Hitesh Sahu
Hitesh SahuHitesh Sahu
  1. Home
  2. ›
  3. work
  4. ›
  5. …

  6. ›
  7. 6 squint

Loading ⏳
Fetching content, this won’t take long…


💡 Did you know?

🦈 Sharks existed before trees 🌳.

🍪 This website uses cookies

No personal data is stored on our servers however third party tools Google Analytics cookies to measure traffic and improve your website experience. Learn more

Cloud-DevOps

    AI & Machine Learning

    Cloud & DevOps
    • Betting Platform

    • Connected Vehicles

    • Payment Communication Platform

    • GhostFleet — Kubernetes GPU Cluster Simulator

    • GPU-Lens — Drop-in GPU Observability for Kubernetes & SLURM

    • Squint — GPU-Aware Terminal UI for SLURM Clusters


    Full-Stack Applications

    Mobile Development

Cover Image for Squint — GPU-Aware Terminal UI for SLURM Clusters
Fork me on GitHub
Cloud & DevOps

Squint — GPU-Aware Terminal UI for SLURM Clusters

Open Source

2026

Creator & Maintainer

Tech Stack
Go
Bubble Tea
Lip Gloss
SLURM
NVIDIA-SMI
GPU Monitoring
CLI Tooling
Terminal UI

Summary

A real-time terminal dashboard for SLURM clusters — GPU utilisation heatmap across all nodes, plain-English job status, and zero-configuration read-only access via squeue, scontrol, and nvidia-smi.


What I Built

Project Overview

squeue output is verbose and hard to scan. Squint transforms it into a real-time terminal dashboard — a GPU utilisation heatmap across all cluster nodes, plain-English translations of cryptic SLURM reason codes, and at-a-glance job status without touching cluster state.

Squint only executes query commands (squeue, scontrol, nvidia-smi) — it cannot modify the cluster. Safe to run anywhere, no configuration required.

Pairs with Caravan — use Caravan to spin up a local SLURM cluster, Squint to monitor it.


Key Features

GPU Heatmap

Colour-coded visualisation of GPU utilisation across every node — instantly identifies underutilised GPUs, hot spots, and idle reservations without running nvidia-smi per node manually.

Plain-English Job Status

Translates SLURM's cryptic reason codes into actionable explanations:

SLURM reason Squint translation
Resources Waiting for enough GPUs to become free
Priority Higher-priority jobs ahead in queue
ReqNodeNotAvail Requested node is down or drained
AssocGrpGRESMinutes GPU-minute quota exhausted for your account

Mock Data Mode

Fully functional demo mode — no cluster access needed. Run squint --mock to explore all views with synthetic data, useful for demos and testing.

Read-Only by Design

Executes only squeue, scontrol, and nvidia-smi. No write paths, no risk of accidental job cancellation or node drain.


Why It Matters

Raw SLURM CLI Squint
squeue — wall of text Real-time table with colour-coded state
nvidia-smi per node — SSH loop GPU heatmap across all nodes at once
Reason codes require documentation lookup Plain-English explanation inline
No overview of cluster GPU waste Idle-reservation waste visible immediately

Technology Stack

Language: Go 1.22+

UI Framework: Bubble Tea (Charm)

Styling: Lip Gloss (Charm)

Cluster Integration: SLURM (squeue, scontrol), NVIDIA-SMI

Platforms: Linux, macOS, Windows (WSL)

← Previous

GPU-Lens — Drop-in GPU Observability for Kubernetes & SLURM

Next →

Flight Booking Platform

Let's work together
+49 176-2019-2523
hiteshkrsahu@gmail.com
WhatsApp
Skype
Munich 🥨, Germany 🇩🇪, EU
Playstore
Hitesh Sahu's apps on Google Play Store
Need Help?
Let's Connect
Navigation
  Home/About
  Skills
  Work/Projects
  Lab/Experiments
  Contribution
  Awards
  Art/Sketches
  Thoughts
  Contact
Links
  Sitemap
  Legal Notice
  Privacy Policy

Made with

NextJS logo

NextJS by

hitesh Sahu

| © 2026 All rights reserved.