Quick reference sheet for SLURM resource manager
Job scheduling commands
| Commands | Function | Basic Usage | Example |
|---|
| sbatch | submit a slurm job | sbatch [script] | $ sbatch job.sub |
| scancel | delete slurm batch job | scancel [job_id] | $ scancel 123456 |
| scontrol hold | hold slurm batch jobs | scontrol hold [job_id] | $ scontrol hold 123456 |
| scontrol release | release hold on slurm batch jobs | scontrol release [job_id] | $ scontrol release 123456 |
Job management commands
| Job Status | Commands |
|---|
| sinfo -a | list all queues |
| squeue | list all jobs |
| squeue -u userid | list jobs for userid |
| squeue -t R | list running jobs |
| smap | show jobs, partitions and nodes in a graphical network topology |
Job script basics
A typical job script will look like this:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
| #!/bin/bash
#SBATCH --nodes=1
#SBATCH --cpus-per-task=8
#SBATCH --time=02:00:00
#SBATCH --mem=128G
#SBATCH --mail-user=netid@gmail.com
#SBATCH --mail-type=begin
#SBATCH --mail-type=end
#SBATCH --error=JobName.%J.err
#SBATCH --output=JobName.%J.out
cd $SLURM_SUBMIT_DIR
module load modulename
your_commands_goes_here
|
Lines starting with #SBATCH are for SLURM resource manager to request resources for HPC. Some important options are as follows:
| Option | Examples | Description |
|---|
| --nodes | #SBATCH --nodes=1 | Number of nodes |
| --cpus-per-task | #SBATCH --cpus-per-task=16 | Number of CPUs per node |
| --time | #SBATCH --time=HH:MM:SS | Total time requested for your job |
| --output | #SBATCH -output filename | STDOUT to a file |
| --error | #SBATCH --error filename | STDERR to a file |
| --mail-user | #SBATCH --mail-user user@domain.edu | Email address to send notifications |
Interactive session
To start a interactive session execute the following:
1
2
3
| #this command will give 1 Node for a time of 4 hours
srun -N 1 -t 4:00:00 --pty /bin/bash
|
You can use slurm database to see how much memory your previous jobs used, e.g. the following command will report requested memory and used residential and virtual memory for job
1
2
| sacct -j <JOBID> --format JobID,Partition,Submit,Start,End,NodeList%40,ReqMem,MaxRSS,MaxRSSNode,MaxRSSTask,MaxVMSize,ExitCode
|
Place these alias’ into your .bashrc
1
2
| alias si="sinfo -o \"%20P %5D %14F %8z %10m %10d %11l %16f %N\""
alias sq="squeue -o \"%8i %12j %4t %10u %20q %20a %10g %20P %10Q %5D %11l %11L %R\""
|
More information about Slurm can be found here:
Table of contents