A starter Python 3 and R environment for data science tasks performed on the server-side using a wide array of development tools for Python 3 and R:

• command-line tools (such as python3 / ipython, R), accessible via SSH clients or docker exec command,
• classic GUI-based IDEs (Spyder, PyCharm CE, and RStudio), designed for local use and thus requiring a graphical Desktop (here: Lubuntu/LXDE - a lightweight alternative to the standard Unity desktop), accessible inside the container through a desktop server (here: the NoMachine server, based on NX, a protocol superior to VNC).

• if you have linked GitHub with the Docker Hub, than create an automated build in the Docker Hub for this Docker image (see automated builds on Docker Hub), so that simply changing the Dockerfile in GitHub repo will automatically build and push the image to the Docker Hub,

• if you don't have automated builds configured in Docker Hub, or if your have cancelled a build in Docker Hub, then you need to build the image locally (possibly cloning the GitHub repo first):

git clone https://github.com/mirekphd/docker-nomachine-desktop.git
cd docker-nomachine-desktop
docker build --tag=mirekphd/docker-nomachine-desktop .

• if automated build is not enabled or not up-to-date (it happens) then the manually built local image has to be pushed to the Docker Hub:

# (note: separate tagging step can be omitted if already done during building with the "--tag" option)
docker tag <your_image_id_SHA256_hash> mirephd/docker-nomachine-desktop
docker login
docker push mirekphd/docker-nomachine-desktop

• on your deployment server execute docker pull:

docker pull mirekphd/docker-nomachine-desktop

docker run -d --rm -p 4000:4000 -p 22:22 --name docker-nomachine-desktop -v /home/nomachine:/home/nomachine --memory-reservation 8G --cap-add=SYS_PTRACE mirekphd/docker-nomachine-desktop:latest

• the -d option will run the container in the background (returning control to the shell at the cost of hiding errors messages displayed inside the container)
• the -rm option will remove the docker image and other objects to release memory after the container is stopped (caution: potential data loss of the data stored inside the container)
• the -v option maps en external storage folder to the internal container folder, thus: -v /external/folder:/internal/folder
• the -p option sets up port forwarding: contenerized SSH and NX servers use their standard ports, but non-standard ports are exposed outside the contained (here incremented by one); these exposed ports were defined in the Dockerfile
• the --memory-reservation option specifies the soft limit on the memory use, an indication of intended memory usage rather than a maximum cap (the container is still allowed to use as much memory as it needs); note that this can be later changed using docker update
• the --cap-add option grants additional priviledges to the container and manages quotas (e.g. memory and CPU quotas, CPU pinning), see Limit a container's resources for details
• on Ubuntu 16.04 (and later), it is absolutely necessary to enable PTRACE capabilities required by NoMachine, because they are not provided by the default docker AppArmor profile - hence the --cap-add=SYS_PTRACE parameter (see Build and Deploy NoMachine Desktops and Applications in Docker for Linux

Caution: avoid saving data inside the container, because this storage is not persistent and will be lost at next docker stop.

To obtain persistent storage outside the ephemeral container, you can map an external folder (e.g. '/host/folder') to a user folder which has been created inside the container (here called '/home/nomachine/' after the default 'nomachine' user) when running the container with the -v option:

docker run ... -v /host/folder:/home/nomachine ...

Caution: you must first grant the unpriviledged user (defined in the Dockerfile by NX_UID) write access to the host directory about to be mapped. For instance, if NX_UID=1000, then execute the following command in the host system (notice the -R (recursive) switch to change also ownership for all subfolders):

sudo -R chown 1000:1000 /host/folder

The container can run as standard user, which you can verify by running it as user (forcing docker to use its default UID of 1000 at run time):

docker run ... -u 1000 ...

The container user (nomachine) does belong to sudoers, but for very narrowly defined situations. Only server startup and logging had to be executed using paswordless sudo narrowly restricted to these two operations. This can be verified by trying to use sudo on other, generic operations (e.g. sudo apt-get install mc), which will ask for password and then be rejected as not permitted.

Security can be further hardened by restricting its capabilities to an absolute minimum (which in case of NoMachine turns out to be very wide), and dropping all others (with --cap-drop=all):

docker run -d --rm -p 4000:4000 --name docker-nomachine-desktop -u 1000 --cap-drop=all --cap-add=SYS_PTRACE --cap-add=chown --cap-add=dac_override --cap-add=dac_read_search --cap-add=fowner --cap-add=fsetid --cap-add=kill --cap-add=net_admin --cap-add=setgid --cap-add=setuid --cap-add=sys_admin --cap-add=sys_nice mirekphd/docker-nomachine-desktop:latest

• the -e option of docker run is used to define container's environmental variables USER and PASSWORD
• USER: the SSH/NoMachine user login
• PASSWORD: the SSH/NoMachine password
• these credentials are needed only to allow an already authorised server user to execute applications inside the Docker container, so passing password in clear text - as argument of docker run (e.g. docker run .. -e USER=nomachine) is acceptable

Download NoMachine client and install it on the statistical server running Docker machine

• IP: localhost
• port: 4000 (as defined by EXPOSE in the Dockerfile)
• user: nomachine
• password: nomachine
• protocol: NX
• (note that all connection details except the IP can be changed in the Dockerfile)

• IP: localhost
• port: 22 (as defined by EXPOSE in the Dockerfile)
• user: nomachine
• password: nomachine
• (note that all connection details except the IP can be changed in the Dockerfile)
• example connection:

ssh nomachine@localhost -p 22	

docker exec -it docker-nomachine-desktop bash

Note: -it = interactive terminal.

• find out the current name of the running containers (if you used --name, then this is not needed, because name is known):

docker ps -a

Note: the -a option will list also the stopped (and failed) containers.

• stop the container:

docker stop docker-nomachine-desktop

• if a subsequent docker run attempt fails, the container has to be removed with force (-f):

docker rm -f docker-nomachine-desktop

• if the -rm option was not supplied to docker run, we need to remove all unwanted objects: containers, networks, images and optionally (--volumes) also volumes (caution: they may contain unsaved data):

docker system prune [--volumes]