Use Rocker based R images

Use Rocker based R images#

The Rocker project is a community led set of containers for use in the R ecosystem, similar to the jupyter/docker-stacks images for the python ecosystem. They contain hand-picked useful libraries determined by the R Community, along with a working RStudio setup. We should treat them as meaningful upstream for users who want an environment that is primarily R and RStudio.

Why?#

Why use the Rocker images when jupyter/r-notebook exists? A few reasons!

  1. r-notebook does not have RStudio, and R users primarily use RStudio. It is very important for us, from the JupyterHub community, to respect this choice and support that to the extent possible with JupyterHub, rather than try to get them to move to Jupyter notebooks. Being able to use RStudio as well as Jupyter Notebooks is one of JupyterHub’s great selling points, and we should capitalize!

  2. r-notebook installs R packages from conda, while Rocker uses the Posit Package Manager CRAN snapshot. Both these provide binary packages, but the latter is multiple orders of magnitude more popular in the R ecosystem. Most R users install packages via install.packages or devtools, and those all work from CRAN with Posit Package Manager, not conda. So for an R native experience, we should continue to use Posit Package Manager’s CRAN archive.

  3. The Rocker team is well respected within the R community, and they pick the base set of libraries to go into these images. This lets us provide a ‘batteries included’ experience to our end users with minimal need to inherit images.

Picking the image to use#

If we want to use the upstream image directly, rocker/binder is the appropriate image to use, as it has JupyterHub, jupyter-rsession-proxy, and whatever else is needed to work on a JupyterHub. It inherits from rocker/geospatial, which contains:

  • tidyverse set of packages

  • A working LaTeX installation, set up to be usable with R’s publishing infrastructure

  • Geospatial packages

The version tags used refer to R versions only, and the upstream repo is only tagged for each new R version.

Hub configuration#

Default user and home directory#

The rocker project uses rstudio as the default user, with uid 1000, for non-root access. This is slightly different from repo2docker or the jupyter/docker-stacks or pangeo-stacks images, which use the jovyan user with uid 1000. This is mostly fine in most people’s cases, but because we explicitly set the user home directory to /home/jovyan in basehub/values.yaml, we will need to explicitly set user’s home directory to /home/rstudio for rocker based images. There isn’t a problem with folks mixing rocker and other images in the same home directory though, as the uid is the same. While this name difference does cause some extra config to be required, we should stick to it as part of respecting the other community’s norms.

Default URL to open#

Mostly, we would want the hub to put users directly into RStudio after they log in. Thanks to the jupyter-rsession-proxy, this just requires us to set singleuser.defaultUrl or Spawner.default_url traitlet config.

Configuration without profileList#

When not using profileLists, the hub can be configured like this:

jupyterhub:
  hub:
    config:
      KubeSpawner:
        # Ensures container working dir is homedir
        # https://github.com/2i2c-org/infrastructure/issues/2559
        working_dir: /home/rstudio
        # Launch into RStudio after the user logs in
        default_url: /rstudio

Configuration with profileList#

When using profileLists, it should probably look like this:

jupyterhub:
  singleuser:
    profileList:
      - display_name: "Some Profile Name"
        profile_options:
          image:
            display_name: Image
            choices:
              geospatial:
                display_name: Rocker Geospatial
                default: true
                slug: geospatial
                kubespawner_override:
                  image: rocker/binder:4.3
                  image_pull_policy: Always
                  # Launch into RStudio after the user logs in
                  default_url: /rstudio
                  # Ensures container working dir is homedir
                  # https://github.com/2i2c-org/infrastructure/issues/2559
                  working_dir: /home/rstudio
              scipy:
                display_name: Jupyter SciPy Notebook
                slug: scipy
                kubespawner_override:
                  image: quay.io/jupyter/scipy-notebook:2024-03-18

User installed packages#

In R, users can install packages with install.packages or similar R commands. Rocker is configured to automatically install packages from the Posit Package Manager, which provides fast binary package installs.

They are installed under /usr/local, and hence cleared on server stop. So installed packages will not persist across user sessions.

If users need to have persistent packages installed, we would need a custom image.

Contents