🤔 Problem

It is currently not obvious to new users how they should go about creating compute environments. They can use the basic admin-provided ones to get started, but quickly they want to move beyond those.

However, even with the ones we provide there is a glaring problem: package installations do not survive hibernation.

We need to address these shortcomings to provide a smoother on-boarding of new users and projects.

We want something (a Point of View) ready for the Oct 24 Hackathon!

🍴 Appetite

3 weeks

🎯 Solution

The goal of this build is to try and understand how much we can do with existing tools with good documentation sprinkled on top. One example is using template repositories that someone can use to quickly bootstrap a versioned environment build (see https://github.com/rokroskar/compute-environment-template).

There are a few things to consider that must be accounted for:

1. The front-end needs to be baked into the image for the time being (consider using devcontainer features for common front-ends like Jupyter)
2. There should be minimal exotic modifications and hacks (i.e. the image should work on renkulab and stand-alone)
3. Assume that users won’t have a deep knowledge of Docker
4. All image modifications should be on top of standard, well-supported images, not our custom-built and maintained image library (e.g. not from https://github.com/SwissDataScienceCenter/renkulab-docker)

🚞 User stories / journeys

Required