Dockerfile Collection for DGX-230

XUbuntu

Online building

If you do not want to change the contents of the dockerfile, you could use such command to build the image:

docker build -t xubuntu:1.7 https://github.com/cainmagi/Dockerfiles.git#xubuntu

This image is compatible for Ubuntu 16.04, 18.04 and 20.04. Please check your base image and confirm that the Ubuntu inside the image is compatible with this dockerfile.

We provide 3 examples:

• Start from pytorch 1.12.0a image:

  docker build -t xubuntu-tc:1.7 --build-arg BASE_IMAGE=nvcr.io/nvidia/pytorch:22.03-py3 --build-arg BASE_LAUNCH=/opt/nvidia/nvidia_entrypoint.sh --build-arg JLAB_VER=3 https://github.com/cainmagi/Dockerfiles.git#xubuntu

• Start from cuda 11.6 image:

  docker build -t xubuntu-cuda:1.7 --build-arg BASE_IMAGE=nvcr.io/nvidia/cuda:11.6.2-cudnn8-runtime-ubuntu20.04 --build-arg BASE_LAUNCH="" --build-arg JLAB_VER=3 https://github.com/cainmagi/Dockerfiles.git#xubuntu

• Start from tensorflow 1.13.1 image (since it is a Python 3.5 image, we recommend to fall back to Jupyter Lab 2.x):

  docker build -t xubuntu-tf:1.7 --build-arg BASE_IMAGE=nvcr.io/nvidia/tensorflow:19.03-py3 --build-arg BASE_LAUNCH=/usr/local/bin/nvidia_entrypoint.sh --build-arg JLAB_VER=2 https://github.com/cainmagi/Dockerfiles.git#xubuntu

There are 3 available options:

Option	Description	Default
BASE_IMAGE	The base image for building this desktop image.	nvcr.io/nvidia/pytorch:22.03-py3
BASE_LAUNCH	The entrypoint script from the base image. If there is no entry script, please use"".	/opt/nvidia/nvidia_entrypoint.sh
JLAB_VER	The version of the Jupyter Lab to be installed. Could be1, 2, 3 or unset. If use unset, nothing would be installed if there is already a Jupyter Lab.	unset
JLAB_EXTIERS	The to-be-installed extra extensions for the Jupyter Lab. IfJLAB_VER is unset, nothing would be installed. To view details about which extensions would be installed, see here.	2
WITH_CHINESE	If set, the image would be built with Chinese support for vscode, sublime and codeblocks.	true
WITH_EXTRA_APPS	The installed extra applications. Each character represents an app or several apps. For example,cgo represents fully installing Cloudreve, GIMP, LibreOffice and Thunderbird. More details could be referred in the following table.	cgo
ADDR_PROXY	Set the proxy address pointing to localhost. If specified, this value should be a full address. (Experimental feature ::)	unset

Here we show the list of extra apps:

Code	Description
c	Cloudreve
p	PyCharm
g	GIMP
k	GitKraken
m	Sublime Text 4
x	TeXLive + TeXstudio
n	Nautilus + Nemo
o	LibreOffice + Thunderbird
a	Atom + Nuclide
e	GNU Emacs

To find your launch script of your base image, use

docker inspect <your-base-image>:<tag>

Offline building

Otherwise, you need to clone the branch firstly:

git clone --single-branch -b xubuntu https://github.com/cainmagi/Dockerfiles.git xubuntu

After that, run such command to build the image:

docker build -t xubuntu:1.7 xubuntu

where xubuntu is the folder of the corresponding branch. The options in online building examples could be also used for offline buliding.

Launching

When launching the image for the first time, please use the following command to configure your user id and VNC password. When you use this image for the first time, please configure your user id by:

docker run --gpus all -it --rm xubuntu:1.7 uid=$(id -u) gid=$(id -g)

Then commit the image by

docker commit --change='CMD [""]' <conatiner-id> xubuntu:1.7

• By built-in noVNC: In default mode, you just need to launch the built image by:

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal -p 6080:6080 xubuntu:1.7

  It is equivalent to use --vnc or not in the above command. However, if you have saved the image in other modes before, you may need this flag to force the image to enter the VNC mode. The --vnc option is required when you need to force the image to switch to VNC mode. The following command would force the vnc launched by root mode.

