sue-lumi-inode-cleanup

What problem does it solve?

Shared HPC parallel filesystems like LUMI Lustre and Snellius GPFS allocate one inode per file or directory, so large-scale experiment runs, caches, and scattered small files can exhaust file quotas long before block storage runs out. This leads to cryptic "Disk quota exceeded" errors and blocked jobs that are time-consuming to debug manually.

Core Features & Use Cases

• Full inode footprint scanning: Scans all user-accessible storage roots including scratch, home, flash, code checkouts, and experiment output directories to identify the top inode-consuming paths and scattered small-file hotspots like caches, checkpoint shards, and loose per-variant metadata.
• Safe, confirmation-gated cleanup: Classifies all findings by safety level (safe delete, tar/remove, reduce/pack, review) and only proposes destructive actions after explicit user approval, delegating tar+remove operations to the hardened sue-cleanup archive tool to prevent accidental data loss.
• Real-world use case: If a researcher encounters quota errors despite having ample free space on their LUMI allocation, this skill locates the hundreds of thousands of small HuggingFace cache or temporary experiment files consuming their inode quota and guides them through a targeted, low-risk cleanup process.

Quick Start

Use the sue-lumi-inode-cleanup skill to scan your LUMI or Snellius storage for inode pressure and receive a prioritized list of safe cleanup actions with estimated inode savings.