Open OnDemand

From RCSWiki
Revision as of 19:00, 16 November 2021 by Dmitri (talk | contribs) (→‎Connect to OnDemand)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Open OnDemand is a web portal for accessing certain clusters operated by Research Computing Services. The web portal provides a convenient way to access the login node, your files, and certain graphical applications such as Jupyter Notebooks and remote desktops. This service is an additional way to access HPC resources in addition to the existing command-line based options.

Open OnDemand is an open source project actively developed by the Ohio Supercomputer Center.

Access

Get an account

Before using Open OnDemand, you will need to have an account on the cluster. If you do not already have an account, please review the cluster's quick start guide for information on getting started.

Connect to OnDemand

You may access Open OnDemand for the following clusters using your UCIT credentials. Sign-on is handled through the University's Single Sign-On mechanism and requires Multi-Factor Authentication enabled.

Cluster Open OnDemand Access
ARC Cluster ood-arc.rcs.ucalgary.ca

Open OnDemand Dashboard

After logging in, you will see the Open OnDemand dashboard. The Message Of The Day (MOTD) will show any news and announcements relating to the cluster. Any quota warnings will also be displayed on your dashboard.

Open OnDemand Dashboard

Other components and applications can be accessed through the top navigation.

File Browser

The file browser interface allows you to manage, upload, or download files from your directories, drag & drop file management, and basic file viewing and editing. You can access all files across all filesystems available to the cluster with this interface.

There is a 128 MB limit on file uploads. Please do not use this interface for large file transfers and instead look at other methods for file transfers.

Job Explorer

You may view and manage your current jobs on the cluster through the Active Jobs page. This may be helpful for users new to visualize scheduled jobs in Slurm.

Shell Access

You can launch a SSH session via Shell Access to the login node of the cluster.

Interactive Applications

We have created a small selection of graphical applications launchable through Open OnDemand. We currently offer:

  • Jupyter Notebook / Jupyter Lab
  • RStudio (via VNC as a desktop application)
  • RStudio Server
  • VNC Remote Desktop (software rendering)
  • VNC Remote Desktop (VirtualGL rendering)

Jupyter Notebooks

Open OnDemand offers Jupyter Notebooks via containers or environment modules. For most users, we recommend using the containers way to launch your Jupyter notebooks as it is the simplest. However, we also offer the environment module based method for users that already have a working workflow based on environment modules.

Container-based Jupyter notebooks

Open OnDemand Jupyter Notebook
Open OnDemand Jupyter Notebook

We offer a selection of different Jupyter Notebooks derived from the official Jupyter Stacks Docker images. We will provide updated images with major ARC upgrades. The following flavors of Jupyter notebooks are currently available:

Image Description
Spark Notebook all-spark-notebook_spark-#.#.#.sif Includes Python, R, and Scala support for Apache Spark.

Spark version is listed in the filename.

Julia Data Science Notebook datascience-notebook_julia-#.#.#.sif Includes libraries for data analysis from the Julia, Python, and R communities.

Julia version is listed in the filename.

Python Scipy Notebook scipy-notebook_python-#.#.#.sif Includes Python support for Apache Spark.

Python version is listed in the filename

Tensorflow Notebook tensorflow-notebook_tensorflow-#.#.#.sif Based on the Scipy notebook but includes Tensorflow.

Tensorflow version is listed in the filename.

gpu-jupyter.sif Includes NVIDIA CUDA along with the data science notebook.

Based on https://github.com/iot-salzburg/gpu-jupyter/

To select the flavour of your notebook, select the desired image in the drop-down menu in the launch page. You may optionally use a custom image as well by selecting 'Custom Singularity Image...' (see the next section below). You may also specify any additional arguments to Jupyter including whether to launch JupyterLab instead.

Create a custom Jupyter Notebook image

To help facilitate the many different workflows you may have, we allow the use of customized Jupyter notebook images. Simply place your custom Jupyter Singularity image at $HOME/ondemand/jupyter and select 'Custom Singularity Image...' on the launch page.

All custom Jupyter images should be based off of the base Jupyter notebook image for compatibility. Additional packages can then be installed via pip or conda. An example Dockerfile is given below.

