Submitting a batch job
Submit a job to the queue with command sbatch:
For example:
When the job is successfully submitted, the command prints out the ID number of
the submitted job. For example, Submitted batch job 3609650.
To check if your job is running correctly:
or
You should see your job ID and other details displayed in the terminal.
To cancel a submitted batch job:
For example,
To find out more about your jobs:
The information includes the state of the jobs (PENDING, RUNNING,
COMPLETED, FAILED, etc.) and the job ID. By default, the sacct command
displays information about a user's own jobs that have been submitted to the
queue on the current date. It also has a wide selection of options and
parameters that can be used to select which data is displayed.
See all sacct options and parameters.
Avoid querying too much data with sacct
Do not query job data from a long time period with sacct. sacct fetches
its data from the Slurm accounting database, and with large queries the
operation can be heavy on the system. In particular, avoid running the
command many times in a row. Instead, you can redirect the output data to a
file and then search that for whatever information you want to analyze.
For example, to save the data from the past 7 days in a file
sacct-output.txt:
Do not mix CPU and GPU architectures on Roihu
Although the scheduler allows submitting jobs from CPU login nodes (x86_64) to GPU queue (aarch64) and vice versa, this is strongly discouraged: binaries are not portable across architectures and jobs may fail unpredictably. Always submit from a login node matching your target compute nodes.