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal -p 6080:6080 xubuntu:1.7 --root

  In current version, users could use either http to get access to the unencrypted noVNC session or https to get access to the ssl-encrypted noVNC session. For users who open the encrypted session firstly, they may need to add the noVNC site into the trusted list.

• Switch the VNCServer to XTigerVNC (experimental): Add the option --xvnc will make the desktop hosted by the Xvnc program. Everything will be run in the same process. There will be no sub-process manager like tigervncserver to manage desktop related programs. A good thing is that, users do not need to run tigervncserver -kill :1 before saving the image. However, currently these desktop related programs are not guaranteed to be closed if hitting Ctrl+C. Therefore, we suggest the users to use ps -aux to validate the running processes before saving the image.

  docker run --gpus all -it --rm -v ~:/homelocal -p 6080:6080 xubuntu:1.7 --xvnc

  Certainly, there is also a root mode for this method:

  docker run --gpus all -it --rm -v ~:/homelocal -p 6080:6080 xubuntu:1.7 --rootxvnc

  After using Ctrl+C to kill the Xvnc program, users can use the following command to relaunch the Xvnc and noVNC services:

  xvnc-launch [--root]

  If adding the option --root, the desktop will be run with root privilege.

• By external VNC viewer (client): If you have installed a VNC viewer on your client side, and want to connect the VNC server of the image directly, please use:

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal -p 5901:5901 xubuntu:1.7

  The root mode could be also applied here.

• By Jupyter Lab: If you want to launch the Jupyter Lab but do not start the desktop, please use

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal -p 6080:6080 xubuntu:1.7 --jlab jlab_password=openjupyter jlab_rootdir=/homelocal

  The jlab_password would override the default random token. The jlab_rootdir is the root folder of the launched jupyter lab. If not set jlab_rootdir, the default root folder would be /homelocal. The --jlab option is required when you need to force the image to switch to Jupyter Lab mode.

• By BASH: If you want to enter the command line but do not start the desktop, please use

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal xubuntu:1.7 --bash

• By any script: If you want run any script inside the docker for only one time, please use

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal xubuntu:1.7 script=<the-path-to-your-script>

• With Cloudreve 🔗: We recommend users to launch Cloudreve by opening a new terminal on the desktop, and using the following command:

  crpasswd  # only used for checking the INITIAL admin password.
  cloudreve  # launch Cloudreve service, requires users to expose 5212 port.

  ⚠️ Using Cloudreve requires users to add the extra app c in the option WITH_EXTRA_APPS when building the image.

  ⚠️ We STRONGLY recommend users to change their admin password, and create a non-admin user for using Cloudreve. You can also configure your data exchanging folder.

  After launching the app, users can get access to Cloudreve by <dgx-ip>:5212 port. Remember to expose the port number by -p 5212:5212 when launching the container.

  If users have configured Cloudreve by the webpage, and commit the image. Then the users can launch the container only with Cloudreve (not opening the desktop):

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal -p 5212:5212 xubuntu:1.7 --cloudreve --bash

  or launch Cloudreve together with the desktop:

  docker run --gpus all -it --rm --shm-size=1g -v ~:/homelocal -p 5212:5212 xubuntu:1.7 --cloudreve

  Cloudreve will show you a lot of logs on the terminal, it may interfere the messages from Xvnc or noVNC. So we do not recommend to launch it together with the desktop. It is always better if you open a terminal on the desktop and use the cloudreve command.

• With FileBrowser 🔗: We recommend users to launch filebrowser by opening a new terminal on the desktop, and using the following command:

  fbrowser  # although users can use filebrowser to launch the app, we still recommend users to use this command, because this command can configure the IP and PORT number automatically.

  ⚠️ We STRONGLY recommend users to change the initial admin password, and keep the modified password by themselves.

  After launching the app, users can get access to FileBrowser by <dgx-ip>:5212 port. Remember to expose the port number by -p 5212:5212 when launching the container.

  Similarly, the users can also launch FileBrowser by adding the option --filebrowser when launching the container. The usage is exactly the same as Cloudreve.

  ⚠️ Note that Cloudreve should not be launched together with FileBrowser, unless you have carefully configured the launching scripts and understood what you want to do.

