nanoslurm

nanoslurm is a zero-dependency Python wrapper for SLURM job submission and monitoring.
It uses a tiny POSIX-compatible shell script to call sbatch and related commands, avoiding any heavy Python dependencies.

Features

• Submit jobs from Python without pyslurm or other packages
• Monitor status (PENDING, RUNNING, COMPLETED, etc.)
• Cancel jobs
• Tail job logs
• Get detailed info via scontrol
• Respects working directory at runtime (sbatch -D)

Requirements

• SLURM cluster with sbatch, squeue, and optionally sacct / scontrol
• Python ≥ 3.11
• Linux operating system

Quickstart

import nanoslurm

job = nanoslurm.submit(
    command=["python", "train.py", "--epochs", "10"],
    name="my_job",
    cluster="gpu22",
    time="01:00:00",
    cpus=4,
    memory=16,
    gpus=1,  # optional; omit or set to 0 for CPU-only jobs
    stdout_file="./slurm_logs/%j.txt",
    stderr_file="./slurm_logs/%j.err",
    signal="SIGUSR1@90",
    workdir="."
)

print(job)                      # Job(id=123456, name='my_job_2025-08-08_09-12-33.123', ...)
print(job.status)               # "PENDING", "RUNNING", ...
print(job.is_running())         # True / False
print(job.is_finished())        # True / False
print(job.info())               # Detailed dict from scontrol
job.tail(10)                    # Last 10 lines of stdout
job.wait(poll_interval=5)       # Wait until completion
job.cancel()                    # Cancel job

For CPU-only workloads simply omit the gpus argument (or set it to 0) and nanoslurm will not add any --gres directives to the submission script.

Releasing

Bump the version in pyproject.toml and merge the change into main. A workflow will tag the commit as vX.Y.Z and publish the package to PyPI.