SLURM Cluster Manager

Manage, monitor, and submit SLURM jobs directly from VS Code.

SLURM Cluster Manager brings your HPC workflow into your editor: monitor jobs in real time, inspect logs instantly, compare GPU usage, and take common actions (cancel/pin/submit) without context-switching to a terminal.

✨ Highlights

• Live job monitoring in a dedicated sidebar (Running / Pending / Completing / Job History)
• Visual progress bars for time usage (Elapsed / Time Limit)
• GPU Partition Usage so you can compare partitions before submitting
• Cluster Overview showing which Slurm accounts are using the most GPUs
• GPU stats via nvidia-smi (when available)
• One-click actions: cancel, cancel pending jobs, batch cancel, pin
• Job History grouped by date with configurable lookback range
• Instant log access for stdout / stderr

🚀 Features

Active Job Management

• Real-time Monitoring: View all active jobs at a glance (Running, Pending, Completing, and other active states).
• Time Awareness: Smart progress bars show elapsed vs. requested wall time.
• Resource Stats: Display allocated CPUs, memory, and node count for each job.
• GPU Visibility: Uses nvidia-smi to surface GPU utilization and memory usage where supported.
• One-Click Actions: Cancel or pin jobs directly from the UI.
• Batch Cancel: Select multiple jobs via checkboxes, then cancel them all at once. The "Cancel All" button becomes "Cancel Selected" when jobs are checked. Selections persist across refreshes.
• Pending Cleanup: Cancel all pending jobs without stopping jobs that are already running.
• Smart Pending Display: Pending jobs hide irrelevant info (Nodes, Elapsed, logs) and instead show human-readable pending reasons, estimated start time, and dependency indicators (🔗).
• Job Dependencies: View dependency info (e.g., afterok:12345) in the expanded job details.

Job Array Management

Smart handling of SLURM job arrays with flexible cancellation options:

• Cancel entire array: Remove all jobs in the array at once
• Cancel pending jobs: Cancel only pending jobs, keep running ones
• Cancel specific job(s): Flexible input supporting:
  • Single index: 3
  • Range: 0-10 (indices 0 through 10 inclusive)
  • Step: 0-20:2 (every 2nd job: 0, 2, 4, ..., 20)
  • List: 1,3,5,7 (non-contiguous jobs)
• Bounds validation: Automatically validates against actual array range
• Safety warnings: Extra confirmation when cancelling >100 jobs

Note: Array-level cancel options only appear for pending jobs. Running array tasks are cancelled directly like any individual job.

Job History & Logs

• Integrated Job History: Browse recent completed, failed, and cancelled jobs (default: last 7 days).
• Date Grouping: Jobs are grouped by completion date with compact elapsed-time and end-time labels.
• Job History Range Control: Use the toolbar action to switch between common lookback windows or enter a custom range.
• Refresh Awareness: Job History includes a "last refreshed" row so you know when the data was fetched.
• Instant Log Access: Right-click any job (active or historical) to open its stdout / stderr.
• Smart Path Resolution: Automatically resolves log locations from sbatch directives and scontrol metadata, including relative paths, ~, Slurm filename placeholders, array IDs, escaped spaces, and unavailable paths like (null).

GPU Partition Usage

• GPU-only Partition View: A dedicated sidebar view shows only partitions that advertise GPUs through Slurm GRES.
• Least-used First: Rows are sorted from least used to most used by allocated GPU share, then pending-job pressure, idle GPUs, running jobs, and name.
• Available vs. Total GPUs: Rows distinguish available GPUs from total GPUs so down/draining nodes do not make a partition look more usable than it is.
• Queue Pressure: Running and pending job counts are shown per GPU partition, including pending jobs that target multiple partitions.
• GPU Type Breakdown: Hover a partition row to see GPU types and capacity, such as a100, h200, or generic GPUs.
• Manual refresh only: The view fetches data when opened or manually refreshed, avoiding background load on the Slurm controller.

