sltop 0.1.0

nvtop-inspired interactive SLURM cluster dashboard

sltop — SLURM Cluster Top

An nvtop-inspired interactive SLURM cluster dashboard.
Monitor partitions, scheduling rules, the full job queue, and your own running/pending jobs — all from a single, keyboard-driven terminal window powered by Textual.

---

Table of Contents

• Features
• Requirements
• Installation
  • Via pip (recommended)
  • Via pipx (isolated)
  • Manual install
• Usage
  • Arguments
  • Key Bindings
• Dashboard Tabs
  • ① Resources
  • ② Rules
  • ③ Queue
  • ④ My Jobs
• How It Works
• Troubleshooting
• Contributing
• License

---

Features

• 📊 Resources tab — per-partition CPU/GPU/node utilisation bars with alloc/mix/idle/drain breakdown and cluster-wide totals
• 📋 Rules tab — scheduling constraints (MaxTime, QoS, GPU limits, node limits, TRES) rendered as Rich panels
• 📜 Queue tab — full SLURM job queue with sortable columns (click any header); PENDING config-errors highlighted as INCORRECT_CONFIG in red with a plain-English explanation
• 👤 My Jobs tab — cards for the current user's jobs showing elapsed/limit time, resource requests, GPU mini-bar, and a human-readable translation of every SLURM reason code
• 🔄 Auto-refresh — all tabs update on a configurable interval (default 10 s) with last-refresh timestamp in the subtitle bar
• 🌈 Rich colour UI — explicit RGB colours via Textual + Rich; stacked node-state bars, colour-coded utilisation bars, per-partition colour coding

---

Requirements

Requirement	Notes
Python ≥ 3.8	Pure Python — no Bash or SSH required
SLURM (squeue, scontrol, sinfo, sacctmgr)	Must be available on the login node
textual ≥ 0.50	Installed automatically as a dependency

---

Installation

Via pip (recommended)

pip install sltop

This places the sltop command on your PATH.

Via pipx (isolated)

pipx installs the tool into an isolated environment and exposes the command globally — ideal for shared HPC login nodes.

pipx install sltop

Manual install

# Clone
git clone https://github.com/whats2000/sltop.git
cd sltop

# Install in editable mode
pip install -e .

---

Usage

sltop [-n SECS] [-p P1,P2] [-u USER]

Simply run sltop from any terminal on your HPC login node:

sltop                      # default 10-second refresh, all partitions, all users
sltop -n 5                 # refresh every 5 seconds
sltop -p gpu,cpu           # filter to specific partitions
sltop -u $USER             # show only your jobs in the Queue tab

Arguments

Argument	Default	Description
-n / --interval	10	Refresh interval in seconds
-p / --partitions	all	Comma-separated partition filter
-u / --user	all	Show only jobs for this user in the Queue tab

Key Bindings

Key	Action
Tab	Next tab
1	Jump to ① Resources tab
2	Jump to ② Rules tab
3	Jump to ③ Queue tab
4	Jump to ④ My Jobs tab
r	Force refresh now
Esc	Focus the Queue table
q	Quit

---

Dashboard Tabs

① Resources

Cluster-wide summary panel (total CPU / GPU / node utilisation) followed by one Rich card per partition showing:

• Availability — UP / DOWN indicator
• MaxTime & QoS — scheduler policy labels
• GRES & per-node memory — hardware totals
• Constraints — min/max GPU and node limits (with implied-node inference)
• CPU / GPU / Node bars — colour-coded stacked bars (alloc / mix / idle / drain)

② Rules

One Rich card per partition with the full set of scontrol show partition fields plus QoS GPU limits from sacctmgr, including:

• Time limits (Max / Default)
• Node and CPU constraints
• GPU totals and per-node limits
• Access lists (AllowGroups / AllowAccounts)
• TRES breakdown

③ Queue

Full squeue output as a sortable DataTable. Click any column header to sort ascending; click again to reverse; third click clears the sort.
PENDING jobs whose reason code indicates a configuration error (e.g. QOSMinGRES, InvalidAccount) are flagged as INCORRECT_CONFIG in red with a human-readable explanation appended.

④ My Jobs

Rich Panel cards for every job belonging to the current Unix user, showing:

• Job state with colour and symbol
• Job ID, partition (colour-coded), user
• Elapsed time / time limit
• Node count and GRES request
• GPU mini-bar (request vs partition total)
• Plain-English translation of the SLURM reason code

---

How It Works

Login Node
┌──────────────────────────────────────────────┐
│ sltop                                        │
│  ├─ sinfo    ──► Resources / Rules tabs      │
│  ├─ scontrol ──► Rules tab                   │
│  ├─ sacctmgr ──► QoS GPU limits              │
│  ├─ squeue   ──► Queue / My Jobs tabs        │
│  └─ Textual TUI render loop                  │
└──────────────────────────────────────────────┘

On mount, sltop fires a single _do_refresh() pass and schedules it to repeat every --interval seconds. Each pass calls the four SLURM CLI tools, builds Rich renderables, and pushes them into the Textual widget tree — no background threads or SSH connections required.

---

Troubleshooting

No partition data.

sinfo returned no output. Make sure SLURM is available on the current node (which sinfo).

Rules or QoS data is missing

scontrol or sacctmgr may not be available, or you may lack the permissions to query QoS data. sltop silently omits unavailable data rather than crashing.

Queue shows no jobs

There are currently no jobs matching the optional --user / --partitions filter. Run without filters to see all jobs.

textual not found

Install it manually: pip install "textual>=0.50".

---

Contributing

Contributions, bug reports and feature requests are welcome!

1. Fork the repository
2. Create a feature branch: git checkout -b feat/my-feature
3. Commit your changes with a descriptive message
4. Open a Pull Request

---

License

This project is licensed under the MIT License — see the LICENSE file for details.