Building Custom Kubeflow Jupyter Notebook Image

Describes how to build the custom Kubeflow Jupyter notebook image.

About this task

Build a custom image with one of the default notebooks available in the Kubeflow dashboard as a base image. The notebook will be created using the custom image.

You can build the custom image for both air-gapped and non-air-gapped environments for all three types of packages – OS level packages, conda packages and pip packages.

To build the custom Kubeflow Jupyter notebook image, perform:

Procedure

  1. Create requirements.txt file.
    For example: The content of the file can be following: 
    ###requirements.txt
# pandas packages
pandas=1.5.0
numpy=1.24.2
# Some other packages 
###
  2. Create Dockerfile. 
    ARG BASE_IMG=gcr.io/mapr-252711/kubeflow/notebooks/jupyter-scipy: <image-tag>
FROM $BASE_IMG
COPY requirements.txt /tmp/requirements.txt
RUN python3 -m pip install -r /tmp/requirements.txt --quiet --no-cache-dir \
 && rm -f /tmp/requirements.txt
  3. Build the image with docker build command. Replace the <image> with actual image name, <tag> with actual tag name, and <base_img> with actual base image. 
    docker build -t  <image>:<example>  .
    (OR) If the default base image is not suitable,
    1. Choose one of the default images as the base image. See Notebook Images.
    2. Run docker build -t <image>:<example> --build-arg BASE_IMG=<base_img> .
  4. Push the image to the registry. 
    docker push <image>:<example>
  5. Use the custom image option when creating the notebook server in the Kubeflow UI.