Submitting SBATCH job scripts

Owned by Josh Burks
Last updated: Dec 28, 2023 by William Dizon
1 min read

SBATCH files are--at the core--simply regular, executable scripts with scheduler resource information.

BASH SCRIPTS

Consider this example script test.sh

#!/bin/bash #SBATCH -N 1 # number of nodes #SBATCH -c 4 # number of cores to allocate #SBATCH -t 0-00:02:00 # time in d-hh:mm:ss echo "hello"

With the executable permission (+x), this script can run in an ordinary terminal session.

By running the command ./test.sh, it is processed by the bash shell indicated in line 1. This means it will use the bash interpreter, echo hello, and exit.

Note, the number of cores, nodes, and length of time have no meaning here--by being interpreted by bash, each line that begins with # is considered a comment, and ignored.

SBATCH SCRIPTS

Consider the same above script test.sh now run through the supercomputer scheduler:

$ sbatch test.sh

By passing test.sh to sbatch, the file no longer needs executable permissions; sbatch will read each of these lines and reproduce the results, but now within a contained process--that is, one that is confined by the amount of resources you specify.

Thus, running this command will:

  • Request 4 CPU cores

  • Ensure all cores share the same physical machine

  • Ensure all cores are available for 2:00 minutes.

When the job completes (within a second), the job will end and the remaining 1:59 minutes will be surrendered and left unused.

When using sbatch, you have the same capability to use different directives as with salloc, so your file can include additional, highly-reproducible constraints:

#SBATCH -o slurm.%j.out # file to save job's STDOUT (%j = JobId) #SBATCH -e slurm.%j.err # file to save job's STDERR (%j = JobId) #SBATCH --export=NONE # Purge the job-submitting shell environment