Running Rstudio Server in a Conda Environment

I usually rely on the conda package manager to manage my environments during development. Thanks to conda-forge and bioconda most R packages are now also available through conda. For production, I convert them to containers as these are easier to share.

Unfortunately, there seems to be no straightforward way to use conda envs in Rstudio server. This repository provides three approaches to make rstudio server work with conda envs.

Running Rstudio Server in a Singularity Container
Running Rstudio Server in a Docker/Podman Container
Running Rstudio Server locally

Running Rstudio Server with Singularity

With this approach Rstudio Server runs in a Singularity container (based on rocker/rstudio).
The conda environment gets mounted into the container - like that there's no need to rebuild the container to add a package and install.packages can be used without issues. The container-based approach has the following benefits:

Authentication works (#3)
Several separate instances of Rstudio server can run in parallel, even without the Pro version.

Prerequisites

Singularity
conda or mamba

Usage

Clone this repository

git clone git@github.com:grst/rstudio-server-conda.git
cd rstudio-server-conda/singularity

Activate the target conda env or set the environment variable CONDA_PREFIX to point to the location of the conda env.
Check the run_singularity.sh script. In particular, you may need to add additional bind mounts (e.g. a global data directory).
Execute the run_singularity.sh script. It will automatically build the container if it is not available.
```
PORT=8787 PASSWORD=notsafe ./run_singularity.sh
```
Log into Rstudio
- open rstudio server at http://localhost:8787 (or whatever port you specified)
- login with your default username and the password you specified via the PASSWORD environment variable.

Running Rstudio Server with Podman/Docker

This approach is similar to Singularity, but uses Docker or Podman and a docker-compose.yml file instead.

Known limitations

No access to shared group directories (#14)

Prerequisites

Usage

Clone this repository

git clone git@github.com:grst/rstudio-server-conda.git

Build the rstudio container (fetches the latest version of rocker/rstudio and adds some custom scripts)
```
cd rstudio-server-conda/docker
docker-compose build     # or podman-compose
```

Copy the docker-compose.yml file into your project directory and adjust the paths.

You may want to add additional volumes with your data.

[...]
   ports:
      # port on the host : port in the container (the latter is always 8787)
      - "8889:8787"
    volumes:
      # mount conda env into exactely the same path as on the host system - some paths are hardcoded in the env.
      - /home/sturm/anaconda3/envs/R400:/home/sturm/anaconda3/envs/R400
      # Share settings between rstudio instances
      - /home/sturm/.local/share/rstudio/monitored/user-settings:/root/.local/share/rstudio/monitored/user-settings
      # mount the working directory containing your R project.
      - /home/sturm/projects:/projects
    environment:
      # password used for authentication
      - PASSWORD=notsafe
      # repeat the path of the conda environment (must be identical to the path in "volumes")
      - CONDAENV=/home/sturm/anaconda3/envs/R400

Run your project-specific instance of Rstudio-server
```
docker-compose up 
```
Log into Rstudio

Open your server at http://localhost:8889 (or whatever port you specified)
Login with the user rstudio (when using Docker) or root (when using Podman) and the password you specified in the docker-compose.yml. If you are using Podman and login with rstudio you won't have permissions to access the mounted volumes.

Running Locally

With this approach a locally installed Rstudio server is ran such that it uses the conda env.

Known limitations

no authentication (#3). Use this approach only in a secure network!

Prerequisites

rstudio server installed locally
conda or mamba

Usage

Clone this repo

git clone https://github.com/grst/rstudio-server-conda.git

Run rstudio server in the conda env

cd rstudio-server-conda/local
conda activate my_project
./start_rstudio_server.sh 8787  # use any free port number here.

Connect to Rstudio

You should now be able to connect to rstudio server on the port you specify. If an R Session has previously been running, you'll need to rstart the Rsession now.

Obviously, if your env does not have a version of R installed, this will either not work at all, or fall back to the system-wide R installation.

How it works

Rstudio server, can be started in non-daemonized mode by each user individually on a custom port (similar to a jupyter notebook). This instance can then run in a conda environment:

> conda activate my_project
> /usr/lib/rstudio-server/bin/rserver \
   --server-daemonize=0 \
   --www-port 8787 \
   --rsession-which-r=$(which R) \
   --rsession-ld-library-path=$CONDA_PREFIX/lib

To avoid additional problems with library paths, also rsession needs to run within the conda environment. This is achieved by wrapping rsession into the rsession.sh script. The path to the wrapped rsession executable can be passed to rserver as command line argument.
```
rserver # ...
    --rsession-path=rsession.sh
```
When using multiple users a unique secret-cookie-key has to be generated for each user. The path to the secret cookie key can be passed to rserver as a command line parameter.
```
uuid > /tmp/rstudio-server/${USER}_secure-cookie-key
rserver # ...
  --secure-cookie-key-file /tmp/rstudio-server/${USER}_secure-cookie-key
```