Container images for PAIS Coder workspaces, built by CI and published to GitHub Packages (GHCR).
| Image | Tag | Purpose |
|---|---|---|
ghcr.io/drai-inn/pais-images/datasci-cpu |
latest, sha-<git>, datasci-cpu-v* |
General Python research / data-science workbench base. Ships conda (miniforge) + uv only; researchers build their own environments. Used by the pais-datasci-cpu Coder template. |
CI (.github/workflows/build.yml) builds on push to main (when the image or workflow changes)
and on datasci-cpu-v* tags, pushing to GHCR. Packages are public, so the cluster pulls with
no image pull secret.
To rebuild manually: run the build workflow via Actions → build → Run workflow.
FROM codercom/enterprise-base:ubuntu (the base pais-cpu uses), plus:
- miniforge (conda) at
/opt/conda. New envs land in~/.conda/envs, on the workspace home PVC, so they persist across restarts. - uv at
/usr/local/bin. Cache defaults to~/.cache/uvon the home PVC. - CLI tools:
tmux(so long jobs survive a web-terminal reload),git-lfs,gfortran+pkg-config+cmake(source builds of scientific packages),ncdu(find what is filling the home PVC),pandoc(nbconvert to non-HTML formats; no LaTeX, so no PDF),ripgrep,fd,tree,zip. - Shell defaults in
/etc/bash.bashrc, not~/.bashrc, because the home PVC masks the latter: the bash-completion loader (plusuv/uvxcompletions) andalias ls='ls --color=auto'. - Timezone
Pacific/Auckland(TZplus/etc/localtime); the base image is UTC.
Not included by design: ML/data-science libraries (build your own env) and CUDA/GPU support (a
future datasci-gpu image).