This page was translated by an AI (LLM) with a cursory human check and is awaiting full review.
Slurm
Jobs are managed across all nodes by the Slurm software.
- To submit a submission script:
sbatch script.slurm - To monitor pending or running jobs:
squeue --me
This command displays the information in the following format:whereJOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)235 part_name test abc R 00:02 1 noeud123- JOBID: the job identifier
- PARTITION: the partition used
- NAME: the job name
- USER: the username of the job owner
- ST: the job execution status (R=running, PD=pending, CG=completing, ...)
- TIME: the elapsed time since the start of the run (ST = R)
- NODES: the number of nodes used
- NODELIST: the list of nodes used.
You can use the option --start so that Slurm
displays an estimate of the start time of your jobs
(column "START_TIME"). It is possible that Slurm is not yet
able to provide a sufficiently reliable estimate of the
start time of certain jobs, in which case the information will
appear as not available ("N/A").
As the list of pending jobs is evolving, it is important
to keep in mind that the information provided by Slurm is only
an estimate that is likely to change depending on the machine load.
- To obtain complete information (allocated resources and execution status)
on a pending or running job:
scontrol show job JOBID - To cancel an execution:
scancel JOBID
Remarks
-
A comprehensive table of Slurm commands is available in the official documentation.
-
In the event of a hardware failure on the machine, the default Slurm configuration ensures that running jobs are automatically restarted. If your code does not support checkpoints, the computation will automatically restart from the beginning. If you want to avoid this behaviour, you need to use the option
--no-requeuewhen submitting, i.e. submit your job withsbatch --no-requeue script.slurmor add the following directive to your submission script:#SBATCH --no-requeue