=============== SEAMM JobServer

.. image:: https://img.shields.io/github/issues-pr-raw/molssi-seamm/seamm_jobserver :target: https://github.com/molssi-seamm/seamm_jobserver/pulls :alt: GitHub pull requests

.. image:: https://github.com/molssi-seamm/seamm_jobserver/workflows/CI/badge.svg :target: https://github.com/molssi-seamm/seamm_jobserver/actions :alt: Build Status

.. image:: https://codecov.io/gh/molssi-seamm/seamm_jobserver/branch/main/graph/badge.svg :target: https://codecov.io/gh/molssi-seamm/seamm_jobserver :alt: Code Coverage

.. image:: https://github.com/molssi-seamm/seamm_jobserver/workflows/CodeQL/badge.svg :target: https://github.com/molssi-seamm/seamm_jobserver/security/code-scanning :alt: Code Quality

.. image:: https://github.com/molssi-seamm/seamm_jobserver/workflows/Release/badge.svg :target: https://molssi-seamm.github.io/seamm_jobserver/index.html :alt: Documentation Status

.. image:: https://img.shields.io/pypi/v/seamm_jobserver.svg :target: https://pypi.python.org/pypi/seamm_jobserver :alt: PyPi VERSION

The JobServer for the SEAMM environment

• Free software: BSD license
• Documentation: https://molssi-seamm.github.io/seamm_jobserver/index.html
• Code: https://github.com/molssi-seamm/seamm_jobserver

Features

The SEAMM JobServer handles executing Jobs for the Dashboard, either as local subprocesses (the default) or, if configured, by submitting each job as a SLURM batch job -- see the User Guide.

Acknowledgements

This package was created with the molssi-seamm/cookiecutter-seamm-plugin_ tool, which is based on the excellent Cookiecutter_.

.. _Cookiecutter: https://github.com/audreyr/cookiecutter .. _molssi-seamm/cookiecutter-seamm-plugin: https://github.com/molssi-seamm/cookiecutter-seamm-plugin

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

.. _MolSSI: https://molssi.org .. _National Science Foundation: https://www.nsf.gov

======= History

2026.8.11 -- Honor a queue's setup= shell commands before running a job * A SLURM queue's new setup directive (see seamm_slurm's 2026.8.11 release) is now prepended to the generated sbatch script, before run_from_jobserver runs -- e.g. setup = module load ORCA for a queue whose own submission environment doesn't otherwise carry whatever a code's installation = modules setting needs. No effect for a queue that doesn't set it.

2026.8.10 -- Route jobs to multiple queues/clusters from one JobServer instance * A JobServer instance can now be configured with more than one queue (cluster/section) in its <root>/<jobserver-name>.ini and route each job to the one it asks for (parameters["queue"]), falling back to the instance's default queue if it doesn't specify one. Requesting an unknown queue, or no queue with no default configured, fails the job immediately (startup error) rather than silently running it somewhere unintended. * Each queue's max_concurrent_jobs is enforced independently, so one full queue never blocks another from accepting new jobs, and SLURM polling is batched per queue so two different clusters' same-numbered SLURM job ids are never compared against each other. * A type = local queue (see seamm_slurm's 2026.8.10 release) and one or more type = slurm queues can now coexist on the same JobServer instance -- some jobs run as local subprocesses, others dispatch to real SLURM clusters, from one process. * A SLURM job's job_data.json now records which queue it ran on and its SLURM job id (found from real usage: nothing previously told a user which cluster a finished job had actually run on). * No <root>/<jobserver-name>.ini at all means behavior is unchanged from before this feature existed -- every job still runs as an uncapped local subprocess.

2026.8.9 -- Bugfix: startup reattachment could misapply SLURM staging to a local job * On restart, a job's running/kill datastore row was reattached based on whether the JobServer instance is currently SLURM-configured, not on what that specific job actually is. A local job predating SLURM being enabled on this instance (or, worse, one still genuinely running) could be treated as a lost SLURM submission: at best retrying a pointless remote file transfer forever instead of reading its already-correct local status, at worst getting resubmitted via SLURM as a real duplicate run of a job still running locally. Reattachment now checks each job's own recorded pid or SLURM job id, not the instance's current mode, so a local job is always recognized as local regardless of whether SLURM is configured. * Bugfix: a locally-run job resumed after a restart could raise an error the next time it finished, from a missing internal field. * A locally-run job whose process ended with no other evidence of its outcome now has its actual result read from the job's own record rather than being assumed successful.

