Dockerfiles related to the R programming language and RStudio's Shiny Server. Greatly inspired by the Version-stable Rocker images.
The Makefile contains commands to build each of the Linux images.
Check the file to see the build arguments for Docker.
I have one folder for each image, but not for different versions.
Instead, each version of R has its own tag.
To get the files for e.g. R version 3.6.2, checkout the tag v3.6.2.
(This approach does not seem to work with automatic builds on Docker Hub, so I push selected images manually 😰)
Once upon a time I relied on Microsoft's MRAN service to install packages as they appeared for each version of R. Now that MRAN is departed I install packages using the {pptm} package to ensure compatability with the R version.
The r-minimal image only contains R.
Available on Docker Hub.
The r-base image is based on r-minimal and contains C, C++ and Fortran compilers as well as the {remotes} package.
Available on Docker Hub.
The r-test image is used as a base image for testing R packages.
The image has the {covr} package, the {devtools} package, the {roxygen2} package and the {testthat} package installed.
The working dir in this image is /home/shiny/package.
My usecase is to copy/clone a package (repository) into this folder and then test it in r-test.
The shiny image contains a Shiny server installed from source as described in Shiny server's wiki as well as the {shiny} and {rmarkdown} packages. Available on Docker Hub.
The configuration file shiny-server.conf is very basic:
It exposes the Shiny server at port 3838 and includes some example apps.
Check out Shiny server's docs to learn more about its configuration.
Example apps from Shiny server are available at <url>/sample-apps.
Example apps from the {shiny} package are available at <url>/<app name>, where <app name> is one of the following:
01_hello02_text03_reactivity04_mpg05_sliders06_tabsets07_widgets08_html09_upload10_download11_timer
To test if the Shiny server is working, run the command
docker run --rm -p <port>:3838 robertdj/shiny:<tag>
with an appropriate value of <port> (e.g. 3838) and navigate to this URL: https://localhost:<port>
The windows folder contain Dockerfiles for an r-base image and an r-test image that are counterparts to the Linux images above, but based on a Windows Server Core image.
These images have R and Rtools installed and the same packages as their Linux counterparts.
The Dockerfile for r-base contain long lines because the usual "newline character" in Windows' CMD (the caret ^) is not parsed correctly by Docker.
The Windows images are not included in Makefile, but from the folder windows/r-base the following command builds the base image with an appropriate value of <R version> and associated MRAN date
docker build --build-arg R_VERSION=<R version> --build-arg MRAN_date=<MRAN date> --tag r-base:<R version> .
The Makefile contains R versions and matching MRAN dates.
In the folder windows/r-test the following command builds the test image:
docker build --build-arg R_VERSION=<R version> --tag r-test:<R version> .
Some technical considerations.
The Linux Dockerfiles use the BuildKit to enable faster builds and the use of heredocs in the Dockerfiles.
I use Container Structure Tests to test images.
The tests for an image is the yaml file in the same folder as the Dockerfile and the commands for testing are in Makefile.
When this repo began, all images started R as a normal/non-privileged user. This is in accordance with the best practices for writing Dockerfiles. However, these best practices also recommend not to change users too often, which I did.
My current approach is that a user shiny is still available in all R images, but no images swich to shiny.
Instead, each image is considered a starting image -- use it in the FROM line of a Dockerfile and then swtich to shiny with USER shiny in the end.
The username of the non-root user is usually not important.
I use the username shiny because this is Shiny Server's default non-root user.
Why am I using Ubuntu as the Linux distribution?
- R packages (with system requirements) are usually tested on popular/larger Linux distributions, making the installation seamless (once the requirements are available).
- I use Ubuntu on my "daily laptop", so I have to figure out how to install R on Ubuntu anyway.
I have experimented with Alpine Linux, but
- It is much more work to install packages with system requirements.
- The runtime dependencies for R actually make the final images so large that they are comparable in size to the Ubuntu images.
For the r-minimal image I have scrutinized the Dockerfiles for the Version-stable Rocker images and therefore I have no choice but to use the GPL 2 license.
The other Dockerfiles are licensed under MIT.