slurm

Submits job scripts to Simple Linux Utility for Resource Management.

Uses the sbatch command. SLURM directives can be provided in the flow.cylc file:

global.cylc

[platforms]
    [[slurm_platform]]
        job runner = slurm

flow.cylc

[runtime]
    [[my_task]]
        platform = slurm_platform
        execution time limit = PT1H
        [[[directives]]]
            --nodes = 5
            --account = QXZ5W2

Note

• Cylc requires long form directives (e.g. --begin not -b).

• Cylc requires an = even if the directive does not have a value (e.g. --hold= not --hold).

• If a directive does not have a value you may use the short form (e.g. -H=). But the directive must still be suffixed with an =.

These are written to the top of the job script like this:

#!/bin/bash
#SBATCH --nodes=5
#SBATCH --time=60:00
#SBATCH --account=QXZ5W2

If execution time limit is specified, it is used to generate the --time directive. Do not specify the --time directive explicitly if execution time limit is specified. Otherwise, the execution time limit known by the workflow may be out of sync with what is submitted to the job runner.

Cylc supports heterogeneous Slurm jobs via special numbered directive prefixes that distinguish repeated directives from one another:

[runtime]
    # run two heterogenous job components:
    script = srun sleep 10 : sleep 30
    [[my_task]]
        execution time limit = PT1H
        platform = slurm_platform
        [[[directives]]]
            --account = QXZ5W2
            hetjob_0_--mem = 1G  # first prefix must be "0"
            hetjob_0_--nodes = 3
            hetjob_1_--mem = 2G
            hetjob_1_--nodes = 6

The resulting formatted directives are:

#!/bin/bash
#SBATCH --time=60:00
#SBATCH --account=QXZ5W2
#SBATCH --mem=1G
#SBATCH --nodes=3
#SBATCH hetjob
#SBATCH --mem=2G
#SBATCH --nodes=6

Note

For older Slurm versions with packjob instead of hetjob, use global.cylc[platforms][<platform name>]job runner=slurm_packjob and directive prefixes packjob_0_ etc.