Mounting path /home/jovyan

[jankogasic]

What docker image(s) are you using?

minimal-notebook

Host OS

Rocky Linux 9.5

Host architecture

x86_64

What Docker command are you running?

I am running docker compose file:

name: py-env
services:
  jupyter:
    build:
      context: .
      args:
        CONDA_ENV_NAME: py-env
    image: py-env:latest
    user: root # This is a must in order for env var user switch to take effect
    restart: always
    ports:
      - "8888:8888"
    volumes:
      - "/path/on/my/host/jovyan:/home/jovyan"  
    environment:
      NB_UID: "1005"
      NB_GID: "1005"
      CHOWN_HOME: "yes"
      CHOWM_HOME_OPTS: "-R"
    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]  # Ensure the container has access to GPUs
    runtime: nvidia  # Specify the use of NVIDIA runtime for GPU support

How to Reproduce the problem?

I manually created the /path/on/my/host/jovyan path with non-root user.

Here is Dockerfile:

FROM quay.io/jupyter/minimal-notebook

# Name your environment and choose the Python version
ARG CONDA_ENV_NAME=my-env
ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:${CONDA_DIR}/envs/${CONDA_ENV_NAME}/lib"

USER root
 
RUN apt-get update && \
    apt-get install -y \
        openjdk-21-jdk \
        libopenmpi-dev && \
    apt-get clean

# USER ${NB_UID}

# Create conda environment from env1.yaml file
COPY --chown=${NB_UID}:${NB_GID} env1.yaml /tmp/
RUN mamba env create -p "${CONDA_DIR}/envs/${CONDA_ENV_NAME}" -f /tmp/env1.yaml && \
    mamba clean --all -f -y && \
    fix-permissions "${CONDA_DIR}" && \
    fix-permissions "/home/${NB_USER}"

# Create Python kernel and link it to jupyter
RUN "${CONDA_DIR}/envs/${CONDA_ENV_NAME}/bin/python" -m ipykernel install --user --name="${CONDA_ENV_NAME}" && \
    fix-permissions "${CONDA_DIR}" && \
    fix-permissions "/home/${NB_USER}"

# Install from the requirements.txt file
COPY --chown=${NB_UID}:${NB_GID} requirements.txt /tmp/
RUN "${CONDA_DIR}/envs/${CONDA_ENV_NAME}/bin/pip" install --upgrade pip  && \
    "${CONDA_DIR}/envs/${CONDA_ENV_NAME}/bin/pip" install --no-cache-dir --requirement /tmp/requirements.txt && \
    fix-permissions "${CONDA_DIR}" && \
    fix-permissions "/home/${NB_USER}"

# seperating tensorrt_llm from requirements.txt because it needs different index url
RUN "${CONDA_DIR}/envs/${CONDA_ENV_NAME}/bin/pip" install --no-cache-dir tensorrt_llm --extra-index-url https://pypi.nvidia.com && \
    fix-permissions "${CONDA_DIR}" && \
    fix-permissions "/home/${NB_USER}"

# This changes the custom Python kernel so that the custom environment will
# be activated for the respective Jupyter Notebook and Jupyter Console
# hadolint ignore=DL3059
# RUN /opt/setup-scripts/activate_notebook_custom_env.py "${CONDA_ENV_NAME}"

# Comment the line above and uncomment the section below instead to activate the custom environment by default
# Note: uncommenting this section makes "${CONDA_ENV_NAME}" default both for Jupyter Notebook and Terminals
# More information here: https://github.com/jupyter/docker-stacks/pull/2047
USER root
RUN \
    # This changes a startup hook, which will activate the custom environment for the process
    echo conda activate "${CONDA_ENV_NAME}" >> /usr/local/bin/before-notebook.d/10activate-conda-env.sh && \
    # This makes the custom environment default in Jupyter Terminals for all users which might be created later
    echo conda activate "${CONDA_ENV_NAME}" >> /etc/skel/.bashrc && \
    # This makes the custom environment default in Jupyter Terminals for already existing NB_USER
    echo conda activate "${CONDA_ENV_NAME}" >> "/home/${NB_USER}/.bashrc"

COPY docker-compose.yaml /home/jovyan/docker-compose-files/docker-compose.yaml

USER ${NB_UID}

Command output

Expected behavior

I expected /home/jovyan to be mounted to host`s directory with everything inside of container working.

Actual behavior

It seems like that my conda environment is not being created as expected. Also, when I open the terminal the conda environment is not automatically activated.

Anything else?

When I mount "/path/on/my/host/jovyan:/home/jovyan/work" everything works fine.

I want to achieve this because some libraries like tensorrt_llm tend to download LLMs in /home/jovyan and not /home/jovyan/work which I want to avoid.

Latest Docker version

• I've updated my Docker version to the latest available, and the issue persists

[benz0li]

Cross reference:

• Populate home dir if it is empty #1478

[mathbunnyru]

Thanks @benz0li 👍

Currently having something in home/jovyan from both container and the bind mount is not supported.

I think the best you can do is mount to /home/jovyan/work, but at the same time change the cmd to start-notebook.py --ServerApp.root_dir=/home/jovyan/work
This way you won't notice that the mount dir is work/ and new tab will open in this directory.

Also, there is a typo in the name CHOWM_HOME_OPTS you use.

[jankogasic]

With combination of HF_HOME environment variable and your proposition, my problem was solved.

Good catch of typo.

Closing the issue. Thanks!