2026.8.8 -- Support SLURM dispatch to a cluster with no shared filesystem * A JobServer can now dispatch jobs via SLURM to a cluster it shares no filesystem with -- for example, a laptop reaching a remote cluster over ssh. A job's working directory is staged there before submission and its results staged back after, automatically. Configure it with new remote_root and remote_run_from_jobserver (or remote_conda_env) options in the relevant section of <root>/<jobserver-name>.ini. See the User Guide for the full option list. * Internal: the JobServer itself, not the running job, now writes a job's final status to the datastore, for every job type (this was already true for the SLURM-losing-track case; now it's the normal path for local jobs too). Fixes a real gap: a local job that crashed hard before it could write its own status previously stayed stuck as running forever, with no recovery -- the JobServer now finalizes it from the process's exit status instead. * Live-validated end to end against a real cluster: a real job staged out via rsync, submitted via sbatch, confirmed completed via SLURM's own accounting, and staged back correctly.

2026.8.7 -- Actively stop deleted or explicitly-killed jobs * Deleting a job (e.g. via the dashboard) removes its row and files, but previously left whatever was actually running it untouched -- it just ran on until it crashed on its own missing files. The JobServer now notices a tracked job's row disappearing and actively stops it (scancel in SLURM mode, terminating the process in local mode), every poll cycle. * A job can also be stopped without deleting it or its files, by setting its status to kill -- an ordinary status update, no new API. The JobServer stops the run the same way, then sets status to killed. A job killed before it was ever started is finalized as killed directly. * Both checks are also applied during startup reattachment, so a kill requested just before a JobServer restart is not lost or, worse, mistaken for a merely-lost job and resubmitted.

2026.8.6 -- Add SLURM submission mode (Option 1: whole-flowchart sbatch) * A JobServer instance can now submit each job as a single SLURM batch job instead of a local subprocess, controlled by a new, opt-in <root>/<jobserver-name>.ini config file (absent means unchanged, local-subprocess behavior). Supports both running the SLURM CLI directly on the JobServer's own host and reaching a remote cluster over SSH. * Added a configurable cap on how many jobs a JobServer instance keeps outstanding in SLURM at once. * If a tracked job's SLURM state goes missing or terminal while its datastore row still shows running (a node failure, an OOM kill, a cancellation, or the JobServer itself having restarted), the JobServer now trusts the job's own recorded outcome when available, otherwise safely resubmits it (flowcharts checkpoint and resume, so this does not redo completed work) up to a configurable retry cap before giving up. * A job can now request different SLURM resources (cores, memory, walltime, ...) than the JobServer instance's own defaults, via parameters["slurm"] on the job row -- gated by a new, optional [<section>.limits] ini section that says which fields a job may override and within what bounds (enumerated choices or numeric/size/ time ranges). Secure by default: no .limits section means nothing is overridable. Always re-validated server-side regardless of what submitted the job. * See the User Guide for the config file format and behavior.

2025.11.22 -- Bugfix: Catching errors when starting a job. * Now catch any error that occurs starting a job, and set its status to 'startup error'. This prevents the JobServer getting into a loop repeatedly trying to submit the job.

2025.11.12 -- Added SEAMM_JOBSERVER and SEAMM_JOB_ID environment variables This PR adds support for passing job-specific metadata to spawned processes through environment variables. Jobs can now access their unique job ID and the name of the JobServer that spawned them.

* Added SEAMM_JOB_ID and SEAMM_JOBSERVER environment variables for spawned job
  processes 
* Added --name command-line argument to specify JobServer name (defaults to
  hostname)
* Cleaned up docstring formatting

2024.4.12 -- Fixed issue with status of finished jobs

• Fixed a problem if a job returned a status of None, which was reported as an error.

2024.4.11 -- Correcting description of this package

2024.4.5 -- Adding support for debugging

• Use the value of the environment variable SEAMM_LOG_LEVEL to set the log level for jobs. DEBUG, INFO, WARNING are three useful levels.

2024.1.17 -- Changes to support running in Docker containers.

2023.12.12 -- Improved the output in the GUI.

• Improved the output to the GUI
• Fixed a bug in the file path for the status file.

2023.3.23 -- Substantial improvements to JobServer

• Switched to independent process for Jobs, which means they are fully independent of the JobServer and continue to run if the JobServer stops
• Discover existing running jobs on startup and monitor them.
• Added status information for the machine the JobServer is on as well as Jobs
• Provide a GUI if run from the commandline, showing the log and status.

0.9.1 (2020-05-29)

• First release on PyPI.