Ubuntu desktop Docker container image with VNC

Ubuntu 16.04 container running the XFCE window manager with GPU support (via nvidia-docker).

The Docker image is installed with the following components:

• Desktop environment Xfce4
• VNC-Server (default VNC port 5901)
• noVNC - HTML5 VNC client (default http port 6901)
• Browsers:
  • Mozilla Firefox

Build

Build the docker container via:

docker build -t ubuntu-vnc-xfce4 .

Usage

The vnc-server is exposed on port 5901 and vnc web access port 6901. The option --net=host is used below, alternatively use -p 5901:5901 -p 6901:6901. It is possible to use the container as root, which is highly discouraged.

nvidia-docker run --rm -d -t --net=host ubuntu-vnc-xfce4  # as root DON'T DO THIS

Running as root is not recommended. A better approach is to run as a user with home directory mounted.

nvidia-docker run --name=mydesk -d -t -u $(id -u):$(id -g) -e USER=$USER \
  -e HOME=$HOME -v $HOME:$HOME --net=host -w $PWD ubuntu-vnc-xfce4

To stop the container when done using it run:

docker stop mydesk && docker rm mydesk

When the home directory is mounted into the container the xfce4 desktop settings will persist when the container is restarted.

Download a VNC client for your system from RealVnc, or use a web-browser.

=> connect via VNC viewer hostip:5901, default password: vncpassword

=> connect via noVNC HTML5 client: http://hostip:6901/?password=vncpassword

Within the container once connected with VNC or noVNC, run nvidia-smi command to see the GPUs. If the nvidia-smi command fails with message:

NVIDIA-SMI couldn't find libnvidia-ml.so library in your system...

Then run ldconfig command as follows:

docker exec -it -u root mydesk bash -c 'ldconfig'

Afterwards the nvidia-smi command should work within the container in the VNC session.

Troubleshooting

If the mouse is not working in noVNC on touchscreen PCs disable "Touch Events API" in Chrome chrome://flags.

Hints

Override VNC environment variables

The following VNC environment variables can be overwritten at the docker run phase to customize your desktop environment inside the container:

• VNC_COL_DEPTH, default: 24
• VNC_RESOLUTION, default: 1280x1024
• VNC_PW, default: my-pw

The resolution can be changed after the container is launched via xrandr as well.

1) Example: Override the VNC password

Overwrite the value of the environment variable VNC_PW. For example in the docker run command:

nvidia-docker run -d -t --name=mydesk --net=host \
  -u $(id -u):$(id -g) -e HOME=$HOME -e USER=$USER -v $HOME:$HOME \
  -e VNC_PW=my-pw \
  -w $HOME ubuntu-vnc-xfce4

2) Example: Override the VNC resolution

Overwrite the value of the environment variable VNC_RESOLUTION. For example in the docker run command:

nvidia-docker run -d -t --name=mydesk --net=host \
  -u $(id -u):$(id -g) -e HOME=$HOME -e USER=$USER -v $HOME:$HOME \
  -e VNC_RESOLUTION=800x600 \
  -w $HOME ubuntu-vnc-xfce4

Or use xrandr once inside the container VNC session.

Experimental: Running docker from VNC container

Using a few additional options when starting up this VNC container it is possible to launch other docker containers. For example if you are using the VNC container as an IDE with IDE tools, but woud like to then run a docker container. Docker enabled VNC container command:

nvidia-docker run --name=mydesk -d -t -u $(id -u):$(id -g) -e USER=$USER \
  -e HOME=$HOME -v $HOME:$HOME --net=host \
  --hostname $(hostname)_mydesk \
  --group-add $(stat -c %g /var/run/docker.sock) \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -w $HOME ubuntu-vnc-xfce4

Once the container is launched and connected with VNC, open a terminal and run a container command. Example:

nvidia-docker run --rm --name=tf_test \
    tensorflow/tensorflow:1.2.1-devel-gpu python -c '
import tensorflow as tf
print("TF version: {}".format(tf.__version__))
from tensorflow.python.client import device_lib
local_device_protos = device_lib.list_local_devices()
gpus_list = [x.name for x in local_device_protos if x.device_type == "GPU"]
print("Available GPUs: {}".format(gpus_list))'

Or an interactive session in a container in the VNC. Open a terminal and run:

tmpdir=${HOME}/tmp
mkdir -p $tmpdir
# Trick to make yourself appear in the container and run with your uid/gid.
getent group > ${tmpdir}/group
getent passwd > ${tmpdir}/passwd

nvidia-docker run --rm --name=tf_test -ti \
    -u $(id -u):$(id -g) -e HOME=$HOME -e USER=$USER -v $HOME:$HOME \
    -v ${tmpdir}/group:/etc/group -v ${tmpdir}/passwd:/etc/passwd \
    --hostname $(hostname)_tftest \
    tensorflow/tensorflow:1.2.1-devel-gpu

The containers being launched from the VNC container are external to the vnc container itself in the sense that a directory not mapped into the vnc container can still be mapped into the container being launched. Say there is a directory /datasets that was not mounted during vnc-container launch. That directory can still be mapped to the containers. Example:

tmpdir=${HOME}/tmp
mkdir -p $tmpdir
# Trick to make yourself appear in the container and run with your uid/gid.
getent group > ${tmpdir}/group
getent passwd > ${tmpdir}/passwd

nvidia-docker run --rm --name=tf_test \
    -u $(id -u):$(id -g) -e HOME=$HOME -e USER=$USER -v $HOME:$HOME \
    -v ${tmpdir}/group:/etc/group -v ${tmpdir}/passwd:/etc/passwd \
    -v /datasets:/datasets \
    tensorflow/tensorflow:1.2.1-devel-gpu bash -c \
'ls -l /datasets'