Persist installed packages

[mindrunner]

rstudio installs packages to a folder in /home/rstudio/R which is impractical since the changes are lost after a container restart. If I try to bind mount that folder, rstudio decides to install packages somewhere else (/usr/local/lib/....).

What is the preferred way of persisting user packages?

[eddelbuettel]

You can set .libPath() in the session when you are installing, and you can also give a 'destination' argument to install.packages().

Where files persists depends on how the container was started / will be restarted so is hard to generalize. Another common solution is to create a new container: you start with the given one in a FROM ... statement and add a RUN statement with your installation step. That way you have your enlarged container as you want it.

[mindrunner]

I am hosting this for end users. When they install packages, these should automatically go into a predefined directory that is bind-mounted, so the data survives image updates, etc.

You can set .libPath() in the session when installing, and you can also pass a destdir argument to install.packages().

This is not an option, since users won’t know how to do that — it needs to be preconfigured.

Where files persist depends on how the container is started / restarted, so it's hard to generalize.

Setting - "R_LIBS_USER=/packages" didn’t help, nor did bind-mounting /home/rstudio/R.

What is the preferred way to handle this?

[eddelbuettel]

There are two aspects to your question. First, how do set options during startup of R is described exhaustively in the help page accessible via help(Startup) (and in other places). There are different ways, my preference for site-wide settings is /etc/R/Rprofile.site (and /etc/R/ is a convenience shortcut I create in r-base-core).

The second aspect is a bit more special and depends on exactly what you do / how you do it. You absolutely can persist directories across restarts, it seems you know how, and now it is a matter of working through that locally, conditional on what you do in the first step. You should be able to make this work without altering the container but worst case you can always build a local container from ours and hard-code your settings.

[mindrunner]

Turned out that a typo in my compose file resulted in rstudio silently skipping my config due to a missing folder.

For reference, here is my setup:

.Renviron

R_LIBS=/home/rstudio/R/library
R_LIBS_USER=/home/rstudio/R/library
R_DEFAULT_PACKAGES='utils,grDevices,graphics,stats'

docker-compose.yml (simplified)

services:
  studio:
    image: rocker/rstudio:4
    restart: always
    ports:
      - 8787:8787
    environment:
      - "USERID=1002"
      - "GROUPID=1002"
    volumes:
      - ./config:/home/rstudio/.config/rstudio
      - ./workspace:/home/rstudio/workspace
      - ./library:/home/rstudio/R/library
      - ./.Renviron:/home/rstudio/.Renviron:ro

Thanks for your help @eddelbuettel

[eitsupi]

docker-compose.yml (simplified)

services:
  studio:
    image: rocker/rstudio:4
    restart: always
    ports:
      - 8787:8787
    environment:
      - "USERID=1002"
      - "GROUPID=1002"
    volumes:
      - ./config:/home/rstudio/.config/rstudio
      - ./workspace:/home/rstudio/workspace
      - ./library:/home/rstudio/R/library
      - ./.Renviron:/home/rstudio/.Renviron:ro

FYI, it is considered an anti-pattern to do many bind mounts under the home directory in a container.
https://rocker-project.org/images/versioned/rstudio.html#userid-and-groupid

[mindrunner]

Btw, the example config is doing so as well, that's were I got started

services:
  rstudio:
    image: rocker/verse:4
    ports:
      - "8787:8787"
    environment:
      PASSWORD: yourpassword
    volumes:
      - ./.rstudio_config:/home/rstudio/.config/rstudio
      - ~/workspace:/home/rstudio/workspace
      - /other_dir:/other_dir

However, for example, another directory such as the one containing CSV files should not necessarily have to be under the home directory, so it is recommended to bind-mount it under its own name directly under the root, e.g. /other_dir.

It doesn't really tell why it is not a good idea

[eitsupi]

Please read the documentation carefully.
Bind mounting to the home directory with the environment variables UID and GID set may cause problems.

If these are set, ownership of the /home/rstudio directory in the container is updated by the root user. This will also overwrite the ownership of any files that are bind-mounted under the /home/rstudio directory.

[benz0li]

@mindrunner You may be interested in b-data's/my JupyterLab R docker stack which allows bind mounting the entire home directory.

[mindrunner]

Please read the documentation carefully. Bind mounting to the home directory with the environment variables UID and GID set may cause problems.

If these are set, ownership of the /home/rstudio directory in the container is updated by the root user. This will also overwrite the ownership of any files that are bind-mounted under the /home/rstudio directory.

Ah, got it. Makes sense, but not an issue in our case here. I moved packages out of the home dir now.

@mindrunner You may be interested in b-data's/my JupyterLab R docker stack which allows bind mounting the entire home directory.

Thanks, will have a look :)