Job Composer¶

The Job Composer provides a graphical interface for creating, configuring, and submitting batch jobs to Derecho or Casper without requiring SSH access or command-line interaction. It is accessible from the Open OnDemand (OOD) portal at ondemand.hpc.ucar.edu by selecting Jobs > Job Composer from the top navigation menu.

Jobs created through the composer are stored in a dedicated workspace directory on the file system. Each job directory contains the submission script and any associated input files, and can be reused or copied for future runs.

Submitting a Job¶

From a Default Template¶

The default template provides a minimal PBS job script that can be edited directly in the browser. This is the recommended starting point for new users.

1. Select New Job > From Default Template from the Job Composer toolbar.
2. A new job entry will appear in the job list with the status Not Submitted. The right-hand panel shows the job details and a file listing for the job directory.
3. Click Job Options to set the job name, submission script filename, and target cluster (Derecho or Casper).
4. Click Open Editor to open the submission script in the browser-based editor. Modify the PBS directives and job commands as needed (see example scripts below), then save the file.
5. Click Submit to queue the job.

The job status will update from Not Submitted → Queued → Running → Completed (or Failed). Output files are accessible through the Folder button in the job listing once the job finishes.

From a Specified Path¶

Use this method when you have an existing directory on the file system containing a job script and input files you want to submit directly.

1. Select New Job > From Specified Path.
2. In the dialog, provide:
   • Path: the full path to the directory containing your job script (e.g., /glade/work/username/myjob/)
   • Job Name: a display label for the job in the composer
   • Script: the filename of the PBS submission script within that directory
   • Cluster: Derecho or Casper
3. Click Create New Job. The composer will reference the files at that path rather than copying them.
4. Review the script via Open Editor and click Submit when ready.

From a Saved Template¶

If you have previously created a job template, you can create new jobs from it to avoid re-entering configuration each time.

1. Select New Job > From Template.
2. Select the desired template from the list.
3. The composer creates a new job directory by copying the template's files into a new workspace.
4. Click Job Options to adjust any per-run settings, then edit the script and submit as usual.

From an Existing Job¶

To duplicate a job you have already run (for example, to rerun with minor changes):

1. Select the job in the job list.
2. Click New Job > From Selected Job.
3. A new job directory is created as a copy of the selected job, including all files and script settings.

Example Scripts¶

Derecho¶

#!/bin/bash
#PBS -N my_job
#PBS -A <project_code>
#PBS -q main
#PBS -l walltime=01:00:00
#PBS -l select=1:ncpus=128:mpiprocs=128
#PBS -j oe

### Load required modules
module reset
module load ncarenv intel

### Run the executable
mpiexec ./my_program

Casper¶

#!/bin/bash
#PBS -N my_job
#PBS -A <project_code>
#PBS -q casper
#PBS -l walltime=02:00:00
#PBS -l select=1:ncpus=4:mem=50GB
#PBS -j oe

### Load required modules
module reset
module load ncarenv

### Run the executable
./my_program

Replace <project_code> with your active allocation. See Job Script Examples for a broader set of templates covering MPI, GPU, and array jobs.

Creating and Managing Job Templates¶

Templates allow you to define a reusable job configuration including the submission script and any supporting files that can be shared across multiple job runs.

Creating a Template¶

1. Navigate to the Templates tab at the top of the Job Composer panel.
2. Click New Template.
3. Provide:
   • Template Name: a descriptive label for the template
   • Path: the full path to a directory containing your template files (submission script, any required input stubs, etc.)
   • Script: the filename of the PBS script within that directory
   • Cluster: the default target cluster
   • Notes (optional): a description of what the template is for
4. Click Save. The template will appear in the Templates list and can be selected when creating new jobs.

Editing a Template¶

1. Select the template in the Templates list.
2. Click Edit to update the template metadata (name, notes, default cluster, or script filename).
3. To modify the script or supporting files, navigate to the template directory via the File Explorer or a terminal session and edit the files directly. Changes take effect for all jobs created from the template going forward.

Deleting a Template¶

Select the template and click Delete. This removes the template entry from the Job Composer but does not delete the source directory or any jobs previously created from it.

Managing Jobs¶

Monitoring Status¶

The Job Composer job list displays each job's current status. For a consolidated view of all running and queued jobs across Derecho and Casper, use Jobs > Active Jobs from the top navigation menu.

Stopping a Job¶

Select a running job in the job list and click Stop to cancel it. This is equivalent to running qdel <job_id> from the command line.

Viewing Output¶

Once a job completes, click the Folder button on the job entry to open the job's working directory in the File Explorer. Output files (e.g., <job_name>.o<job_id>) are listed there and can be viewed or downloaded directly in the browser.

Deleting a Job¶

Select a job and click Delete to remove it from the Job Composer list. This also deletes the associated job workspace directory and all files within it.

Getting Help¶

For assistance with the Job Composer or batch job submission at NSF NCAR, contact the Consulting Services Group User Support Team or submit a ticket to the NSF NCAR Research Computing Help Desk.