DigitalOcean Droplet中Docker镜像安装detectron2失败求助

本地构建Docker镜像安装detectron2完全正常，但在2GB内存/2核Intel CPU、60GB磁盘的Ubuntu 22.10 x64 DigitalOcean Droplet上构建时，出现ERROR: Failed building wheel for detectron2错误，日志核心信息为：

gcc: fatal error: Killed signal terminated program cc1plus
compilation terminated.
error: command '/usr/bin/gcc' failed with exit code 1

补充情况：使用--no-cache参数重建镜像时detectron2未被安装，此前无此问题；第二个未指定detectron2版本的Dockerfile在本地也无法完成安装。

测试的两个Dockerfile #

Dockerfile 1（指定v0.4版本） #

FROM python:3.10-slim-buster

# Update package lists
RUN apt-get update && apt-get install ffmpeg libsm6 libxext6 gcc g++ git build-essential libpoppler-cpp-dev pkg-config poppler-utils tesseract-ocr libtesseract-dev -y

# Make working directories
RUN  mkdir -p  /app
WORKDIR  /app
# Copy the requirements.txt file to the container
COPY requirements.txt .

# Install dependencies
RUN pip install --upgrade pip

RUN pip install torch torchvision torchaudio

RUN pip install -r requirements.txt

RUN pip install 'git+https://github.com/facebookresearch/detectron2.git@v0.4#egg=detectron2'

# Copy the .env file to the container
COPY .env .

# Copy every file in the source folder to the created working directory
COPY  . .

# Expose the port that the application will run on
EXPOSE 8080

# Start the application
CMD ["python3.10", "uvicorn", "-m", "main:app", "--proxy-headers", "--host", "0.0.0.0", "--port", "8080"]

Dockerfile 2（未指定版本） #

FROM python:3.10-slim-buster

# Update package lists
RUN apt-get update && apt-get install ffmpeg libsm6 libxext6 gcc g++ git build-essential libpoppler-cpp-dev pkg-config poppler-utils tesseract-ocr libtesseract-dev -y

# Make working directories
RUN  mkdir -p  /app
WORKDIR  /app
# Copy the requirements.txt file to the container
COPY requirements.txt .

# Install dependencies
RUN pip install --upgrade pip

RUN pip install torch torchvision torchaudio

RUN pip install -r requirements.txt

RUN pip install 'git+https://github.com/facebookresearch/detectron2.git'

# Copy the .env file to the container
COPY .env .

# Copy every file in the source folder to the created working directory
COPY  . .

# Expose the port that the application will run on
EXPOSE 8080

# Start the application
CMD ["python3.10", "-m", "uvicorn", "main:app", "--proxy-headers", "--host", "0.0.0.0", "--port", "8080"]

解决方案 #

1. 给Droplet添加Swap分区（核心解决内存不足） #

编译detectron2需要大量内存，2GB内存无法支撑，添加Swap作为虚拟内存补充：

# 在Droplet上执行，创建4GB Swap文件
fallocate -l 4G /swapfile
chmod 600 /swapfile
mkswap /swapfile
swapon /swapfile
# 设置开机自动挂载Swap
echo '/swapfile none swap sw 0 0' >> /etc/fstab

构建完成后若不需要Swap，可执行以下命令删除：

swapoff /swapfile
rm /swapfile
sed -i '/swapfile/d' /etc/fstab

2. 使用预编译轮子跳过源码编译 #

直接安装对应环境的预编译detectron2包，避免在容器内编译：
修改Dockerfile中安装detectron2的命令，替换为匹配你torch版本的预编译包地址（以torch 1.13+、CUDA 11.7为例）：

RUN pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu117/torch1.13/index.html

注意：需根据实际安装的torch、CUDA版本选择对应轮子地址，版本不匹配会导致安装失败

3. 限制编译核心数降低内存占用 #

编译时默认使用全部CPU核心，会加剧内存消耗，限制核心数减少压力：
在安装detectron2前设置环境变量：

ENV MAX_JOBS=1
RUN pip install 'git+https://github.com/facebookresearch/detectron2.git@v0.4#egg=detectron2'

4. 优化Dockerfile编译效率 #

在apt-get install命令中添加ninja-build，避免编译时fallback到慢的distutils后端，加快编译速度：

RUN apt-get update && apt-get install ffmpeg libsm6 libxext6 gcc g++ git build-essential ninja-build libpoppler-cpp-dev pkg-config poppler-utils tesseract-ocr libtesseract-dev -y

内容的提问来源于stack exchange，提问作者neil_ruaro