This is a minimal Centos-based docker image for running the Jupyter notebook server. It is an overly-optimistic redesign of the original Alpine-based version used by the nbgallery team. For more information about the team and project overall, please check out this post on our github.io site.
- base the image off of Centos7
- replicate the core functionality of the Alpine-based version:
- launch a secure Jupyter notebook on startup
- support Py3 kernel out of the box
- include the nbgallery additions to Jupyter
- allow for dynamic kernel installation for other languages
- attempt to keep the size low (< 1G)
- This will not be as small as the Alpine version
- incorporate conda for package management
- launch into a user account rather than root
- unclear if this is necessary, but we'll see
Alpine is small. That's about its only benefit. If that's the main consideration, then use it. Through usage, however, the limitations of using Alpine as the base system for a complex and dynamic environment meant to support flexible data-science efforts have started to show up. There were a lot of tweaks that needed to be done to allow the core systems to work correctly, and in order to manage the post-container, on-demand installation of python packages, a home-grown package management system (ipydeps) was created. It tries to leverage Alpine packages (apks) when it can and pip packages when it can't, but regular problems are encountered by pip packages that need libraries Alpine doesn't have, and the challenge of maintaining an independent package management system has led to questions about whether the trade-offs that have been made just for size are worth it.
This branch is an exploration of an alternative. It may not work, but should at least be attempted to determine if a small enough non-Alpine-based image can be created in a way to make the other pieces easier and allow us to leverage a python package system with a larger community (like conda).
To build the image from source, clone or download the repo. Then build with something like this:
docker build -t nbgallery/jupyter-centos:<version> <source-directory>
You will usually launch a container something like this:
docker run --rm -p 443:443 nbgallery/jupyter-centos
The default entrypoint is jupyter-notebook-secure, which will generate a self-signed certificate and then launch the jupyter notebook server under HTTPS with an automatically-generated authentication token.
Launching the container with the command:
docker run -p 8888:8888 -it jupyter-centos bash
Will launch you into a bash prompt inside the container. The shortcut command ji will launch an insecure version of Jupyter that should be accessible on your system's localhost:8888
- testing fresh updates of all the packages to see if it fixes some of the security issues in docker hub.
- attempt to fix issue #3 (permission to open port 443 as
jovyaninstead of having to be root)
- still alpha testing, but this version uses the latest versions of Jupyter notebook and tornado
- initial Alpha release rolled into the docker repo (I'll do better about tag naming moving forward)