Difference between revisions of "Submitting a job to a slurm queue"

From HPC Guide
Jump to navigation Jump to search
Line 26: Line 26:
 
'''srun'''
 
'''srun'''
 
<pre>
 
<pre>
'''srun -intasks=2 --label hostname''' (--label means that before the output line write the task id)
+
srun -intasks=2 --label hostname  (--label means that before the output line write the task id)
 +
<pre>
 
0:compute-0-1
 
0:compute-0-1
 
1:compute-0-1
 
1:compute-0-1
 +
</pre>
 
</pre>
 
</pre>
  

Revision as of 15:15, 13 March 2023

submit commands

sbatch - submits script

salloc - submit interactive job - allocates what it needs, but will not start to work on the node/s

srun - submits interactive job w mpi ("job step")

sattach - connect stdin/out/err for an existing job (or job step)

So for example, may submit a job with command:

sbatch script.sh

Examples

sbatch

sbatch --ntasks=1 --time=10 pre_process.bash (time is 10 minutes)
(Submitted batch job 45001)
sbatch --ntasks=128 --time=60 --depend=45001 do_work.bash
(Submitted batch job 45002)
sbatch --ntasks=1 --time=30 --depend=45002 post_process.bash
(Submitted batch job 45003)

srun

srun -intasks=2 --label hostname  (--label means that before the output line write the task id)
<pre>
0:compute-0-1
1:compute-0-1

Using 2 nodes:

srun -innodes=2 --exclusive --label hostname
0:compute-0-1
1:compute-0-2

opening bash

srun --ntasks=56 --pty bash
[dolevg@compute-0-12 beta16.dvory]$....

Specifying compute node (which is available)

srun --ntasks=56 -p gcohen_2018 --nodelist="compute-0-12" --pty bash

See available nodes: salloc

salloc --ntasks=8 --time=10 bash
salloc: Granted job allocation 45000
(gives us a bash prompts of a node:)
env | grep SLURM
SLURM_JOBID=45000
SLURM_NPROCS=4
SLURM_JOB_NODELIST=compute-0-1,compute-0-2
hostname
powerlogin
srun --label hostname
0:compute-0-1
1:compute-0-1
2:compute-0-2
3:compute-0-2
exit (terminates the shell)

info commands

sinfo -- to see all queues (partitions)

squeue -- to see all jobs

scontrol show partition -- to see all partitions

scontrol show job <number> -- to see job's attributes