在 docker 中激活 conda 环境

Question

我需要在 docker 中激活环境并在此环境中运行命令。我创建了环境，但随后我尝试激活此环境并以这种方式运行命令：

CMD [ "source activate mro_env && ipython kernel install --user --name=mro_env" ]

但是当我运行 docker 时出现错误：

[FATAL tini (8)] exec source activate mro_env && ipython kernel install 
--user --name=mro_env failed: No such file or directory

这是整个 Dockerfile：

FROM continuumio/miniconda3

ADD /src/mro_env.yml /src/mro_env.yml
RUN conda env create -f /src/mro_env.yml

# Pull the environment name out of the mro_env.yml
RUN echo "source activate $(head -1 /src/mro_env.yml | cut -d' ' -f2)" > ~/.bashrc
ENV PATH /opt/conda/envs/$(head -1 /src/mro_env.yml | cut -d' ' -f2)/bin:$PATH

CMD [ "source activate mro_env && ipython kernel install --user --name=mro_env" ]

Answer 1

对我来说，这里介绍的解决方案工作顺利：

FROM continuumio/miniconda3
RUN conda create -n {env} python=3.6
RUN echo "source activate {env}" > ~/.bashrc
ENV PATH /opt/conda/envs/{env}/bin:$PATH

UPD：“env”只是一个示例，如果需要，请调整“env”名称。

Answer 2

按照本教程进行操作，它奏效了。示例 Dockerfile：

FROM continuumio/miniconda
WORKDIR /usr/src/app
COPY ./ ./
RUN conda env create -f environment.yml

# Make RUN commands use the new environment:
SHELL ["conda", "run", "-n", "myenv", "/bin/bash", "-c"]

EXPOSE 5003
# The code to run when container is started:
ENTRYPOINT ["conda", "run", "-n", "myenv", "python3", "src/server.py"]

更新：

如果您使用 4.9 版本的 conda，则可以使用“conda run --no-capture-output”来不缓冲 IO。更新的 Dockerfile：

FROM continuumio/miniconda
WORKDIR /usr/src/app
COPY ./ ./
RUN conda env create -f environment.yml

# Make RUN commands use the new environment:
SHELL ["conda", "run", "--no-capture-output", "-n", "myenv", "/bin/bash", "-c"]

EXPOSE 5003
# The code to run when container is started:
ENTRYPOINT ["conda", "run", "--no-capture-output", "-n", "myenv", "python3", "src/server.py"]

Answer 3

对我来说，问题是安装后在 docker 内运行命令conda activate env导致 conda 要求我使用该conda init bash命令。但是，此命令要求您重新启动 shell，我们不想在 docker 内执行此操作。因此，解决方案是认识到要求conda您重新启动 shell 的原因是因为它已修改并想要重新加载~/.bashrc. 我们可以手动执行此操作，并且无需使用以下命令重新启动 shell：

. ~/.bashrc

对于那些需要它的人来说，这是完整的 dockerfile：

FROM ubuntu:18.04
# update apt and get miniconda
RUN apt-get update \
    && apt-get install -y wget \
    && wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh


# install miniconda
ENV PATH="/root/miniconda3/bin:$PATH"
RUN mkdir /root/.conda && bash Miniconda3-latest-Linux-x86_64.sh -b

# create conda environment
RUN conda init bash \
    && . ~/.bashrc \
    && conda create --name test-env python=3.7 \
    && conda activate test-env \
    && pip install ipython

编辑-处理评论

评论者是正确的，因为上面的示例不适用于在 docker 中切换 conda 环境。最近，我发现要使其正常工作，您应该RUN以conda init bash && .~/.bashrc && conda activate env.

这是另一个例子：

FROM quay.io/pypa/manylinux2014_x86_64

RUN yum install -y wget
RUN wget https://repo.anaconda.com/miniconda/Miniconda3-py39_4.9.2-Linux-x86_64.sh
RUN bash Miniconda3-py39_4.9.2-Linux-x86_64.sh -b -p /Miniconda3

RUN /Miniconda3/bin/conda create -y --name py37 python=3.7 pytest
RUN /Miniconda3/bin/conda create -y --name py38 python=3.8 pytest
RUN /Miniconda3/bin/conda create -y --name py39 python=3.9 pytest

RUN /Miniconda3/bin/conda init && bash ~/.bashrc && . ~/.bashrc

ENV conda /Miniconda3/bin/conda
ENV bashrc /root/.bashrc

# install numpy in each env
RUN $conda init && . $bashrc && conda activate py37 && pip install numpy 
RUN $conda init && . $bashrc && conda activate py38 && pip install numpy
RUN $conda init && . $bashrc && conda activate py39 && pip install numpy

Answer 4

您可以设置 CONDA_DEFAULT_ENV

像这样：

FROM continuumio/miniconda3

ARG conda_env=mro_env

ADD /src/environment.yml /src/environment.yml
RUN conda env create -f /src/environment.yml

ENV PATH /opt/conda/envs/$conda_env/bin:$PATH
ENV CONDA_DEFAULT_ENV $conda_env

CMD [ "python", "test.py" ]

更新：

更好地使用激活。为我工作：

FROM continuumio/miniconda3

ADD /src/environment.yml /src/environment.yml

RUN conda env create -f /src/environment.yml
ENV PATH /opt/conda/envs/mro_env/bin:$PATH
RUN /bin/bash -c "source activate mro_env"

CMD [ "python", "test.py" ]

Answer 5

As the user merv points out in one of the comments above (sorry don't have enough rep to vote up the comment) conda run buffers all stdout/stderr, thus making it not workable for applications interacting or even just displaying logs over I/O.

I noticed there was no accepted answer, so I just post what has worked very well for me:

You can use an entrypoint script to activate the conda enviroment and let it take over further input from the Dockerfile such that the python script can be executed within the activated conda environment

Example Dockerfile:

FROM continuumio/miniconda3

# make docker use bash instead of sh
SHELL ["/bin/bash", "--login", "-c"]

# copy all necessary files
COPY environment.yml .
COPY ownchain/mypyscript.py .
COPY entrypoint.sh /usr/local/bin/

# make entrypoint script executable
RUN chmod u+x /usr/local/bin/entrypoint.sh
# create environment
RUN conda env create -f environment.yml

ENTRYPOINT ["/usr/local/bin/entrypoint.sh"]
CMD ["python", "mypyscript.py"]

Where the entrypoint.sh looks like this:

#!/bin/bash --login
set -e

# activate conda environment and let the following process take over
conda activate myenv
exec "$@"

All credit to David R. Pugh from this post that has more details, in particular with regards to Jupyter.

Answer 6

以下是如何在 conda 环境中运行脚本而不缓冲所有标准输入/输出。

该--no-capture-output选项自 conda 4.9 版起可用。

FROM continuumio/miniconda3

COPY ./environment.yml ./environment.yml

RUN conda env create && conda clean --all -f --yes

ENTRYPOINT ["conda", "run", "--no-capture-output", "-n", "the_environment_name", "python", "myscript.py"]