FROM jupyter/base-notebook

# Install any additional packages here
RUN set -ex; \
        conda install --yes \
                -c conda-forge \
                ipywidgets ;\
        conda clean --all -f -y; \
    fix-permissions "${CONDA_DIR}"; \
    fix-permissions "/home/${NB_USER}"

USER $NB_UID

WORKDIR $HOME

Build the docker image and convert it into a Singularity image.

Environment module based Jupyter notebooks

This is the alternative way of launching Jupyter notebooks on Open OnDemand. This option automates our previous recommended way of running Jupyter on ARC (described on our Jupyter Notebooks page) by loading python from an Anaconda environment and creating a port forwarding. We recommend this option only if you have an existing workflow that uses Jupyter with modules.

When using this method, you will be asked for an initialization bash script that sets up and starts Jupyter. An example script is given below. You may modify and add to this example script to customize your Jupyter Notebook setup. If no script is given when launching the notebook or if the script is not found, this example script will be used instead.

#!/bin/bash

# TODO: Set up your environment as required.

# Purge the module environment to avoid conflicts
module purge
module load python/anaconda3-2018.12
# TODO: Add any additional modules desired here.
module list

# As per https://rcs.ucalgary.ca/index.php/Jupyter_Notebooks
unset XDG_RUNTIME_DIR

# Launch the Jupyter Notebook Server
set -x
jupyter notebook --config="${CONFIG_FILE}"

Remote Desktop

Remote desktops may be helpful when exploring data or to run certain GUI based applications. Previously, we required users to manually set up and use X11 forwarding or VNC via SSH tunnels. The Open OnDemand Desktop app simplifies the process of launching a full Linux desktop session or other graphical based applications all within your web browser. We currently offer only XFCE as the desktop manager. The desktop can run with either software rendering or hardware (via VirtualGL) rendering. VirtualGL rendering may only be used only on specific nodes and is currently not publicly available.

Remember, your desktop will run as an interactive job on the Slurm cluster and will be subject to a maximum time limit before it is terminated. As a result, you will need to restart your desktop after the time limit is reached. When using this desktop feature, be sure to save your work frequently and be mindful of the remaining time left for your session to avoid loss of work.

To get started, click on 'Desktop (Software)' from the Interactive Apps menu. Select the desired partition that your desktop will run on.

Open OnDemand Desktop
Open OnDemand Desktop

Use the launch page to specify the resources your session should need. You may optionally request for a dedicated GPU for VirtualGL based sessions.

Once you launch the remote desktop job, you will be presented with a notification once it starts. Depending on how busy the cluster is and the requested partition and resources, you may need to wait. Once started, you will see the option to launch the Remote Desktop in green:

Open OnDemand Desktop Launch
Open OnDemand Desktop Launch

Click on the blue "Launch Desktop" button to connect to your desktop.

Remote Desktop on Open OnDemand
Remote Desktop on Open OnDemand

You may connect to CIFS/Samba shares through the desktop. Do so by opening the Thunar file manager and entering a SMB address:

Thunar SMB connection
Thunar SMB connection


RStudio

You may launch RStudio (a desktop application) through Open OnDemand. After launching a RStudio session, connect to it as

Launch RStudio on Open OnDemand
Launch RStudio on Open OnDemand

The desktop application is similar to a Remote Desktop session but will automatically load and start RStudio.

RStudio via VNC
RStudio via VNC

RStudio Server

RStudio Server is a web-based version of RStudio and functions similar to Jupyter Notebooks. After requesting for a RStudio Server session through Open OnDemand, you will see a session information box similar to the one below. Each session will use a randomized password.

RStudio Server launch on Open OnDemand
RStudio Server launch on Open OnDemand

After clicking on 'Connect to RStudio Server', you will need to enter the username and password that are displayed:

RStudio Server Login
RStudio Server Login

Troubleshooting

VNC session cannot reconnect

If your VNC session closed, you may need to re-connect by going back to the Open OnDemand dashboard, listing your sessions under "My Interactive Sessions" page, then click on the blue "Launch Desktop" button.