Note: This still uses (a newer) gitpod base image and thus also the gitpod user.

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 76.48%. Comparing base (75d8c83) to head (c5dd2d2).
Report is 15 commits behind head on dev.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Bit late to the party here, but note that we went through this process when switching to Devcontainers for the Nextflow Training. In the end we went all-in on the devcontainer spec and stopped thinking about base images entirely.

See the current Devcontainer spec: https://github.com/nextflow-io/training/blob/master/.devcontainer/codespaces-dev/devcontainer.json

• Uses mcr.microsoft.com/devcontainers/base:ubuntu the simplest base Devcontainer image
• Uses devcontainer features instead of Docker layers for stuff like Java, conda etc.
• Uses local features (basically bash scripts) to install stuff
• Bungs all of this into a pre-build container image for faster start-up times (workflow)

I'm not sure that we need anything in the devcontainer image that we haven't already sorted in the training image? We already have Nextflow, conda, nf-core/tools etc etc. The simplest approach would be to forget about image entirely and just use this devcontainer spec file with the training base image but mounting a different repo / workdir.

If the image being called "training" feels off then we could always move that. Feels more sensible to share effort on the setup there rather than reinventing the wheel though..

@ewels Yep, devcontainer features are great! We are using those and docker-in-docker works like a charm on codespaces 🚀
(We are now using mcr.microsoft.com/devcontainers/python:3.11)
I can add conda and nextflow as a feature as well, like it was done over at nextflow.io/testcontainers.

Just using the devcontainer from nextflow.io/training would not install the most recent version of tools after a merge, which is the purpose of this exercise, no?

But we can use their base image and reinstall tool with dev dependencies and --editable flag set?

Our image for use with devcontainer is now based on ghcr.io/nextflow-io/training:latest to remove a lot of redundancy and hopefully maintenance effort. The setup.sh had to be added to make that base image work.

The devcontainer.Dockerfile sets up the image for use with the non-root user vscode, installs apptainer and most importantly copies the current state of the tools repository for installation.

If we don't need the unprivileged user (the tutorial image just uses root) and the repository can be mounted into the container, we could move those things to the setup.sh script and drop our own devcontainer image entirely.

any update on this? I keep pointing people to codespaces and then the run into docker-in-docker issues...

I was messing around using pixi in a container to install needed tools. Seems to work OK. I haven't got it fluid yet though.

https://github.com/mahesh-panchal/Nextflow_sandbox

Feel free to play around. Just FYI.