # 设备唯一标识，每台设备必须不同
|
||
edge_id: edge-demo-001 # 设备名称，便于识别
|
||
cloud_url: http://101.36.73.102:8004 # 云端 API 地址
|
||
mqtt_broker: tcp://101.36.73.102:1883 # MQTT Broker
|
||
mqtt_user: "" # MQTT 认证用户名
|
||
mqtt_pass: "" # MQTT 认证密码
|
||
edge_token: "" # 设备认证 Token
|
||
rtsp_urls:
|
||
- http://101.36.73.102:8080/rtp/xxx.live.flv # 视频流地址 (支持多路)
|
||
infer_socket: /tmp/edge-infer.sock # Go与Python通信的Unix Socket
|
||
infer_fps: 2 # 推理帧率 (每秒抽样数)
|
||
infer_workers: 3 # 并发推理 worker 数
|
||
conf_threshold: 0.2 # 检测置信度阈值
|
||
dedup_window_sec: 30 # 事件去重窗口 (秒)
|
||
ota_url: http://101.36.73.102:8087 # OTA 自动更新地址
|
||
version: 1.0.0 # 固件版本号
|
||
```
|
||
|
||
> 新设备部署时 **必须修改**: `device_uuid`、`edge_id`、`rtsp_urls`
|
||
|
||
---
|
||
|
||
## 服务管理
|
||
|
||
### 启动/停止/重启
|
||
|
||
```bash
|
||
sudo systemctl start edge-infer # 启动 NPU 推理服务
|
||
sudo systemctl start edge-agent # 启动边缘代理
|
||
sudo systemctl stop edge-infer edge-agent
|
||
sudo systemctl restart edge-infer edge-agent
|
||
```
|
||
|
||
### 开机自启
|
||
|
||
```bash
|
||
sudo systemctl enable edge-infer edge-agent
|
||
```
|
||
|
||
### 查看状态
|
||
|
||
```bash
|
||
sudo systemctl status edge-agent edge-infer
|
||
sudo journalctl -u edge-agent -f # 实时查看边缘代理日志
|
||
sudo journalctl -u edge-infer -f # 实时查看推理服务日志
|
||
```
|
||
|
||
### 手动调试模式
|
||
|
||
```bash
|
||
# 停止 systemd 服务
|
||
sudo systemctl stop edge-agent edge-infer
|
||
|
||
# 手动启动推理服务
|
||
bash -lc 'source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
|
||
NAMES_FILE=/opt/tianyan-edge/model/names.txt \
|
||
OM_MODEL=/opt/tianyan-edge/model/model.om \
|
||
CONF_THRESHOLD=0.2 \
|
||
OUTPUT_FORMAT=raw \
|
||
python3 /opt/tianyan-edge/python/infer_server.py'
|
||
|
||
# 另一个终端启动边缘代理
|
||
/opt/tianyan-edge/bin/edge-agent -config /opt/tianyan-edge/config/edge.yaml
|
||
```
|
||
|
||
---
|
||
|
||
## 运行验证
|
||
|
||
```bash
|
||
# 检查进程
|
||
ps aux | grep -E 'edge-agent|infer_server'
|
||
|
||
# 检查 NPU 状态
|
||
npu-smi info
|
||
|
||
# 检查网络连接 (应看到连接到 101.36.73.102)
|
||
ss -tunp | grep 101.36.73.102
|
||
|
||
# 检查 Unix Socket
|
||
ls -la /tmp/edge-infer.sock
|
||
|
||
# 检查模型文件
|
||
ls -lh /opt/tianyan-edge/model/model.om /opt/tianyan-edge/model/names.txt
|
||
|
||
# 查看系统日志
|
||
tail -f /opt/tianyan-edge/logs/agent.log 2>/dev/null
|
||
tail -f /opt/tianyan-edge/logs/infer.log 2>/dev/null
|
||
```
|
||
|
||
---
|
||
|
||
## Docker 部署（可选）
|
||
|
||
```bash
|
||
cd AI-tianyan
|
||
docker compose up -d
|
||
docker compose logs -f tianyan-edge
|
||
```
|
||
|
||
要求：Docker 已安装，CANN 驱动已就绪，NPU 设备节点 `/dev/davinci0` 等存在。
|
||
|
||
---
|
||
|
||
## 常见问题
|
||
|
||
| 问题 | 解决方法 |
|
||
|---|---|
|
||
| `acl.init failed` | 确认 `source set_env.sh` 已执行，检查 `LD_LIBRARY_PATH` 包含 CANN 路径 |
|
||
| `model.om not found` | 确认 `/opt/tianyan-edge/model/model.om` 存在，重新执行 ATC 转换 |
|
||
| `python3 -m pip: command not found` | `apt install -y python3-pip` |
|
||
| `opencv-python-headless 安装失败` | `apt install -y python3-dev gcc g++ libgl1-mesa-glx` 后再 pip install |
|
||
| 无法连接 MQTT | 检查 `mqtt_broker` 地址是否正确，确认网络可达 |
|
||
| 拉流失败 | 确认 ZLMediaKit 端口 8080 可达，流地址正确 |
|
||
| Go 编译失败 | `export GOPROXY=https://goproxy.cn,direct && go mod download` |
|
||
| systemd 启动后立即退出 | `journalctl -u edge-agent -n 100` 查看详细错误 |
|
||
| 多设备 UUID 冲突 | 每台设备必须有不同的 `device_uuid`，用 `cat /proc/sys/kernel/random/uuid` 生成 |
|
||
|
||
---
|
||
|
||
## 卸载
|
||
|
||
```bash
|
||
bash scripts/uninstall.sh
|
||
# 或 make uninstall
|
||
```
|
||
|
||
---
|
||
|
||
## OTA 远程更新
|
||
|
||
云端推送新版本后，edge-agent 会自动检测并下载更新到 staging 目录。
|
||
也可手动更新：
|
||
|
||
```bash
|
||
cd /root/AI-tianyan
|
||
git pull
|
||
make build
|
||
make install
|
||
sudo systemctl restart edge-agent edge-infer
|
||
```
|