slurm-jobmanager 0.1.4

CLI tool to manage SLURM jobs via sacct/squeue/sbatch/scancel

jobmanager

A command-line tool to manage SLURM jobs via sacct, squeue, sbatch, and scancel.

Requirements

• Python 3
• SLURM (sacct, squeue, scontrol, scancel)

Installation

pip install slurm-jobmanager

Or install directly from the repository:

git clone https://github.com/Dannyzimmer/SlurmJobManager
cd SlurmJobManager
pip install .

Intended workflow

[!Important] jobmanager is designed to run only one job at a time for a given directory. This encourages an organized workflow where each job has its own directory with associated metadata and logs. Running multiple jobs in the same directory is not supported and may lead to incorrect behavior.

Each job lives in its own directory containing a metadata.json file (auto generated via sacct -j <jobid> --json > metadata.json). The typical usage pattern is:

jobs/
├── run_001/
│   ├── metadata.json
│   └── job_logs
│       ├── LOG_run_001_6552064.log
│       └── LOG_run_001_6552081.log
├── run_002/
│   ├── metadata.json
│   └── job_logs
│       ├── LOG_run_001_6552092.log
│       └── LOG_run_001_6552101.log

Navigate to the job directory and run commands without extra flags — jobmanager picks up metadata.json automatically:

cd jobs/run_001
jobmanager status
jobmanager watch

Job identification

All job-specific commands resolve the target job in this order:

1. --jobid ID — fetch metadata from SLURM on the fly
2. --metadata FILE — use an explicit metadata file
3. (default) — use metadata.json in the current directory

Commands

list

List your active jobs (wraps squeue).

jobmanager list
jobmanager list --user alice

status

Print the current state of a job (RUNNING, PENDING, COMPLETED, …).

jobmanager status

summary

Print a full summary: partition, CPUs, memory, elapsed time, state, working directory, etc.

jobmanager summary

elapsed

Print the elapsed wall-clock time of a job.

jobmanager elapsed

watch

Stream the job's stdout in real time until it reaches a terminal state.

jobmanager watch
jobmanager watch --interval 10   # poll every 10 s (default: 5)

stop

Cancel a job (scancel). Prompts for confirmation unless --yes is passed.

jobmanager stop
jobmanager stop --yes

print-request

Print a SLURM batch script header reproducing the resource requests of the job. Redirect to a file to reuse it.

jobmanager print-request > job.sh

resubmit

Resubmit a job using the original submission command recovered from sacct. If the current job is still active, prompts to cancel it first.

jobmanager resubmit

template

Create a run.sh SLURM batch script in the current directory. Job name is set automatically from the directory name.

jobmanager template            # blank placeholders (must be filled before submitting)
jobmanager template --test     # 1 node, 1 task, 1 CPU, 1 GB, 1 min
jobmanager template --quick    # 1 node, 1 task, 2 CPUs, 4 GB, 1 h
jobmanager template --medium   # 1 node, 4 tasks, 8 CPUs, 32 GB, 24 h

submit

Validate and submit a batch script via sbatch. Checks for zero-value resource directives and warns if the current directory's job is still active.

jobmanager submit              # uses run.sh in the current directory
jobmanager submit path/to/script.sh