coexist.schedulers.SlurmScheduler

class coexist.schedulers.SlurmScheduler(time, qos=None, mail_type='FAIL', ntasks='1', mem=None, commands='set -e\n', interpreter='python', script='access_slurm_submission.sh', **kwargs)

Bases: Scheduler

Launch simulations on a SLURM distributed cluster using sbatch.

First a bash script must be defined for launching each simulation job; this class generates this script, but some details must be defined by you; they are specified as class parameters, see examples below.

Parameters:

timestr

The time allocated for a single simulation, given as a string, e.g. “1:0:0”. Will be added as “#SBATCH –time 1:0:0”.

qosstr, optional

The “#SBATCH –qos bbdefault” sbatch command.

mail_typestr, default “FAIL”

The “#SBATCH –mail-type FAIL” sbatch command.

ntasksstr, default “1”

The “#SBATCH –ntasks 1” sbatch command.

memstr, optional

The “#SBATCH –mem 4G” sbatch command.

commandsstr or list[str], default “module load Python”

Any other non-SLURM commands to run in the job submission script before executing the simulation; this is normally the setup work, e.g. loading necessary modules, environments, etc.

interpreterstr, default os.path.split(sys.executable)[1]

Name of the Python interpreter which will be used to execute the simulation script; this is normally set to the name of the executable used to run ACCES, e.g. “usr/bin/python3” -> “python3”.

**kwargsother keyword arguments

Other “#SBATCH” commands to include at the top of the job submission script; e.g. constraint = "cascadelake" is transformed into "#SBATCH --constraint cascadelake"; mem_per_cpu = "4" is transformed into "#SBATCH --mem-per-cpu 4".

Examples

>>> from coexist.schedulers import SlurmScheduler
>>> scheduler = SlurmScheduler(
>>>     "10:0:0",          # Time allocated for a single simulation
>>>     commands = """
>>>         # Commands to add in the sbatch script after `#`
>>>         set -e
>>>         module purge; module load bluebear
>>>         module load BEAR-Python-DataScience
>>>     """,
>>>     qos = "bbdefault",
>>>     constraint = "cascadelake",   # Any other #SBATCH --<CMD> = "VAL"
>>> )

__init__(time, qos=None, mail_type='FAIL', ntasks='1', mem=None, commands='set -e\n', interpreter='python', script='access_slurm_submission.sh', **kwargs)

Methods

__init__(time[, qos, mail_type, ntasks, ...])
generate(scriptpath)
schedule(dirpath, index)

generate(scriptpath)

schedule(dirpath, index)