How to Use¶
How to access¶
Access to our cluster can be done through Open OnDemand or via the JupyterLab (K8S) Terminal. In both options, it is essential to have a valid account in LIneA's computational environment. If you don't have an account, contact the Service Desk by email (helpdesk@linea.org.br) for more information.
Attention
Even with an active LIneA account, access to the HPC processing environment is not automatic. For more information contact the Service Desk at helpdesk@linea.org.br.
Accessing via JupyterLab terminal
In the home screen of your Jupyter Notebook, in the "Other" section, you will find the terminal button. When clicking it, you will be redirected to a Linux terminal, initially located in your home directory. To access the Apollo Cluster, simply execute the following command:
ssh loginapl01
$HOME and $SCRATCH
Compute nodes don't have access to your user (home) directory. Move or copy all files needed for job submission to your SCRATCH directory.
How to use the SCRATCH area¶
Your SCRATCH directory is the place to send essential files for job submission, as well as to check results after code execution. It's crucial to note that all results and generated files must be transferred back to your user directory (home). Otherwise, there's a risk of losing these files stored in your SCRATCH.
- To access your SCRATCH directory:
ssh $SCRATCH
- To send files to your SCRATCH directory:
cp <FILE> $SCRATCH
EUPS Package Manager¶
EUPS is an alternative package manager (and official LSST one) that allows loading environment variables and including paths to programs and libraries in a modular way.
- To load EUPS:
. /mnt/eups/linea_eups_setup.sh
- To list all available packages:
eups list
- To list a specific package:
eups list | grep <PACKAGE>
- To load a package in current session:
setup <PACKAGE NAME> <PACKAGE VERSION>
- To remove loaded package:
unsetup <PACKAGE NAME> <PACKAGE VERSION>
How to Submit a Job¶
A Job requests computing resources and specifies applications to be launched on those resources, along with any input data/options and output directives. Cluster task and resource management is done through Slurm. Therefore, to submit a Job you need to use a script like below:
#!/bin/bash
#SBATCH -p PARTITION #Name of the Partition to use
#SBATCH --nodelist=NODE #Name of the Node to be allocated
#SBATCH -J simple-job #Job name
#----------------------------------------------------------------------------#
##path to executable code
EXEC=/lustre/t0/scratch/users/YOUR.USER/EXECUTABLE.CODE
srun $EXEC
In this script you need to specify: the queue name (Partition) to be used; the node name to be allocated for Job execution; and the path to the code/program to be executed.
WARNING
It is strictly prohibited to submit jobs directly to the loginapl01 machine. Any code running on this machine will be immediately interrupted without prior notice.
- To submit the Job:
sbatch script-submit-job.sh
If the script is correct there will be an output indicating the job ID.
- To check job progress and information:
scontrol show job <ID>
- To cancel the Job:
scancel <ID>
Internet access
Compute nodes do not have internet access. Packages and libraries must be installed from loginapl01 in your scratch area.
Useful Slurm Commands¶
To learn about all available options for each command, enter man <command> while connected to the Cluster environment.
| Command | Definition |
|---|---|
| sbatch | Submits job scripts to execution queue |
| squeue | Displays job status |
| scontrol | Used to display Slurm state (various options available only to root) |
| sinfo | Displays partition and node status |
| salloc | Submits a job for execution or starts a real-time job |