Features

This is the minimal desktop test based on ubuntu 16.04, 18.04 or 20.04 image, it has:

• Fully installed xfce4 desktop: it has most of the useful plug-ins for xfce4 desktop. While libreoffice and texlive are not installed.
• Fully inherit the base image: some base image may already have the entrypoint script. We provide options for including the the entry-script of base image.
• Modern VNC server: it contains tigervncserver, which is a modern VNC server and could provide more features than tightvncserver and vnc4server, like cutomizing display settings, fully implemented animated cursor and shadow effects.
• Compatible for multiple Ubuntu versions: including Ubuntu 16.04, 18.04 and 20.04.
• Useful apps: including nomacs, notepadqq, visual studio code, peazip, okular, smplayer and chrome.
• Fully installed Jupyter Lab: if user needs, a full Jupyter Lab with several extensions could be installed, the details could be checked here.
• Multiple launching method: including VNC server, jupyterlab, bash and arbitrary script mode.
• Chinese language support: for some apps including edge, chrome (chromium), firefox, vscode, kate, codeblocks, ...
• Cloudreve Service (Chinese only) 🔗: a private cloud storage service, allowing users to expose their personal folder as an "online drive" available on LAN. If users are interested, they can dig into the configurations and enable more features (like WebDAV and offline downloading). Currently this feature is designed for using a browser-based app to replace the WinSCP client.
• FileBrowser Service 🔗: an alternative of Cloudreve. It supports multi-language and is more flexible for exchanging files with a single server. Although it does not support so many online drive features like Cloudreve, with FileBrowser, users can upload / download files, share links, and even run commands (need to be added to the whitelist) easily.
• Extra scripts: we also provide some extra scripts for compiling specific libraries (like ffmpeg and gcc). These files will be convenient examples for users who want extra features.

Update records

ver 1.7 @ 4/17/2022

1. Provide Microsoft Edge in the basic tier of the desktop apps.
2. Add more extensions for Visual Studio Code.
3. Upgrade the input method fcitx to version 5 when using Ubuntu 20.04. This is a previewed version in Ubuntu 20.04.
4. Upgrade the other basic tier dependencies, including PeaZip, GitFiend, and Sublime 4.
5. Move Sublime to the extra tier, because it is not a free software. Use Kate as the replacement.
6. Support a cloud file transfer tool: Cloudreve. This tool may be able to replace the functionality of WinSCP.
7. Update the versions of packages in the icon / theme bundle.
8. Switch from Jupyter Lab 2 to Jupyter Lab 3 by default. This configuration is not recommended for those images with Python 3.5. Users may need to configure the J-lab version manually in that case.
9. Provide the Oh-my-posh (OMP) 🔗 terminal theme.
10. Provide some optional scripts in the folder /home/xubuntu (~).
11. Fix small bugs about dependency problems (in detach), and the OMP installation.
12. Fix a bug caused by get-conda-path.py. When there is neither conda nor python, the script incorrectly returned True value before.
13. Prefer a local snapshot to install VS Code extensions. This change may help avoid the Server 503 error 🔗 during the extension installations.
14. Adjust the formats of some installation scripts.
15. Provide Bing Wallpaper.
16. Wrap up of all tests and scripts. This version can be released now.

Inherit from the update of xubuntu-minimal:1.1

1. Fix a bug caused by missing of LibreOffice libs (Fixed by ~/.config/xfce4/xinitrc).
2. Enable users to run GUIs with sudo (Fixed by ~/.config/xfce4/xinitrc).
3. Make the desktop get launched properly. This fixture corrects a bug where the desktop may be launched by twice in Ubuntu 16.04 (Fixed by ~/.vnc/xstartup).
4. Fix a bug caused by using get-pip.py with Python 3.6. Since the pip may be downgraded by NVIDIA configurations, this fixture will also correct the pip version. (Fixed by install-desktop).
5. Prefer conda/mamba when updating python packages (Fixed by install-desktop).
6. Upgrade TigerVNC to 1.12.80. (Fixed by install-vnc).
7. Add more path to sudo/secure_path, now users are allowed to use conda / mamba / pip directly with sudo (Fixed by sudoers).
8. Finish the launching mode --xvnc (Fixed by docker-entrypoint and xvnc-launch).

