Configuration

Docker Image

JupyterHub must be started with a Jupyter Docker Stacks image, or a descendant from it.

Jupyter Docker Stacks provides a wide array of Jupyter Notebooks to get you started:

These Docker images are however NOT GPU-enabled. For your convenience we created a fork of the public Jupyter Docker Stacks with GPU-processing capability. These images use nvcr.io/nvidia/cuda:10.0-cudnn7-runtime-ubuntu18.04 as base image instead of a vanilla Ubuntu.

You can use these images by changing the prefix from jupyter/ to gitlab.ilabt.imec.be:4567/ilabt/gpu-docker-stacks/. For example: a Jupyter notebook with the GPU-enabled Tensorflow and Keras-library can be found under gitlab.ilabt.imec.be:4567/ilabt/gpu-docker-stacks/tensorflow-notebook.

Adding custom software

JupyterHub supports all Docker images which descend from the Jupyter Docker Stacks

You can add additional software by creating a docker image which starts from an existing Jupyter Docker Stacks-image, and which executes the necessary installation commands. Check the section Using pip install or conda install in a Child Docker image of the Jupyter Docker Stacks documentation for a full example.

If you want to use GPU’s, you’ll need to descend your image from a GPU-enabled image. These are available at
iLab.t GPU Docker Stacks Container Registry.

For example, to add a library to the GPU-enabled tensorflow-notebook, you can use the following Dockerfile:

# Start from a core gpu-enabled stack version
FROM gitlab.ilabt.imec.be:4567/ilabt/gpu-docker-stacks/tensorflow-notebook:cuda10.1-ubuntu18.04
# Set user as root so next commands run correctly
USER root
# Install dependencies
RUN apt-get update && apt install graphviz libgraphviz-dev libcgraph6 && rm -rf /var/lib/apt/lists/*
# Install spektral in the default python3 environment
RUN pip install spektral
# Restore the correct user
USER jovyan
After building this image, upload it to a Docker Registry with docker push. You can use any docker internet-accessible
Docker registry, however, we advice you to use the private Docker Registry available via gitlab.ilabt.imec.be.

See dockerImage for more info.

For example to build and push this Dockerfile to the public GPULab Docker Registry:

docker build -t gitlab.ilabt.imec.be:4567/myusername/myprojectname/spektral:v1 .
docker push gitlab.ilabt.imec.be:4567/myusername/myprojectname/spektral:v1

Once pushed to the registry, you can start your custom Jupyter Docker Image by specifying the full registry path in the field dockerImage. In this example: gitlab.ilabt.imec.be:4567/myusername/myprojectname/spektral:v1.

When using a docker registry with authentication, this path will be of the form <deployusername>:<password>@gitlab.ilabt.imec.be:4567/<project>/<image>:<tag>. When using a Gitlab Registry you can create a deploy key to get a username and password for letting JupyterHub pull this image.