speek 0.0.3.dev1

Peek into slurm's resource info such as GPU avaiability, usage per user, job status, and more.

🔍 speek

speek is a suite of SLURM cluster monitoring tools — from quick one-shot snapshots to a full interactive TUI.

Installation

pip install speek

For the latest development version:

pip install --pre speek

Commands

Command	Description
speek0	Classic one-shot cluster overview — GPU availability, per-user usage, job status
speek-	Compact snapshot — per-model GPU bars, trends, pending pressure
speek+	Full interactive TUI — queue, nodes, users, stats, events, shell

speek0 — Classic Overview

speek0 [-u USER] [-f FILE] [-t T_AVAIL]

Option	Description
-u USER	Highlight a specific user (default: self)
-f FILE	User info CSV file
-t T_AVAIL	Time window for upcoming release, e.g. 5 m, 1 h

Shows a table of GPU usage per partition, ranked users with 🥇🥈🥉, utilization-colored counts, and your current jobs.

speek- — Compact Snapshot

speek- [-u USER]

Per-GPU-model view with utilization bars, free/total counts, pending pressure (⏸N), availability trends (↑↓), and your running/pending jobs. Detects down nodes and shows them as DEAD.

speek+ — Interactive TUI

speek+

Full-featured Textual TUI with:

• Cluster — speek0-style usage table (tab 1)
• Queue — all cluster jobs grouped by partition, foldable
• Nodes — per-partition node status with usage bars
• Users — per-user GPU usage, fairshare, per-partition breakdown
• Stats — GPU usage charts, per-user stacked view, issue dashboard
• Logs — session CLI output (not persisted)
• Settings — theme, refresh rates, cache management, log scanning
• Info — cluster probe results, scheduling factors, error detection rules
• Help — keybindings reference

Features

• 70+ color themes (base16 standard)
• OOM and error detection (11 error types) with log scanning
• Job detail popup with stdout, stderr, GPU stats, analysis
• Built-in shell with tab completion, history, sbatch suggestions
• Per-job log hints in the table
• Event notifications with read/unread tracking
• Down node detection with DEAD indicators

Requirements

• Python 3.8+
• SLURM cluster with squeue, scontrol, sinfo
• Optional: sacct, sprio, sshare, sreport, scancel for full features
• rich (all commands), textual>=0.50.0 (speek+ only)