Testing report v1.7

This docker file has been tested sucessfully on:

• nvcr.io/nvidia/pytorch:22.03-py3 (Ubuntu 20.04, python 3.8)
• nvcr.io/nvidia/cuda:11.6.2-cudnn8-runtime-ubuntu20.04 (Ubuntu 20.04)
• nvcr.io/nvidia/pytorch:19.08-py3 (Ubuntu 18.04, python 3.6)
• nvcr.io/nvidia/tensorflow:19.03-py3 (Ubuntu 16.04, python 3.5)

Finished at 4/19/2022.

❗ This testing build is configured by WITH_EXTRA_APPS=cpgkmxnoe indicating an incomplete build. The build test for Atom (a) is currently skipped, because the plugin market of Atom is currently down, see the issues atom/atom#25417 and atom/apm#946. :exclamation: The build for VS Code extensions also falls back to the local build method, because currently VS Code extension market is not stably working. See the issue microsoft/vscode#147670. Despite this situation, we still manage to install all extensions with a full snapshot of the .vsix files.

ver 1.6.1 @ 7/6/2021

1. Fix a bug caused by the upgrade of noVNC.

ver 1.6 @ 6/22/2021

1. Support the proxy value for the built image, this value is important for the devices protected by the firewall.
2. Move GitKraken to the optional packages. Instead, the default Git client is switched to GitFiend.
3. Replace the default system monitor by stacer, the previous app gnome-system-monitor is dropped.
4. Change the default configurations of XFCE4.
5. Bump Pycharm, tigervncserver to the newest versions.
6. Upgrade the Jupyter Lab script to 1.3.
7. Fix a fatal bug caused by the user authority. We may need to find a method for forwarding the current user to the docker image.
8. Fix a bug of caused by dbus initialization. In the previous version, the bug would cause strange behaviors (for example, the screen savers would not work).
9. Fix a bug caused by the changed address of get-pip.py.

ver 1.5 @ 4/10/2021

1. Enhance the launchers of system menu and panels.
2. Set the user as xubuntu, and provide the --root mode.
3. Make python version auto-detected, which means XUBUNTU_COMPAT has been removed.
4. Add options for installing some extra apps, like PyCharm and GIMP.
5. Enhance the VNC support by OpenSSL entryption.
6. Upgrade the Jupyter Lab script.
7. Add some themes.
8. Fix the compatibility problem caused by pip 21.0.
9. Fix small bugs, related to nvm path, root authority, and detach script.
10. Fix small bugs, related to jlab path and the authority of creating SSL certificate.

ver 1.4 @ 1/17/2021

1. Arrange the theme pack installations.
2. Add fully supported Jupyter Lab installation.
3. Adjust the usage of some options.
4. Add some packages for the desktop.
5. Correct the format of some launchers.

ver 1.3 @ 1/13/2021

1. Re-craft the icons and themes for ubuntu 20.04.
2. Add gcc and gfortran supports for building the image.
3. Add some packages for the desktop.
4. Add check for ubuntu version.
5. Add meta-data in the dockerfile.

ver 1.2 @ 1/12/2021

1. Add extra Chinese supports for some packages.
2. Add vscode package installation script.
3. Add exposed ports (5901, 6080).
4. Add Jupyter Lab supports.

ver 1.1 @ 1/10/2021

1. Support Ubuntu 20.04.
2. Fix the font issues.
3. Finish all testings for Ubuntu 16.04, 18.04 and 20.04.

Testing report v1.1

This docker file has been tested sucessfully on:

• nvcr.io/nvidia/pytorch:20.12-py3 (Ubuntu 20.04, python 3.8)
• nvcr.io/nvidia/cuda:11.1-cudnn8-runtime-ubuntu20.04 (Ubuntu 20.04)
• nvcr.io/nvidia/pytorch:20.11-py3 (Ubuntu 18.04, python 3.6)
• nvcr.io/nvidia/tensorflow:19.03-py3 (Ubuntu 16.04, python 3.5)

ver 1.0 @ 12/18/2020

Re-craft this dockerfile.

ver 1.0 @ 20181201

Create the dockerfile branch.