HPC OnDemand

Overview

Open OnDemand (OOD or OoD) is a service providing access to HPC resources through a web browser. Manage your files, check the cluster job status, submit and monitor your own jobs, obtain a shell session, or choose from a set of "interactive" applications -- each provided through a browser tab.

Access

Access the OnDemand server using your Case Single-SignOn (SSO) credentials. It doesn't require campus network or VPN connection.

Use ondemand-pioneer.case.edu to access Pioneer cluster.

Alternatively, use  ondemand.case.edu for access to Markov cluster

OnDemand will need to write files to disk, in the directory(ies), to record information about the conditions requested for the app session in the location  /home/<caseid>/ondemand/data/sys/dashboard/batch_connect/.

Write capability may be denied if the group storage usage reaches the group hard quota.  To still launch an interactive app will require either resolving the disk usage situation, or the following.  Logout of OnDemand, use AMARA to change the primary linux group.  Initiate a new Ondemand session (private window or delete any cookies) and proceed to request the interactive session resources.

Important Notes

• The Shell Access  (via 'cluster' pulldown) doesn't support Visual Access. Use Desktop under Interactive Apps for visual access. The details on visual access is available at HPC Guide to Graphical Access.

• The field under Desktop should be appropriately filled else you will get an error. For an example, If the "Node Type (partition)" field is set to GPU, "Number of GPU" and "CPU or GPU Node Type" should be filled appropriately.

• We have observed that commands with progress output (like a progress bar or a percent progress) can sometimes run slower in the OnDemand terminal than when using a regular terminal. (see User Quick Start Guide for details)

The OOD Dashboard is the landing page:

Browser support

Most major browsers support all features of OnDemand; however, the Safari browser does not implement all features, due to the authentication method used in the CWRU implementation of OOD. The Markov Shell using google chrome may produce truncated output (e.g. may not display compiler error properly) or command (e.g. systemctl) won't work as desired.

Features

A video overview of features is available.

The dropdown menus provide:

• • Files:  Home Directory, with view, edit, upload and other capabilities.

  • Jobs:  Choose Active Jobs to review status, or Job Composer to prepare job submissions

  • Clusters:  A bash shell on the selected cluster

  • Interactive Apps:  These sessions run on compute nodes, scheduled through Slurm manager. Applications include Jupyter/Tensorflow, RStudio, a desktop session and/or CodeServer. Check the cluster for up-to-date application availability.

Passenger Apps

The apps organized under Files, Jobs and Clusters access headnode resources and personal account data. These apps aid in managing, and conducting your work.

Interactive Apps

To see the current set of interactive apps, use the 'Interactive Apps' pulldown menu. These apps require computational resources for their graphics environment, and to run the underlying app, therefore they run on compute node resources obtained through the Slurm scheduler.  A form is presented after clicking on the item in the pulldown menu, through which compute resources are specified. Some applications, particularly on the Pioneer cluster, are modified to allow customization of the environment.  Examples are presented below.

Kernels for Jupyter App 

RStudio Server

• RStudio Server: within a batch job on a compute node.

  • Slurm Account

  • Node Type (slurm partition)

  • Duration -- up to 60 hours

  • Cores -- up to 16

  • GPU -- applies only for gpu node type

  • Memory -- from 4gb to 64gb

The is an alternative approach when starting the rstudio session from the ood 'connect' icon fails.  

1. Open a new tab.

2. Create a url with the format https://ondemand-pioneer.case.edu/rnode/<compute-node>/<port> 

   • Obtain the <port> from the file connection.yml, accessible through the 'session id' link of the rstudio session information 

   • The <compute-node> is visible in the 'launch' window.

3. Once the url is complete in the appropriate format, navigate to the url

4. Follow the above image to select 'Global options' from the 'Tools' menu.

Desktop Sessions

• • Desktop Sessions: general purpose HPC graphical interface

    • Account

    • Desktop application

      • xfce4

      • mate

    • Duration -- up to 60 hours

    • Cores -- up to 16

    • Memory -- from 4gb to 64gb

RStudio Session Startup problem

Startup problems for RStudio, such as https errors 502/502, or just a lengthy delay in the app loading, may be due to settings that load large data files at start up.  Changing the settings may be done through the graphical interface after it loads.  Starting in Safe Mode may be necessary. If prompted to start Safe Mode, select this option and update the settings.  Look for the 'tools' pull down menu to access the 'Global options', where you may deselect the various options for restoring data or settings from previous sessions by default. 

There is an alternative approach when starting the rstudio session from the ood 'connect' icon fails.  

1. Open a new tab.

2. Create a url with the format https://ondemand-pioneer.case.edu/rnode/<compute-node>/<port> 

   • Obtain the <port> from the file connection.yml, accessible through the 'session id' link of the rstudio session information 

   • The <compute-node> is visible in the 'launch' window.

3. Once the url is complete in the appropriate format, navigate to the url

4. Follow the above image to select 'Global options' from the 'Tools' menu.

Issue with studio app

Issue with plots in R studio app

The newer R installs are using a different default setting when calling png()

The way to restore the default behavior is to add the following somewhere in your workflow:

options(bitmapType="cairo")

This could be:

1) Called in the RStudio interpreter prior to running your main script

2) Added to the top of the R script itself

3) Added to /home/<CaseID>/.Rprofile prior to launching the RStudio job in ondemand

Issue Installing Packages in R studio app

We advise installing packages directly in R (https://sites.google.com/a/case.edu/hpcc/hpc-cluster/pioneer-software/programming-languages/r-project).  The RStudio app presents a container environment which lacks the full environment necessary to install software in many cases.

Blank Screen at startup

A session may begin where at launch, the connection is made to the compute node 'remote host', and yet the RStudio session is not displayed in the window.  The url will show the details, and for a few seconds the 'spinning wheel' may be observed, but after that a blank window persists for many mnutes, and if patient, even multiple hours.

In this case, it is worth checking in the directory for any settings relating to previous 'active' sessions. The information would be in the following location by defauilt:
/home/<caseid>/.local/share/rstudio/sessions/active/

Look for subdirectories that would contain session information, and specifically, a file 'session-persistent-state',
/home/<caseid>/.local/share/rstudio/sessions/active/session-efe57ea5/session-persistent-state

 that file may contain information similar to:
abend="1"
active-client-id="99ee0b2a-6998-48ea-a099-0d7eab5b0327"

If so, delete the file and/or directory contents, and relaunch the connection to the 'blank' session.

If you don't request enough memory, your job may timeout or run out of memory. Using OnDemand Desktop, you may get the blank page. Check your job status using sacct command:

sacct -u aps223 -S 080124

JobID           JobName  Partition    Account  AllocCPUS      State ExitCode

------------ ---------- ---------- ---------- ---------- ---------- --------

2042066      sys/dashb+      batch  axa1826_1          4    TIMEOUT      0:0

2042066.bat+      batch             axa1826_1          4 OUT_OF_ME+    0:125

2042066.ext+     extern             axa1826_1          4  COMPLETED      0:0