SLURM HPC Plugin

https://slurm.schedmd.com/documentation.html

This HPC environment can be selected via --execenv=hpc.slurm. In this HPC environment, SIERRA will run experiments spread across multiple nodes allocated by the SLURM scheduler. The following table describes the SLURM-SIERRA interface. Some SLURM environment variables are used by SIERRA to configure experiments during stage 1,2; if they are not defined SIERRA will throw an error.

SLURM-SIERRA interface

Environment variable	SIERRA context	Command line override
PARALLEL	Used to transfer environment variables into the GNU parallel environment.	N/A
PARALLEL_SHELL	Used to set the shell used by GNU parallel to execute all commands in. Overwritten by SIERRA to /bin/bash.	N/A
LD_LIBRARY_PATH	Exported by SIERRA via PARALLEL to child GNU parallel processes. Can be undefined when SIERRA starts.	N/A
PYTHONPATH	Exported by SIERRA via PARALLEL to child GNU parallel processes. Can be undefined when SIERRA starts.	N/A
PATH	Exported by SIERRA via PARALLEL to child GNU parallel processes. Can be undefined when SIERRA starts.	N/A
SLURM_CPUS_PER_TASK	Used to set # threads per experimental node for each allocated compute node.	N/A
SLURM_TASKS_PER_NODE	Used to set # parallel jobs per allocated compute node.	--exec-jobs-per-node
SLURM_JOB_NODELIST	Obtaining the list of nodes allocated to a job which SIERRA can direct GNU parallel to use for experiments.	N/A
SLURM_JOB_ID	Creating the UUID nodelist file passed to GNU parallel, guaranteeing no collisions (i.e., simultaneous SIERRA invocations sharing allocated nodes if multiple jobs are started from the same directory).	N/A