Jupyter / JupyterHub

GWDG offers JupyterHub as a service for users of Python, Julia or R.

A valid GWDG Full Account.

The service is intended for users in education to be used in seminars and lectures or by individual users who want to learn and try out the respective programming languages and their tools.

For larger use cases and when working with large datasets, complex models or parallel computations it is recommended to use the JupyterHub service of the SCC.

Jupyter makes it possible to work interactively with Python, Julia or R with only a browser. Source Code is written, executed and edited directly in the user's browser. This happens in a so called “notebook”. Each notebook has a “kernel” which determines the type of notebook, which can be a Python, Julia or R notebook.

JupyterHub is the portal users log in to and start and manage their notebooks and associated files.

Please note that you can also use Jupyter on our SCC if you do need direct access to data on the SCC and have medium to high compute resource requirements.

Storage and calculation of notebooks happens server side, the client does not need to install any software or meet any other prerequisites other than having a fairly modern browser to work with. To log into the service a GWDG full account is required.

After successfully logging in a selection screen appears with a choice of notebook images to start the notebook server with:

• GWDG default image (based on jupyter/datascience-notebook)
  • This was also the default image in the past.
• Python Stack w/ TensorFlow (jupyter/tensorflow-notebook)
• Python and R Spark Jupyter Notebook (jupyter/all-pyspark-notebook)
• Data Science Jupyter Notebook (jupyter/datascience-notebook)

The notebook image provides the environment for the notebook server, in particular the pre-installed software. While the GWDG default image is heavily extended from the regular data science notebook the regular notebooks from the Jupyter project may be preferable in some cases or provide a more specialized environment for a specific software set.

Irrespective of the selected image the user's home directory and data remains the same.

The notebook image can only be changed when the current notebook server is stopped and restarted. The server does not automatically stop when logging off or closing the browser, although this will cause it to timeout after a while and then stop. The server can be explicitly stopped from the menu File → Hub Control Panel → “Stop my server”. This actually stops the running server which can then be restarted via “Start my server” and a new image can be selected.

A simple changelog with changes to the “GWDG default image” is available.

After successful login at Jupyter-Hub there is a drop down menu at the top left corner. Under „File - New“ a new notebook can be created. Previously used notebooks and their files are listed on the left hand side.

Detailed information about the user interface: https://jupyterlab.readthedocs.io/en/stable/index.html

An active notebook can be closed through the “File - Close and shutdown notebook” menu or by closing the browser window or tab. The actual notebook is preserved in its last state and can be launched again from the overview page of Hub.

Deleting a notebook is done through the directory listing - left click - delete.

• ⚠ Please note that all directories except the home directory are volatile and will be lost when the notebook server is closed.
• A maximum of 50 GB disk space and 10 GB RAM can be used.
• jupyter-cloud.gwdg.de is not suitable for continuous computations over multiple days.

If, after upgrading packages / installing new packages / installing a new kernel, it is no longer possible to start a notebook or the kernel can not connect, it can help to rename the '.local' folder in order to restart. Please note that all custom kernels and locally installed packages will get deactivated:

• File - New - Terminal

•  mv -v .local/ .local.gwdg-disable 

• Restart notebook server: File - Hub Control Panel - Stop My Server

Additional Python modules can be installed via the terminal and the Python package manager “pip”. To do this, a terminal must be opened via the menu “New” → “Terminal”. Afterwards

pip install --user <module>

installs a new module in the home directory.

The installation of large Python modules like “tensorflow” may fail with a message “No space left on device”. This is caused by the temporary space under “/tmp” being too small for pip to work the downloaded packages. The following steps use a temporary directory in the much larger user home directory for this one installation:

mkdir -v ~/.user-temp
TEMP=~/.user-temp pip install --user <module>

Prefixing the installation with the TEMP variable makes pip use that location for this one installation.

If every notebook fails to start after a package installation or upgrade the issue can be resolved by renaming the folder .local Please note that all custom kernels and local installed python packages will be disabled:

• File - New - Terminal

•  mv -v .local/ .local.gwdg-disable 

• Afterwards the notebook server should be restarted: File - Hub Control Panel - Stop My Server

mv -v .local/ .local.gwdg-disable

Management of software packages and environments with Conda/Mamba requires a terminal session started from the notebook server. The terminal ist available after login via New → Terminal.

Before working with conda or “mamba” commands the necessary conda functions need to be loaded (mind the dot at the beginning!):

. /opt/conda/etc/profile.d/conda.sh

The following describes the creation of a new, simple environment wikidoku, the installation of the package jinja2 as an example and how to make the environment available in the kernel selection of the notebook.

