seamm_slurm

A small library that wraps the SLURM command-line tools (sbatch, squeue, sacct, scancel) behind one interface, SlurmBackend, with two transports:

• LocalSlurm -- runs the SLURM CLI directly on the current host (the host running the code is itself a SLURM submit host, e.g. a cluster head/login node).
• SshSlurm -- runs the same commands on a remote host over passwordless SSH (the host running the code is not a SLURM submit host).

Both transports share the same submit/poll/cancel logic and SLURM-version handling (some clusters support squeue --json/sacct --json, others only have plain columnar output -- this library prefers JSON when available and falls back to --parsable2/--format= text parsing otherwise, transparently to the caller).

submit() takes the full script text (shebang, #SBATCH directives, payload) and feeds it to sbatch --parsable on stdin -- it never needs the script to exist as a file on the target host, so submission works identically whether the caller and the SLURM cluster share a filesystem or not. seamm_slurm.script.build_script() is a small helper for building that script text from a directives dict (partition/account/qos/nodes/ntasks/time/ mem/gpus/...) plus a payload command, matching the <root>/<jobserver-name>.ini config shape used by seamm_jobserver.

poll_many() takes a batch of SLURM job IDs (not one call per job) and returns a JobStatus per ID, merging live state from squeue with historical/terminal state from sacct -- sacct is the source of truth once a job has left the live queue.

seamm_slurm.config reads a JobServer's <root>/<jobserver-name>.ini (load_slurm_config/SlurmSection) -- the same system/machine config file seamm_jobserver itself uses, moved here so any dependency-light consumer (a future job-submission UI, for instance) can read and validate it without pulling in the rest of the SEAMM stack. Includes an optional [<section>.limits] companion section and SlurmSection.merge_overrides(), for sites that want to let a job override some of the section's defaults (cores, memory, walltime, ...) within enumerated choices or numeric/size/ time bounds -- secure by default, nothing is overridable unless a site's .limits section says so.

This package intentionally has no SEAMM-core/molsystem/dashboard dependency and no notion of SEAMM's own job-status vocabulary -- it only speaks SLURM's. seamm_jobserver's whole-flowchart SLURM submission mode is the first consumer, validated end-to-end against a real cluster; a future seamm_exec Slurm executor (per-step submission) is expected to reuse the same backend rather than duplicating the SLURM-CLI handling.

See docs/developer_guide/campaigns/2026-08-06/index.rst in this repo, and seamm_jobserver's own docs/developer_guide/campaigns/2026-08-05/ (the full cross-repo SLURM-integration campaign, moved there from a workspace-level scratch doc), for the design rationale.