stats-server

RStudio + RShiny

Prerequesites

make and docker are presumed to already exist on your system.

Set up

Recommended: edit add_users.sh and change the default password, add more users by copy/pasting the example user.

Quickstart:

make run

And access RStudio at http://<your-ip-or-localhost>:8787 (with RShiny running on 3838).

By default you will log in as

admin
changethispassword

Terminal

You may notice the terminal output looks incorrect (unless you are running with GPUs as described below) Under Terminal > Terminal Options, deselect Hardware Acceleration.

Adding Users

Refer to /tmp/add_users.sh to see how to create users with home directories. In short, run useradd -m -g users <username> to create the user and passwd <username> to set their password. Once added using the Terminal from the admin's RStudio session, other users will be able to log in.

By default, the admin user has sudo privileges. You can use standard UNIX users/groups organization to assign your users to various groups and manage file permissions that way.

GPU Support

You must build this image from an nvidia docker base that is associated with compatible drivers on your system. Then you can pass --gpus all to your make run command and have access to your GPUs.

Data Persistence

(PLEASE OPEN ISSUES IF YOU NEED HELP CONFIGURING THIS TO YOUR NEEDS)

Users should save data in their respective home folders.

make run by default runs an emphemeral container for demonstration purposes only.

Before doing any real work inside this environment, you need to make sure your user's data will persist.

Fortunately you can do this by creating a local home directory and bind-mounting ./home:/home/ and various other directories into the container and run it without the --rm flag when deploying the server.

The makefile entry for run would look like:

run:
		docker run -d \
		-v ./home:/home/ \
		--name stats-server \
		-p 8787:8787 \
		-p 3838:3838 \
		-p 13939:13939 \
		-p 3939:3939 \
  stats-server 

Please take time to understand docker volumes prior to using this in production.

Integration with Jupyterhub

(POTENTIAL FUTURE ADD-ON) The maintainer of this project has (in the past) successfully managed to set up Jupyterhub to manage users and proxy traffic to an RStudio server that runs in isolation for each user (as opposed to this project which runs one RStudio server for all users to share). If you are interested in this use case (perhaps managing an environment where Jupyter notebooks and RStudio are used by members of the same research team), please open an issue in ml-starter-packs/jupyterhub-deploy-docker.