seamm_scheduler

Queueing systems for SEAMM: one module per queueing system behind a shared Scheduler interface, with local and ssh transports and staging.

• seamm_scheduler.slurm.Slurm -- SLURM: #SBATCH directives, sbatch, squeue/sacct (with --json where the cluster has it, text otherwise), scancel.
• seamm_scheduler.pbs.Pbs -- PBS Professional / OpenPBS: #PBS directives with one select statement, qsub, qstat, qdel. Tested against recorded output only so far.
• QueueBackend(scheduler, transport) -- submit a script, poll many jobs in as few commands as the scheduler allows, cancel, count the user's jobs.
• LocalTransport / SshTransport -- run the commands on this host, or on a login node over passwordless ssh.
• LocalStager / RsyncStager -- move job (or task) directories when the caller and the cluster share no filesystem.
• seamm_scheduler.config -- the JobServer's <root>/<jobserver-name>.ini: each section is a target, describing where a job's flowchart evaluator runs and, with the optional task keys, where its tasks run.

Every scheduler translates the same scheduler-neutral resources (ntasks, cpus_per_task, mem_per_cpu, ngpus, walltime, partition, account, qos, nodes) into its own directives, and reports job states in one small vocabulary (pending, running, completed, cancelled, failed).

The package has no dependencies. It is used by seamm_jobserver (whole flowcharts as batch jobs) and seamm_exec's task layer (bundles of tasks as batch jobs). seamm_slurm, which this package generalizes, remains as a compatibility shim that re-exports from here.

Features

• SLURM 20.11 (no --json) to 25.11 (nested JSON) handled transparently
• Per-user queue limits (count_jobs())
• Poll failures (an ssh outage) are reported as such, never as vanished jobs
• One rsync for many directories (push/pull)

Acknowledgements

Developed by the Molecular Sciences Software Institute (MolSSI), which receives funding from the National Science Foundation under award CHE-2136142.