Docker构建时conda安装r-base初始求解失败如何解决？

执行make build触发Docker构建时，在执行conda install -c conda-forge r-base步骤失败，报错如下：

Step 12/25 : RUN conda install -c conda-forge r-base
 ---> Running in 828e7f234ed3
Collecting package metadata (current_repodata.json): ...working... done
Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve.

最终构建终止，错误信息：

The command '/bin/sh -c conda install -c conda-forge r-base' returned a non-zero code: 1
make: *** [Makefile:2: build] Error 1

相关Dockerfile和Makefile内容如下：

Dockerfile #

FROM continuumio/anaconda3

ENV INSTALL_PATH /environment_name
RUN mkdir -p $INSTALL_PATH
WORKDIR $INSTALL_PATH

#COPY environment.yml environment.yml
#RUN conda env create -f environment.yml
RUN conda create -n environment_name 
RUN /opt/conda/bin/activate environment_name
#RUN conda config --add channels conda-forge
#RUN conda config --set channel_priority strict
#RUN conda install r-cairodevice
RUN conda install -n environment_name bokeh h5py numpy scipy pandas scikit-learn sphinx spyder jupyter statsmodels
RUN conda install -n environment_name -c bioconda pymzml
RUN conda install -n environment_name -c bioconda netcdf4
RUN conda install -c conda-forge fonts-conda-forge
RUN conda install -c conda-forge r-base
RUN conda install -c bioconda bioconductor-ropls
RUN pip install flask werkzeug gunicorn 
RUN conda install seaborn
RUN echo "source activate environment_name" > ~/.bashrc
ENV PATH /opt/conda/envs/environment_name/bin:$PATH

COPY . .

RUN chmod +x run_server.sh
RUN chmod +x start.sh
ADD start.sh /
CMD /start.sh
RUN conda install -c conda-forge libjpeg-turbo
RUN git clone https://bitbucket.org/iAnalytica/mshub_process.git
RUN conda install -n environment_name -c bioconda pyteomics
#CMD gunicorn -b 0.0.0.0:8000 --access-logfile - "api.app:app"

Makefile #

build:
        docker build -t environment_name .

1. 修正conda环境安装目标 #

Docker的RUN命令是独立shell执行，/opt/conda/bin/activate environment_name仅在当前RUN生效，后续命令默认使用base环境。安装r-base时必须显式指定目标环境，否则会在base环境中安装，与已安装包产生依赖冲突：

RUN conda install -n environment_name -c conda-forge r-base

2. 启用conda-forge通道并设置严格优先级 #

打开Dockerfile中被注释的通道配置，让conda优先使用conda-forge的包，减少跨通道依赖冲突：

RUN conda config --add channels conda-forge
RUN conda config --set channel_priority strict

将这两行放在创建环境之后、安装包之前的位置。

3. 合并conda安装命令 #

分散的conda install会逐步累积依赖冲突，同时增加镜像图层。将同通道的包合并到一个命令中，还能提升构建效率：

# 合并defaults通道包安装
RUN conda install -y -n environment_name \
    bokeh h5py numpy scipy pandas scikit-learn sphinx spyder jupyter statsmodels seaborn

# 合并bioconda通道包安装
RUN conda install -y -n environment_name -c bioconda \
    pymzml netcdf4 bioconductor-ropls pyteomics

# 合并conda-forge通道包安装
RUN conda install -y -n environment_name -c conda-forge \
    fonts-conda-forge r-base libjpeg-turbo

添加-y参数自动确认安装，避免构建过程中需要交互。

4. 固定基础镜像版本 #

continuumio/anaconda3默认拉取最新版，可能存在依赖兼容性问题。指定一个稳定版本，比如：

FROM continuumio/anaconda3:2023.09-0

5. 调整pip安装顺序 #

将pip安装放在所有conda安装之后，且确保使用目标环境的pip（已通过PATH环境变量配置），避免安装到base环境：

# 先完成conda安装，再执行pip安装
ENV PATH /opt/conda/envs/environment_name/bin:$PATH
RUN pip install flask werkzeug gunicorn

完整修改后的Dockerfile示例 #

FROM continuumio/anaconda3:2023.09-0

ENV INSTALL_PATH /environment_name
RUN mkdir -p $INSTALL_PATH
WORKDIR $INSTALL_PATH

# 创建conda环境并配置通道
RUN conda create -y -n environment_name
RUN conda config --add channels conda-forge
RUN conda config --set channel_priority strict

# 批量安装各类包
RUN conda install -y -n environment_name \
    bokeh h5py numpy scipy pandas scikit-learn sphinx spyder jupyter statsmodels seaborn

RUN conda install -y -n environment_name -c bioconda \
    pymzml netcdf4 bioconductor-ropls pyteomics

RUN conda install -y -n environment_name -c conda-forge \
    fonts-conda-forge r-base libjpeg-turbo

# 配置环境变量并执行pip安装
ENV PATH /opt/conda/envs/environment_name/bin:$PATH
RUN pip install flask werkzeug gunicorn

# 后续文件操作
COPY . .
RUN chmod +x run_server.sh
RUN chmod +x start.sh
ADD start.sh /
RUN git clone https://bitbucket.org/iAnalytica/mshub_process.git

RUN echo "source activate environment_name" > ~/.bashrc
CMD /start.sh

内容的提问来源于stack exchange，提问作者anaclc