Linux下terminal环境变量在Jupyter-lab不生效问题

1. 问题还原

(1)环境

• GPU: A100*4
• python3.10
• cuda-12.2.2
• cudnn-8.8.1
• Tensorflow2.15
• TensorRT-8.6.1.6

FROM nvidia/cuda:12.2.2-cudnn8-devel-ubuntu22.04
COPY ["dependence/*", "/root"]
COPY ["dependence/cudnn-linux-x86_64-8.8.1.3_cuda12-archive/include/*", "/usr/local/cuda/include/"]
COPY ["dependence/cudnn-linux-x86_64-8.8.1.3_cuda12-archive/lib/*", "/usr/local/cuda/lib64/"]

# 安装相关软件，可自行调整
RUN apt update && \
    apt install -y python3-pip net-tools \
    less curl iputils-ping telnet nmon zip cron gcc \
    language-pack-zh-hans sudo \
    ssh dos2unix tmux \
    gawk htop libgl1-mesa-glx zsh vim git && \
    pip install jupyter -i http://pypi.douban.com/simple --trusted-host pypi.douban.com && \
    ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && \
    tar -zvxf /root/TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-12.0.tar.gz -C /usr/local/ && \
    rm /bin/sh && \
    ln -s /bin/bash /bin/sh && \
    echo export LC_ALL=zh_CN.utf8 >>/etc/profile && \
    echo export LANG=zh_CN.utf8 >>/etc/profile && \
    echo export EDITOR=vim >>/etc/profile && \
    echo export PATH=$PATH:/usr/local/cuda/bin >>/etc/profile && \
    echo export CUDA_HOME=$CUDA_HOME:/usr/local/cuda >>/etc/profile && \
    echo export LD_LIBRARY_PATH=/usr/local/TensorRT-8.6.1.6/targets/x86_64-linux-gnu/lib:$LD_LIBRARY_PATH >>/etc/profile && \
    echo export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH >>/etc/profile && \
    echo export LD_LIBRARY_PATH=/usr/local/cuda/extras/CUPTI/lib64:$LD_LIBRARY_PATH >>/etc/profile && \
    echo SHELL=/bin/bash >> /etc/profile && \
    echo alias ll=\'ls -al\' >> /etc/profile && \
    echo alias ls=\'ls -a --color\' >> /etc/profile && \
    echo source /etc/bash.bashrc >> /etc/profile && \
    sed -i 's/#PermitRootLogin prohibit-password/PermitRootLogin yes/' /etc/ssh/sshd_config && \
    useradd -ms /bin/bash hoshinory && \
    echo "hoshinory:hoshinory" | chpasswd && \
    groupadd -g 4000 policy && \
    usermod -aG policy hoshinory && \
    usermod -u 997 hoshinory && \
    groupmod -g 997 hoshinory && \
    echo "hoshinory  ALL=NOPASSWD:ALL">>/etc/sudoers

USER hoshinory
COPY ["dependence/.bashrc", "/home/hoshinory/"]
COPY ["dependence/.vimrc", "/home/hoshinory/"]
COPY ["dependence/start.sh", "/home/hoshinory/"]

RUN pip install jupyter -i http://pypi.douban.com/simple --trusted-host pypi.douban.com  && \
    pip install install jupyterlab-lsp -i http://pypi.douban.com/simple --trusted-host pypi.douban.com && \
    pip install python-lsp-server[all] -i http://pypi.douban.com/simple --trusted-host pypi.douban.com && \
    jupyter lab --generate-config && \
    sed -i '2i\import os' /home/hoshinory/.jupyter/jupyter_lab_config.py && \
    sed -i '3i\os.environ["LD_LIBRARY_PATH"] = os.environ.get("LD_LIBRARY_PATH", "") + ":/usr/local/cuda/lib64"' /home/hoshinory/.jupyter/jupyter_lab_config.py && \
    sed -i '4i\os.environ["LD_LIBRARY_PATH"] = os.environ.get("LD_LIBRARY_PATH", "") + ":/usr/local/TensorRT-8.6.1.6/targets/x86_64-linux-gnu/lib"' /home/hoshinory/.jupyter/jupyter_lab_config.py && \
    sed -i '5i\os.environ["LD_LIBRARY_PATH"] = os.environ.get("LD_LIBRARY_PATH", "") + ":/usr/local/cuda/extras/CUPTI/lib64"' /home/hoshinory/.jupyter/jupyter_lab_config.py && \
    sed -i '6i\os.environ["CUDA_HOME"] = os.environ.get("CUDA_HOME", "") + ":/usr/local/cuda:/usr/local/cuda"' /home/hoshinory/.jupyter/jupyter_lab_config.py && \
    sed -i '7i\os.environ["PATH"] = os.environ.get("PATH", "") + ":/usr/local/cuda/bin"' /home/hoshinory/.jupyter/jupyter_lab_config.py

EXPOSE 8888
#CMD ["sh", "-c", "/home/hoshinory/start.sh && tail -f /dev/null"]
CMD jupyter lab --notebook-dir=/home/hoshinory \
                --ip=0.0.0.0 \
                --no-browser \
                --allow-root \
                --port=8888 \
                --config=/home/hoshinory/.jupyter/jupyter_lab_config.py \
                --NotebookApp.token='' \
                --NotebookApp.password='' \
                --NotebookApp.allow_origin='*' \
                --NotebookApp.base_url=/ \
                --NotebookApp.iopub_data_rate_limit=1e10 \
                --FileContentsManager.delete_to_trash=False;

The main problem is enviromental variables not worked in the notebook. It prevent me to develop codes with GPU.

There is a Infinitie loops BUG in prophet, detail shows as following:

• All programms are running on a Linux image built by docker.

• You must keep a main progress alive when you running a Linux image. Main progress can not be changed once you runing image. And image will exit if the main progress died.

• Notebook is running in a Linux image as a main progress in platform of prophet. I donot kown which config file(maybe no config file, assert there is a config file) was loaded when Notebook running. What's worse, the config file loaded does not include enviromental variables we needed.

• I search the answer in Google, and something interesting happend. >(1) I must source ~/.bashrc first and then restart the progress of notebook

  2. If I restart the progress of notebook, I must kill the progress of notebook at beginning.
  3. If I kill the progress of notebook at beginning(I remind that it is a main progress), the image will exit and restart. When Notebook auto restart, it loads config file that nobody knows for sure.
  4. If I changed main progress(someone else expcept Notebook), image cannot be running.