Skip to main content
Version: 最新版本(unreleased)

Wrapper Image Build Process

The process of building a business image​

1. Build requirements​

  • Docker (20.10.0 or above)
  • Service based on conda environment (one conda environment corresponds to one inference service)

2. Envd install​

pip install --pre --upgrade envd
envd bootstrap

3. Export the conda environment list of the current inference service​

conda env export > env.yaml

4.Write build.envd file​

Select the base image to load and the python version of conda​

  • base(language="python3.8", os="ubuntu20.4")
  • Optional base mirror
    • aiges_cpu
    • aiges_gpu

Install the conda env environment list under the specified path​

  • install.conda_packages(env_file="env.yaml")

Install system dependencies (write the dependencies to be installed into the name list, separated by commas)​

  • install.system_packages(name = ["..."])

Copy the service code to the business image​

  • io.copy(src="./detectron2", dest="/")

Offline installation (requires file copy and then install)​

  • io.copy(src="./detectron2", dest="/")
  • run(commands=["pip install -e /detectron2",])

Set the installation source​

  • config.pip_index()
  • config.conda_channel()

Example (there are optional other installation interfaces)​

def build():
    mirror_config() 
    base(language="python3.8", os="ubuntu20.4")    # Load the base image and set the python version built into conda
    install.python_packages(name = [
        "torch==1.10",
        "openmim"
    ])
    install.conda_packages(channel= ["pytorch"], env_file = "env.yaml")   # Install conda yaml environment in the specified path
    install.python_packages(requirements="build.txt")   # Install requirements.txt in the specified path
    install.system_packages(name = [                    # System dependent installation
        "libgl1-mesa-glx"
    ])
    run(commands=[
        "mim install mmcv-full",                        # Install through the third-party tool mim
    ])

    io.copy(src="./detectron2", dest="/")               # Copy of local files to mirror
    run(commands=[                                      # Dependency package installation offline
        "pip install -e /detectron2",
    ])
def mirror_config():                                    # Configure download source
    config.pip_index(url = "https://pypi.tuna.tsinghua.edu.cn/simple")
    config.conda_channel(channel="""
    channels:
    - defaults
    show_channel_urls: true
    default_channels:
    - https://repo.anaconda.com/pkgs/main
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/menpo/
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/bioconda/
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/msys2/
    - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
    custom_channels:
    conda-forge: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    msys2: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    bioconda: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    menpo: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    pytorch: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    pytorch-lts: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    simpleitk: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
    """)

5. Image building via envd​

envd build -t ImageName:TAG -f build.envd

example：envd build -t yolo:v1.0.0 -f build.envd