在 Docker 镜像中安装 pydrill

Question

我有这个基于alpine 的 docker 文件，它使用conda 安装了几个包。最后安装pydrill 和pip，因为没有安装conda。

from jcrist/alpine-dask

RUN /opt/conda/bin/conda update -n base -c defaults conda -y
RUN /opt/conda/bin/conda update dask
RUN /opt/conda/bin/conda install -c conda-forge dask-ml
RUN /opt/conda/bin/conda install scikit-learn -y
RUN /opt/conda/bin/conda install flask -y
RUN /opt/conda/bin/conda install waitress -y
RUN /opt/conda/bin/conda install gunicorn -y
RUN /opt/conda/bin/conda install pytest -y
RUN /opt/conda/bin/conda install apscheduler -y
RUN /opt/conda/bin/conda install matplotlib -y
RUN /opt/conda/bin/conda install pyodbc -y

USER root
RUN apk update
RUN apk add py-pip
RUN pip install pydrill

当我构建 docker 映像时，一切正常。但是当我运行容器时，命令行启动gunicorn，但它失败并显示以下消息：

  File "/code/app/service/cm/exec/run_drill.py", line 1, in <module>
    from pydrill.client import PyDrill
   
   ModuleNotFoundError: No module named 'pydrill'

pip 安装正确吗？这是码头工人撰写：

version: "3.0"
services:

  web:
    image: img-dask
    volumes:
      - vol_py_code:/code
      - vol_dask_data:/data
      - vol_dask_model:/model
    ports:
      - "5000:5000"
    working_dir: /code
    environment:
      - app.config=/code/conf/py.app.json
      - common.config=/code/conf/py.common.json     
    entrypoint:
      - /opt/conda/bin/gunicorn
    command:
      - -b 0.0.0.0:5000
      - --reload
      - app.frontend.app:app


 scheduler:
    image: img-dask
    ports:
      - "8787:8787"
      - "8786:8786"
    entrypoint:
      - /opt/conda/bin/dask-scheduler

  worker:
    image: img-dask
    depends_on:
      - scheduler
    environment:
      - PYTHONPATH=/code
      - MODEL_PATH=/model/rfc_model.pkl
      - PREPROCESSING_PATH=/model/data_columns.pkl
      - SCHEDULER_ADDRESS=scheduler
      - SCHEDULER_PORT=8786
    volumes:
      - vol_py_code:/code
      - vol_dask_data:/data
      - vol_dask_model:/model
    entrypoint:
      - /opt/conda/bin/dask-worker
    command:
      - scheduler:8786
      
volumes:
  vol_py_code:
     name: vol_py_code
  vol_dask_data:
     name: vol_dask_data
  vol_dask_model:
     name: vol_dask_model
  

更新

如果我在容器内运行命令行，我可以看到 pydrill 已安装，但我的代码看不到库。

/code/conf # pip3 list
Package    Version  
---------- ---------
certifi    2020.12.5
chardet    4.0.0    
idna       2.10     
pip        18.1     
pydrill    0.3.4    
requests   2.25.1   
setuptools 40.6.2   
urllib3    1.26.4   
You are using pip version 18.1, however version 21.1.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.

Answer 1

问题在于 pydrill 和所有其他 conda 包处于不同的环境中。服务器启动时，看不到 pydrill，只有 conda 包。

要解决这个问题，请在 conda 的环境中安装 pip 本身：

from jcrist/alpine-dask

USER root
RUN /opt/conda/bin/conda create -p /pyenv -y
RUN /opt/conda/bin/conda install -p /pyenv dask scikit-learn flask waitress gunicorn \
    pytest apscheduler matplotlib pyodbc -y
RUN /opt/conda/bin/conda install -p /pyenv -c conda-forge dask-ml -y
RUN /opt/conda/bin/conda install -p /pyenv pip -y
RUN /pyenv/bin/pip install pydrill

Answer 2

你可以试试conda install pip 代替apk

类似

from jcrist/alpine-dask
WORKDIR /opt/conda/bin

RUN conda update -n base -c defaults conda -y
RUN conda update dask
RUN install -c conda-forge dask-ml 
RUN conda install stickit-learn flask waitress gunicorn \
    pytest apscheduler matplotlib pydobc pip -y
RUN pip install pydrill

Answer 3

我已经为conda-forge打包了pydrill，所以你可以简单地conda install -c conda-forge pydrill。