Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 17 Next »

Slurm Job Examples

Included here are a few key examples to demonstrate the workflow of batch jobs. If you find that your use case is not represented below, please check out our github page for a more complete set of examples. If there is something you would like us to include here, please let us know!

 Serial Job

Serial Job

This script runs a serial, single-CPU job using the standard queue.

Any line beginning with #SBATCH issues a SLURM directive that controls aspects of your job such as job name, output filename, memory requirements, number of CPUs, number of nodes, etc.

To run the script, replace YOUR_GROUP with the name of your PI’s group on HPC. To find this information, you can use the command va. You can submit the job with sbatch <script_name>. This job assumes there is a python script in your working directory (included in downloadable example above).

Submission Script

#!/bin/bash

#SBATCH --job-name=Sample_Slurm_Job
#SBATCH --ntasks=1
#SBATCH --nodes=1
#SBATCH --time=00:01:00
#SBATCH --partition=standard
#SBATCH --account=hpcteam

module load python/3.8

python3 hello_world.py

Example Python Script

#!/usr/bin/env python3
import os
print("Hello world! I'm running on compute node: %s"%os.environ["HOSTNAME"])

Example Job Submission

(ocelote) [netid@junonia ~]$ ls
hello_world.py serial-job.slurm
(ocelote) [netid@junonia ~]$ sbatch serial-job.slurm
Submitted batch job 73224

Output

(ocelote) [netid@junonia ~]$ ls
slurm-73224.out hello_world.py serial-job.slurm
(ocelote) [netid@junonia ~]$ cat slurm-73224.out
Hello world! I'm running on compute node: i4n0
Detailed performance metrics for this job will be available at https://metrics.hpc.arizona.edu/#job_viewer?action=show&realm=SUPREMM&resource_id=5&local_job_id=73224 by 8am on 2021/08/05.
(ocelote) [netid@junonia ~]$
 Single Node MPI Job

Single Node MPI Job

Script that compiles and runs an MPI job using 30 CPUs on a single node.

Note: the C file can also be compiled manually in an interactive session.

Slurm Script:

#!/bin/bash
#SBATCH --job-name=Single-Node-MPI-Job
#SBATCH --ntasks=30
#SBATCH --nodes=1             
#SBATCH --time=00:01:00   
#SBATCH --partition=standard
#SBATCH --account=YOUR_GROUP

module load gn8 openmpi3
mpicc -o hello_world hello_world.c  
/usr/bin/time mpirun -np $SLURM_NTASKS ./hello_world

Companion MPI Script

The following C script was used to create the executable for this workflow. It is included in the example available for download above

#include <mpi.h>
#include <stdio.h>

int main(int argc, char** argv) {
    MPI_Init(NULL, NULL);
    int world_size;
    MPI_Comm_size(MPI_COMM_WORLD, &world_size);
    int world_rank;
    MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
    char processor_name[MPI_MAX_PROCESSOR_NAME];
    int name_len;
    MPI_Get_processor_name(processor_name, &name_len);
    printf("Hello world from node %s. My rank is %d out of %d processors\n",
           processor_name, world_rank, world_size);
    MPI_Finalize();
}

To compile the script manually, start an interactive terminal session using interactive, then:

module load gnu8 openmpi3
mpicc -o hello_world hello_world.c

Script Submission

(puma) [netid@junonia ~]$ sbatch Single-Node-MPI-Job.slurm 
Submitted batch job 1694351

Output Files

(puma) [netid@junonia ~]$ ls *.out
slurm-1694351.out

Additionally, the executable hello_world will be generated and stored in your working directory

File Contents

(puma) [netid@junonia ~]$ head slurm-1694351.out 
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 28 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 8 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 14 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 17 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 6 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 9 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 10 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 2 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 4 out of 30 processors
Hello world from node r2u05n1.puma.hpc.arizona.edu. My rank is 5 out of 30 processors
 Basic Array Job

Basic Array Job

Array jobs are used to execute the same script multiple times with different input.

What problem does this help fix?

To execute multiple analyses, a user may be tempted to submit jobs with a scripted loop, e.g.:

for i in $( seq 1 10 ); do sbatch script.slurm <submission options> ; done

This isn’t a good solution because it submits too many jobs too quickly and overloads the scheduler. Instead, an array job can be used to achieve the same ends.

Example

#!/bin/bash
#SBATCH --ntasks=1
#SBATCH --nodes=1             
#SBATCH --time=00:01:00   
#SBATCH --partition=standard
#SBATCH --account=YOUR_GROUP
#SBATCH --array 1-5

echo "./sample_command input_file_${SLURM_ARRAY_TASK_ID}.in"

Script Breakdown

What differentiates the script above from standard submissions is the --array directive. This is what tells SLURM that you’re submitting an array. Following this flag, you will specify the number of jobs you wish to run. In this case, we’re running 5:

#SBATCH --array 1-5

Each job in the array has its own associated environment variable SLURM_ARRARY_TASK_ID that can be used to differentiate subjobs. To demonstrate how we can use each of these to read in different input files, we’ll print a sample command:

echo "./sample_command input_file_${SLURM_ARRAY_TASK_ID}.in"

Script Submission

(ocelote) [netid@junonia ~]$ sbatch basic_array_job.slurm 
Submitted batch job 73958

Output Files

Each of the subjobs in the array will produce its own output file of the form slurm_jobid_arrayid.out as seen below:

echo "./sample_command input_file_${SLURM_ARRAY_TASK_ID}.in"

For more information on naming SLURM files, see our online documentation.

File Contents

Below is a concatenation of the job’s output files. Notice how the array indices function to differentiate the input files in the sample command:

(ocelote) [netid@junonia ~]$ cat slurm-73958_* | grep sample
./sample_command input_file_1.in
./sample_command input_file_2.in
./sample_command input_file_3.in
./sample_command input_file_4.in
./sample_command input_file_5.in
 Blank

Heading

Text

  • No labels