Creating and activating the environment:

mamba create -y --prefix ./wikidoku
conda activate ./wikidoku

As an example the package jinja2 will be installed next. This is the step to install the desired software packages from various Conda channels.

mamba install -y jinja2

Next the new environment will be registered with the notebook. Terms for –name and –display-name (optional) can be chosen freely, the later being shown in the kernel selection of the notebook. The second command lists all known environments. The last command exists the environment.

python3 -m ipykernel install --user --name wikidoku --display-name "Python (wikidoku)"
jupyter kernelspec list
mamba deactivate

If installation of the kernel fails with the message /usr/bin/python: No module named ipykernel the additional package jupyter needs to be installed and installation of the kernel repeated:

python3 -m pip install jupyter

After installation of a new environment it is recommended to restart the notebook server. Leave all existing terminals and close all open notebooks. In the Jupyter overview page click on Control Panel and stop the server by clicking Stop My Server and restart it with Start My Server.

Via New → a new notebook with the new environment can be started. In existing notebooks the kernel can be changed after starting the notebook and selecting Kernel → Change Kernel.

Installing a new, independent Python kernel für the current environment is possible. As an example an older Python 2.7 kernel will be installed next.

A new environment needs to be created and activated as per the steps above. Next follows the installation of the kernel, the jupyter module for that kernel and finally the kernel is made available for selection in the kernel list.

mamba install -y python=2.7
python3 -m pip install jupyter
python3 -m ipykernel install --user --name oldpython --display-name "Python 2.7 (oldpython)"

The new kernel is now available for new and existing notebooks after restarting the notebook server. The current kernel version can be queried from within Python:

import sys
print (sys.version)

In order to remove an environment it has to be de-registered from the notebook server and then its files removed (optional but recommended). We list the installed kernels, de-register and remove the environment's files:

jupyter kernelspec list
jupyter kernelspec remove wikidoku
rm -rf ./wikidoku

1) create a file "/home/jovyan/.Renviron" with 2 lines:
"R_LIBS_USER=/home/jovyan/R/library" and "TMPDIR=/home/jovyan/tmp"

execute the following in a terminal ("New" -> "Terminal"):

2) mkdir -p ~/R/library; mkdir ~/tmp
3) R
4) source("https://bioconductor.org/biocLite.R")
5) biocLite()

This is because R downloads and installs packages to and from the default tmp directory,
from which it cannot execute files. Using a tmp directory inside the home directory solves
this problem.

How to install packages from Github (in R):

1) library(devtools)
2) options(unzip = "internal")
3) install_github("repo/package")

In order to facilitate access to larger amounts of data on jupyter-cloud.gwdg.de, the Unix / Linux home directory can be used. To do this, data is transferred using the rsync tool. Here is an example that needs to be adapted to the user's environment:

Open a new jupyter terminal via the menu “New” → “Terminal”

jovyan@0d5793127e96:~$ ls mynotebooks/
myfile.txt
jovyan@0d5793127e96:~$ rsync -av ~/mynotebooks/ bbrauns@login.gwdg.de:/usr/users/bbrauns/mynotebooks/
sending incremental file list
./
myfile.txt

sent 145 bytes  received 44 bytes  75.60 bytes/sec
total size is 12  speedup is 0.06
jovyan@0d5793127e96:~$

If necessary, the respective ssh private key must be stored in the .ssh / directory in jupyter-cloud. The associated ssh public key must also be available on login.gwdg.de.

For accessing the data in the Unix / Linux home directory from a Windows machine, see: Samba Server

Open a new jupyter terminal via the menu “New” → “Terminal”

pip install pipenv --user
mkdir myproject
cd myproject
export PATH=~/.local/bin/:$PATH
pipenv --python /usr/bin/python3.6 #needed because of conda
pipenv install ipykernel networkx
pipenv shell
ipython kernel install --user --name=projectname

• Stop and restart server via control panel
• Afterwards “projectname” is usable as new kernel

!experimental!

The jupyter docker stacks image sets the variable JULIA_DEPOT_PATH to the path /opt/julia. However, this is volatile, since only the home directory is kept persistent. The following describes the installation of a new julia kernel, which has its package directory pointed to the home directory:

Start terminal  
Temporarily change julia package directories:  

export JULIA_DEPOT_PATH=/home/jovyan/.julia-depot
export JULIA_PKGDIR=/home/jovyan/.julia-depot

Create directory for custom packages and new julia kernel:

> mkdir /home/jovyan/.julia-depot
> julia
julia > # switch to pkg with ']' character
pkg > add IJulia # switch back to julia with CTRL+C
julia > using IJulia
	   installkernel("My-Julia-kernel", env=Dict("JULIA_DEPOT_PATH"=>"/home/jovyan/.julia-depot"))
Restart notebook server
Create new notebook with "My-Julia-kernel" kernel
Add package example: 

using Pkg
Pkg.add("DataFrames")

See: public folder.