SLURM jobscript explained#
Batch jobs can be submitted to SLURM scheduler so they can run when the requested resources are available for use. This section introduces what a jobscript is and how to configure it to request different allocatable resources.
Basic jobscript#
A typical jobscript has two major sections:
SLURM Directives
The commands to run on allocated computational resource
Below is an example jobscript which request 32 cores of a CPU on a KSL system, all collocated on the 1 compute node, with a wall time/duration of 1 hour. It also mentions which SLURM account should the usage be charged to what name should SLURM label it as to make querying it convenient for the user.
#!/bin/bash
#SBATCH --job-name=my-first-job
#SBATCH --ntasks=32
#SBATCH --ntasks-per-node=32
#SBATCH --partition=workq
#SBATCH --account=<project_id>
#SBATCH --time=01:00:00
srun -n 32 ./helloworld
The very first line in the example above activates a new bash shell. This will ensure the environment from the place the job was submitted does not forward to the job. You may append #!/bin/bash --login to this line to allow activating the environment in the file ~/.bashrc in home directory.
Note
The user must ensure that nothing unwanted is loaded in ~/.bashrc or else the job can have unintended behavior.
The lines starting with the clause #SBATCH are directives to SLURM. For any other program, these are treated as comments since they all start with # symbol. As explained above, the job requests 32 cores one node for 1 hour and the usage be charged to k1001 account. The jobscript also prescribes that the job be queued in the SLURM partition workq. Users should check which partitions exist on the KSL system they want to submit their jobs on.
Note
Some SLURM directives have default values. If a jobscript doesn’t have it set, the default values will be assumed. It is always a good practice to set the SLURM directives explicitly for better readability.
To submit the jobscript, sbatch command is used.
sbatch jobscript.slurm