User Tools

Site Tools


doku:jupyterhub

JupyterHub

The VSC offers a JupyterHub service available for all VSC users at https://jupyterhub.vsc.ac.at

Login works with any cluster user (both VSC3 and VSC4) and uses the OTP like on the cluster. A VPN connection is not needed.

Also make sure to checkout the FAQ to see if your question was already answered.

Important Information: Resource Allocation

The started Jupyter Server instance runs on a separate hardware partition ('jupyter') in each cluster in order to not interfere with the regular operation of VSC.

Since the machines in this partition are shared between all users it is important to be aware of the memory requirements of the processing you are doing.

In order to help the Jupyter users the VSC Team recently added a Memory Resources Monitor plugin to the JupyterLab IDE. It is displayed in the right upper corner of the JupyterLab IDE and shows the current memory consumption of the Jupyter Server instance.

If your job goes out of memory it will affect your job as well as other users running on the same node. This is a hardware/system limitation of how slurm and linux work together. So make sure to have the right amount of memory selected for your purposes.

Important

Always make sure to select the right amount of memory suited to your requirements.

Start a new Jupyter server

choose a profile:

  • VSC-4 Singularity Image (this is the default)
  • VSC-3 Singularity Image
  • VSC-3 GPU nodes - a singularity image with CUDA support (currently only GTX1080)
  • VSC-4 python venv - a python virtual environment
  • If you are participating in a training, there is most likely only a single fixed profile or special training profiles available

Note: You need a VSC4 user for the VSC4 profiles and a VSC3 user for the VSC3 profiles. You also need to have logged into the respective cluster via SSH at least once after getting a username.

In all profiles, you can choose the IDE, either Jupyter Notebook or the more modern Jupyter Lab (which is the default)

For all the singularity profiles, you can choose between a predefined image (default) or a custom image:

  • Predefined image gives a dropdown list of different singularity images. Each image shows a short description, a version, an author and a link to the image. The GPU option only has one predefined image. See also some examples here
  • Custom image shows a line where the complete path to the image has to be entered. Note that the image has to reside on the cluster already. Web-upload is not possible at the moment.

Note: training profiles usually do not allow for customization.

Independent of the profile you can choose the number of CPUs, size of the RAM and the maximum running time.

After everything is selected, press start - your server will start and show you the JupyterLab interface or the simpler Notebook interface. In both cases, there is a list of all files in your home directory, which includes previously created notebooks.

Working with JupterLab IDE

Simply click on the desired notebook in your list - it will open in the main panel.

On first use, no notebook is present yet. To create one (or to create new ones), go to File → New → Notebook. You will be asked what kernel to use (the default is Python3): choose select and the notebook is started. It will show up as untitled in the file list, and can be renamed via right-click.

It is possible to log out of the system and leave the server running with File → Log Out. Keep in mind that this also keeps the selected resources blocked for other users.

To stop the server, user FileHub Control Panel

On the page that opens click the button labelled 'Stop My Server'.

Working with Jupyter Notebook IDE

If you have chosen to work with the Jupyter Notebook IDE, simply click on the desired notebook in your list - it will be opened in a new tab.

On first use, no notebook is present yet. To create one (or to create new ones), click on 'new' in the upper right area above the file list and choose the appropriate type. The new notebook will start in a new browser tab. It will show up as untitled in the file list, and can be renamed by selecting the checkbox to the left of the name and choose 'rename' at the top.

It is possible to log out of the system and leave the server running with the 'Logout' button at the top right. Keep in mind that this also keeps the selected resources blocked for other users.

To stop the server, click on 'Control Panel' at the top right and choose 'Stop My Server'.

BYOI: Bring your own (Singularity) Image

It is possible to use a custom singularity image with our JupyterHub profiles.

A good starting point is to have a look at the official JupyterHub Docker Stacks Images @ Docker-Stacks.

In a nutshell its necessary to install the `batchspawner` package into your docker image (https://pypi.org/project/batchspawner/) - this package provides functionality needed to run in the slurm environment of VSC (e.g. `batchspawner-singleuser` script).

Optionally: If your image needs to run hooks before starting the script the original docker stacks provide a folder for such startup scripts `/usr/local/bin/before-notebook.d` (e.g. the pyspark image depends on this PySpark Dockerfile.

Unfortunately the batchspawner package has not been updated yet so in order to restore this functionality we are using an (optional) startscript called `vsc-singleuser.sh`. All it does is to run the hooks from `/usr/local/bin/before-notebook.d` before finally executing the `batchspawner-singleuser` script.

The script can be found at vsc-singleuser.sh. Note: Contact us to get read rights to the repository.

In addition we also install the following jupyterlab extensions in our images: * jupyterlab-system-monitor * jupyterlab-git * jupyterlab-widgets

An up 2 date list of packages can always be found in our repo: requirements.txt

In order to then use the image with JupyterHub you have to convert it to a singularity image and place it into a folder that is accessible from the VSC compute nodes.

Building a singularity image from e.g. a docker container can easily be done with the `singularity build` command. See the singularity documentation for examples: Build a (singularity) container

FAQ

  • The OTP SMS says 'login at vsc3'? Is this an issue?

No this is a current limitation of our OTP system and should not cause any problems.

  • My Server instance is stuck and I get a timeout when I try to reload the window. Going to the VSC jupyterhub website also results in a timeout.

In order to to control the running instance, navigate directly to the following URL: https://jupyterhub.vsc.ac.at/hub/home. From there you can click 'Stop My Server' to stop the running instance (if there is any).

doku/jupyterhub.txt · Last modified: 2022/05/23 12:32 by katrin