Open OnDemand¶
Some HPCCF clusters have Open OnDemand (OOD). OOD allows access to cluster resources using a web browser. All OOD apps are automatically launched through Slurm jobs, so you have access to your normal cluster resources. Just like sbatch jobs, OOD apps (jobs) run even when your browser is not attached, so you can reattach to a running OOD app just by going back to the OOD website.
Clusters with Open OnDemand:¶
Applications offered with Open OnDemand¶
- JupyterLab
- [Cluster Name] Desktop
- VSCode Server
- RStudio Server
Desktop notes¶
HPCCF provides a full Ubuntu Desktop using the XFCE4 desktop environment. Chrome and Firefox are included, and can be used to download data. This desktop is not intended for long term use. To prevent issues, your Firefox and Chrome data saved on the cluster is deleted every time you launch a new desktop session. This prevents left-behind lock files from blocking browser launch when the Slurm job terminates before you exit the desktop session.
User OnDemand debugging¶
If your OOD jobs launches, but fails very quickly, please click the long, random Session ID:. From there, open the
output.log file and carefully look over the output.
Incorrect Conda environment specified¶
Some OOD Apps allow you to specify a specific conda environment that will load. If you see a line like this in
output.log, you have specified an invalid conda environment:
EnvironmentNameNotFound: Could not find conda environment: r-4.3.3-typo
You can list all discoverable environments with `conda info --envs`.
Out of Memory (OOM) events¶
If your Jupyter or RStudio jobs keep failing with errors like abnormally terminated due to an unexpected crash, and,
when your job finishes, you see a line like this in output.log, then you have not requested enough RAM:
slurmstepd: error: Detected 3 oom_kill events in StepId=20294700.batch. Some of the step tasks have been OOM Killed.
Other indications of an OOM event are an RStudio or Jupyter window like these:
Screenshots
Self installed conda or miniconda break RStudio Server¶
If you self-install conda, or miniconda, it will conflict with our centrally installed conda. If you see lines like this, then you need to remove that to use RStudio Server:
ERROR: CONDA_EXE is currently defined: /home/omen/conda/bin/conda.
This module will almost certainly interfere with your conda installation.
Remove existing conda installation and its shell hooks from your PATH before proceeding.
See Migrating from User-installed Conda
Time limit¶
If you see output like this, it means you did not request enough hours during the app configuration in OOD:
slurmstepd: error: *** JOB 20296118 ON cpu-6-96 CANCELLED AT 2025-01-21T17:34:19 DUE TO TIME LIMIT ***
Failed to submit session with the following error:¶
If your job is rejected at submission time for reasons like below (basically anything that starts with
sbatch: error:), please work with your PI to figure out an appropriate partition and resource request.
Other error¶
If your issue is not listed here, please email hpc-helpat ucdavis.edu and send the cluster name, the OnDemand app
you tried to run, the Session ID of the failed job, any errors you noticed, and any troubleshooting you performed.