Slurm Dashboard (Enhanced)

A dashboard for interacting with the Slurm workload manager in VSCode. It allows you to see running jobs, cancel jobs, submit new jobs, and more.

Note: This is an enhanced version of the original Slurm Dashboard by Daniel Nichols, featuring enhanced GPU monitoring, job templates, resource filtering, and improved user interface. The original work is licensed under Apache License 2.0.

Enhanced Features

This enhanced version adds:

• Enhanced GPU Monitoring: Accurate GPU allocation tracking and display
• Job Script Templates: Create new job scripts from customizable templates
• Resource Filtering: Filter nodes by name patterns
• Nested Job Display: Better organization with user information in nested views
• Time Tracking: Display elapsed time and walltime for jobs
• Template Management: Add, edit, and manage job script templates

Features

There are two main views in this extension: Job Queue and Job Scripts. The first view is a list of all current user jobs in the queue. Actions are provided for canceling, resubmitting, and inspecting the jobs. The second view shows all detected job scripts and provides actions for inspecting and/or running them. These views are shown in the highlight below.

Requirements

This extension requires Slurm to be on the system where the workspace is located. Currently it only works with Slurm, but PBS, BSUB, and Flux support is planned.

Extension Settings

This extension contributes the following settings:

• slurm-dashboard.job-dashboard.showJobInfo: Show all job metadata in the job list. Default: false
• slurm-dashboard.job-dashboard.promptBeforeCancel: Prompt user before canceling a job. Default: true
• slurm-dashboard.job-dashboard.promptBeforeCancelAll: Prompt user before cancelling all jobs. Default: true
• slurm-dashboard.job-dashboard.refreshInterval: How many seconds between refreshes of the job queue view. Set to null to turn off auto-refresh. Default: 300
• slurm-dashboard.job-dashboard.extrapolationInterval: Extrapolate the job run times in the UI without querying the workload manager. Allows you to set a high job-dashboard.refreshInterval, but still see more realistic job run times. This has the danger of coming out of sync with the real times or not portraying job completion/failure, so it is turned off by default. Provide a number to specify the interval in seconds or null to disable extrapolation. Default: null
• slurm-dashboard.job-dashboard.persistJobs: Keep completed jobs in the job dashboard until they are removed with the X button. Default: false
• slurm-dashboard.job-dashboard.useNativeIcons: Instead of the job status icons shipped with the extension, use VSCode native codicons. Default: false
• slurm-dashboard.job-dashboard.sortBy: Sort the job list by this column. Set to null to leave the order returned by the workload manager. Choices: id, name, time left, status. Default: null
• [DEPRECATED] slurm-dashboard.submit-dashboard.jobScriptExtensions: Use jobScriptPatterns setting instead. File extensions used to identify job scripts. Default: [".sbatch", ".slurm", ".job"]
• slurm-dashboard.submit-dashboard.jobScriptPatterns: Glob patterns to identify job scripts. Default: ["**/*.slurm", "**/*.sbatch", "**/*.job"]
• slurm-dashboard.submit-dashboard.promptBeforeSubmitAll: Prompt user before submitting all job scripts. Default: true
• slurm-dashboard.submit-dashboard.sortBy: Sort the job script list by this column. Set to null to leave the order the glob pattern discovered the files. Choices: filename, rel path, last modified, newest, oldest. Default: last modified
• slurm-dashboard.setJobWorkingDirectoryToScriptDirectory: Launch job scripts with the working directory as the location of the job script. Default: true
• slurm-dashboard.backend: Scheduler backend. Choices: slurm, debug. Default: slurm
• slurm-dashboard.slurm-backend.squeueUserArg: By default --me is passed to squeue to get the users jobs. Slurm versions older than 20.02 do not support --me. This settings provides a way to fallback to the --user flag instead. Choices: me, user. Default: me
• slurm-dashboard.resource-dashboard.refreshInterval: How many seconds between refreshing resource data. Set to null to disable automatic refresh. Default: 30
• slurm-dashboard.resource-dashboard.nodeFilter: Filter nodes by name patterns. Empty array shows all nodes. Use glob patterns like ["gpu*", "dgx*"] to show only GPU and DGX nodes. Default: []
• slurm-dashboard.resource-dashboard.showNodeDetails: Show detailed resource information for each node when expanding in the tree view. Default: true

Most notable is the job-dashboard.refreshInterval setting. The job queue view refreshes its data at regular intervals. To avoid overloading the login nodes or the workload manager the refresh interval is set to a high value. By default it is 300 seconds. This can be changed to the users preference or auto-refresh can be turned off entirely by setting it to null.

If you still want to see live job times, you can set job-dashboard.extrapolationInterval, which will extrapolate the current job time based on the last available value from the workload manager. This will make the UI more informative, but has the danger of coming out of sync with the real job times and/or not capturing job completion and failure. However, these will be corrected the next time the jobs are refreshed from the queue. Due to these reasons extrapolation is turned off by default. An example of what this looks like when enabled is shown below.

Known Issues

Submitting Issues

Issues can be submitted under the issues tab on GitHub.

Release Notes

See the Changelog for the entire change history.

0.0.14

• add job error script button to quickly pull up the error output path of a running job
• update and clean up documentation and ci
• add contributing guidelines

Contributing

This enhanced extension is completely open source and contributions are welcome. See the contributing guidelines for more info.

The original Slurm Dashboard is also open source by Daniel Nichols.

License

Original Work: Copyright Daniel Nichols 2024
Enhanced Version: Copyright Francesco Ortu 2025

Both the original work and this enhanced version are published under the Apache 2.0 license.

This enhanced version is a derivative work based on the original Slurm Dashboard by Daniel Nichols, and is distributed under the same Apache 2.0 license terms.