Dockerfile在图像识别服务部署,揭秘CUDA环境配置、多阶段构建优化等核心技术

FROM nvidia/cuda:11.3.1-base-ubuntu20.04

# 安装系统依赖
RUN apt-get update && apt-get install -y \
    python3-pip \
    libgl1 \
    && rm -rf /var/lib/apt/lists/*

# 设置工作目录
WORKDIR /app

# 复制依赖清单优先于代码（利用Docker层缓存）
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# 复制完整项目代码
COPY . .

# 指定启动命令（加载模型并启动API服务）
CMD ["python3", "inference_api.py"]

/image-recognition-service
├── app
│   ├── models
│   │   └── resnet50.pth
│   ├── utils
│   │   └── image_processing.py
│   └── inference_api.py
├── Dockerfile
└── requirements.txt

# 多阶段构建优化镜像体积（阶段1：构建环境）
FROM python:3.9-slim as builder

WORKDIR /build
COPY requirements.txt .
RUN pip install --user -r requirements.txt

# 阶段2：运行时环境
FROM python:3.9-slim

# 从builder阶段复制已安装的包
COPY --from=builder /root/.local /root/.local
ENV PATH=/root/.local/bin:$PATH

# 安装必要系统库（OpenCV依赖）
RUN apt-get update && apt-get install -y \
    libglib2.0-0 \
    libsm6 \
    && rm -rf /var/lib/apt/lists/*

# 设置非root用户
RUN useradd -m appuser && chown -R appuser /app
USER appuser

COPY --chown=appuser:appuser . /app
WORKDIR /app

EXPOSE 8000
CMD ["gunicorn", "--bind", "0.0.0.0:8000", "inference_api:app"]

docker build -t img-recog:v1 .
docker run -p 8000:8000 img-recog:v1

stages:
  - build
  - test

build_image:
  stage: build
  script:
    - docker build -t $CI_REGISTRY_IMAGE:$CI_COMMIT_SHA .
    - docker push $CI_REGISTRY_IMAGE:$CI_COMMIT_SHA

inference_test:
  stage: test
  image: $CI_REGISTRY_IMAGE:$CI_COMMIT_SHA
  script:
    - pytest tests/

# 错误示例：在基础镜像中使用root用户
FROM python:3.9
RUN pip install flask  # 以root身份运行

# 正确做法
FROM python:3.9-slim
RUN useradd -r -u 1001 appuser
USER appuser

# 错误：频繁变动的层放在前面
COPY . .
RUN pip install -r requirements.txt  # 每次代码修改都会导致重新安装依赖

# 正确：稳定层优先
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .  # 代码变更不会影响依赖层

# 将模型文件挂载为Volume
docker run -v /path/to/models:/app/models img-recog:v1

version: '3.8'

services:
  inference:
    build: .
    ports:
      - "8000:8000"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  redis:
    image: redis:alpine

apiVersion: apps/v1
kind: Deployment
metadata:
  name: img-recog
spec:
  replicas: 3
  selector:
    matchLabels:
      app: img-recog
  template:
    metadata:
      labels:
        app: img-recog
    spec:
      containers:
      - name: inference
        image: registry.example.com/img-recog:v1.2
        resources:
          limits:
            nvidia.com/gpu: 1