SLURM Script Syntax

Last Updated 2018-11-16

---

References

• Full online SLURM documentation
• UVA ARCS’s introduction to SLURM

Template

#!/bin/bash
#SBATCH --nodes=2               # number of nodes (computers) to use
#SBATCH -N 2
#SBATCH --ntasks-per-node=20    # number of processes to run per node
#SBATCH --ntasks=1              # number of processes to run
#SBATCH -n 1
#SBATCH --mem=4000              # memory to use in MB
#SBATCH --mem-per-cpu=100       # memory to use per CPU in MB 
                                #   (cannot use with --mem)
#SBATCH --cpus-per-task=1       # number of CPUs (cores) per process
#SBATCH -c 1
#SBATCH --time=1-12:10:00       # amount of time for the whole job d-hh:mm:ss
#SBATCH -t 1-12:10:00
#SBATCH --partition=standard    # the queue/partition to run on 
                                #   (standard/parallel/largemem)
#SBATCH -p standard
#SBATCH --output=myResults.out  # file for saving standard output
                                # Note: by default, both standard output 
                                #   and standard error goes to slurm-<jobid>.out
#SBATCH -o myResults.out
#SBATCH -o myjobs.%A_%a.out
#SBATCH --error=myResults.err   # file for saving standard error
                                # Note: by default, both standard output 
                                #   and standard error goes to slurm-<jobid>.out
#SBATCH -e myResults.err
#SBATCH -e myjobs.%A_%a.err
#SBATCH --account=blabber       # the account/allocation to use
#SBATCH -A blabber
#SBATCH --mail-user=<email>     # address to e-mail
#SBATCH --mail-type=<type>      # send e-mail on certain events; type can be:
                                #   BEGIN, END, FAIL, REQUEUE or ALL
#SBATCH --job-name=<name>       # name of the job
                                #   Note: job arrays should be named
#SBATCH -J <name>

# Actual job script to run
./myexec < input > output

System variables

Note: Do not assign to (only read from) any of these variables inside the SLURM script

SLURM_SUBMIT_DIR                # directory from which job was submitted
SLURM_JOB_NODELIST              # list of nodes
SLURM_JOB_ID                    # job ID
SLURM_NTASKS                    # number of tasks
SLURM_NTASKS_PER_NODE           # number of tasks per node
SLURM_CPUS_PER_TASK             # number of cores per task
SLURM_ARRAY_JOB_ID              # represented by %A in the SLURM script
SLURM_ARRAY_TASK_ID             # represented by %a in the SLURM script