⭐️sbatch

Overview

sbatch is the command in Slurm for submitting batch jobs:

• Returns to the command line immediately after submission (runs in the background)
• Runs automatically on compute nodes when resources are available (not on login nodes)
• Use cases: any task that does not require interaction

Common Options

Option	Description	Example
-p, --partition=partition	Specify partition	-p gpu
-J, --job-name=jobname	Set job name	-J my_job
-o, --output=out	Output log path	-o /path/output.log
-e, --error=err	Error log path	-e /path/error.log
-N, --nodes=N	Number of nodes	-N 2 (2 nodes)
-n, --ntasks=ntasks	Total CPU cores	-n 8 (8 cores)
-c, --cpus-per-task=ncpus	Cores per task	-c 4 (4 cores per task)
-t, --time=minutes	Time limit	-t 1:30:00 (1h 30m)
-w, --nodelist=hosts...	Specify nodes	-w node[1-3]
-x, --exclude=hosts...	Exclude nodes	-x node5
-W, --wait	Wait for job completion	sbatch -W ...
--wckey=wckey	Specify wckey	--wckey="project1"
--wrap[=command string]	Wrap command string in an sh script and submit	--wrap "hostname"
-h, --help	Show full help	sbatch -h

Examples

Run a single command

sbatch -p compute -o make.log --wrap "make test"

Effect: Run make test in the compute partition and save logs to make.log

⚠️ Note: Add -W to wait for the job to finish (useful for chaining tasks in scripts)

Submit a job via script

Script example (myjob.sh):

#!/bin/bash
echo "Task started at $(date)"
hostname
sleep 10
echo "Task finished at $(date)"

The user's sbatch command is:

sbatch -p compute -o job.log myjob.sh

This submission command submits the mybash.sh script to the compute partition and prints the standard output and error of the myjob.sh script to job.log.

Put common options in the script header

📌 Tip: It's recommended to put common options in the script header (example below for mybash.sh):

#!/bin/bash
#SBATCH -J my_job
#SBATCH -p compute
#SBATCH -N 2
#SBATCH -o %j.out
#SBATCH -e %j.err
srun your_program

%j is a filename pattern substitution; see the section below

The user's sbatch command is:

sbatch mybash.sh

Filename pattern substitution

The following substitution tokens can be used in output filenames (must be used inside double quotes):

Common tokens

Token	Description	Example
%j	Job ID	job-%j.out → job-12345.out
%x	Job name	%x-%j.out → myjob-12345.out
%u	Username	%u-%j.out → user1-12345.out
%a	Job array index	output-%a.out → output-1.out
%N	Short hostname	Generate a separate file per node

Zero-padding examples

# Generate job0123.out (4-digit padding)
sbatch -o "job%4j.out" script.sh

# Generate job123-01.out (2-digit task ID padding)
sbatch -o "job%j-%2t.out" script.sh

💡 Tip: For more complex pattern substitution, see the Slurm official documentation or man sbatch

Common environment variables

Input environment variables

Some options can be set via environment variables. Command-line options take precedence and will override environment variable settings. The environment variables and their corresponding options are:

Variable	Option	Example
SBATCH_PARTITION	-p	export SBATCH_PARTITION=gpu
SBATCH_JOB_NAME	-J	export SBATCH_JOB_NAME=myjob
SBATCH_TIME	-t	export SBATCH_TIME=1:00:00
SBATCH_OUTPUT	-o	export SBATCH_OUTPUT=job-%j.out

Output variables

Slurm outputs the following variables in job scripts, and job scripts can use these variables:

Variable	Description
SLURM_JOB_ID	Current job ID
SLURM_JOB_NAME	Job name
SLURM_JOB_NODELIST	Allocated node list
SLURM_SUBMIT_DIR	Job submission directory
SLURM_CPUS_PER_TASK	CPUs per task
SLURM_GPUS	Allocated GPU count