Workflow Integrations

• Pinning: Keep critical long-running jobs visible even while filtering or sorting.
• Search & Filter: Quickly find jobs by name.
• Cluster Hog Indicators: Two fun indicators at the top of your job list:
  • Job Hog: The user with the most running jobs (🐷 Job Hog, 🔥 Cluster Dominator, 🤗 CUDA Cuddler, 😋 Node Nom-Nom)
  • GPU Hog: The user hoarding the most GPUs (🧛 VRAMpire, 🎮 GPU Gobbler, ⚡ Watt Wizard, 🏋️ Tensor Titan)

Hall of Shame

• Hall of Shame: A dedicated sidebar view ranking GPU users by allocated GPUs and running GPU job count.
• GPU-only rankings: CPU-only jobs and CPU-only users are excluded from the Hall of Shame.
• Slurm account context: Rows show the Slurm account responsible for the GPU jobs, with all accounts listed in the tooltip when a user has jobs under multiple accounts.
• GPU type breakdown: Hover a row to see how many GPUs are allocated by type, such as a100, h200, or generic GPUs.
• Cluster GPU share: Rows show a progress bar for how much of the currently allocated cluster GPU pool each user is holding.
• Configurable size: Use the Hall of Shame toolbar action or leaderboardTopUserCount setting to choose how many top GPU users to show.
• Manual refresh only — no background polling, so it won't add load to your cluster.
• Last refreshed timestamp: Shows when the Hall of Shame data was fetched so stale data is easy to spot.
• Your row stays visible: Your own Hall of Shame row is highlighted and shown even when you're outside the configured top count.
• Top 3 hogs get shame emojis: 💀 🔥 👹

Cluster Overview

• Account-level GPU Usage: A dedicated sidebar view shows which Slurm accounts are using the most GPUs.
• GPU-only accounting: CPU-only jobs are excluded so the view stays focused on GPU pressure.
• Top users per account: Hover an account row to see the heaviest users under that account.
• GPU type breakdown: Tooltips show how each account's GPU allocation is distributed by GPU type.
• Cluster share bars: Rows show each account's share of currently allocated GPUs as a compact progress bar.
• Last refreshed timestamp: Shows when the Cluster Overview was fetched.

Script Intelligence

• Quick Submit: A ▶ button appears in the editor title bar when viewing any file containing #SBATCH directives. One click to submit — no dialogs.
• Partition Hover Stats: Hover over a partition name in #SBATCH --partition= to see real-time GPU usage, running/pending jobs, and node availability with a visual usage bar.
• Visual Hints: Partition names get a dotted underline to show they're hoverable.

⚙️ Configuration

Configure the extension via VS Code Settings (Cmd+, on macOS / Ctrl+, on Windows/Linux):

Tip: If you monitor many jobs, increasing autoRefreshInterval reduces SLURM command load.

✅ Requirements

This extension must run on a machine with direct access to SLURM commands.
In practice, that means you should install it only on the cluster side (e.g., a login node / head node / SLURM-accessible node — whichever your site provides), not on your local computer.

Required commands:

• squeue
• sinfo
• scontrol
• sacct
• sbatch
• scancel

GPU Partition Usage requires GPU partitions to be exposed through Slurm GRES (sinfo %G). If your cluster tracks GPUs outside GRES, those partitions may not appear in the GPU Partition Usage view.

Important: No Remote Connection (Yet)

At the moment, the extension cannot connect to a remote cluster by itself. It does not SSH into a server, tunnel commands, or forward SLURM calls.

✅ Supported setup: Run VS Code on the SLURM-accessible node (or use VS Code Remote - SSH to open a remote VS Code session on that node) and install the extension on the Remote target.

🚧 Remote connection support is work-in-progress (WIP) and will be added in a future release.

🤝 Contributing

Contributions are welcome — bug fixes, documentation improvements, and feature requests.

• Report issues / request features: https://github.com/dhimitriosduka1/sCode/issues
• Pull requests are welcome!

If you’re opening a PR, please include:

• A short description of the change and why it helps
• Screenshots/GIFs for UI updates (when applicable)

📄 License

MIT — see LICENSE.

If this extension helps you, consider giving it a ⭐ on GitHub!

Made with ❤️ by Dhimitrios Duka with Google Antigravity