Jean Zay: Slurm commands

Jobs are managed on all the nodes by the software Slurm .

  • To submit a submission script:

     $ sbatch script.slurm 
  • To monitor jobs which are waiting or in execution:

     $ squeue -u $USER 

    This command displays information in the following form:

    JOBID  PARTITION  NAME  USER  ST   TIME  NODES  NODELIST(REASON)   
      235  part_name  test   abc   R  00:02      1  r6i3n1 

    Where
    JOBID: Job identifier
    PARTITION: Partition used
    NAME: Job name
    USER: User name of job owner
    ST: Status of job execution ( R=running, PD=pending, CG=completing )
    TIME: Elapsed time
    NODES: Number of nodes used
    NODELIST: List of nodes used

    Note: You can use the --start option to display an estimated start time for your jobs (“START_TIME” column). Slurm might not have a reliable estimation for the start time of some jobs, in this case the information will show as not available (“N/A”). Since the list of pending jobs is always evolving, it is important to note that the information given by Slurm is only an estimate which might change depending on the machine load.

  • To obtain complete information about a job (allocated resources and execution status) :

     $ scontrol show job JOBID 
  • To cancel an execution:

     $ scancel JOBID 

Comments

  • A complete reference table of Slurm commands is available here .
  • In case of a problem on the machine, the SLURM default configuration is such that the running jobs are automatically restarted from scratch. If you want to avoid this behavior, you should use the --no-requeue option in the submission process, that is, submit your job doing

     $ sbatch --no-requeue script.slurm 

    or add the following line in your submission script:

    #SBATCH --no-requeue