fault-injection-loop

Official

Automate closed-loop fault-injection benchmarking.

AuthorNVIDIA
Version1.0.0
Installs0

System Documentation

What problem does it solve?

Closed-loop fault-injection experimentation is automated end-to-end, enabling rapid benchmarking of resilience workflows for large-scale ML training by coordinating a prioritized pool of experiments and submitting them two at a time to bound filesystem load.

Core Features & Use Cases

  • Closed-loop fault pool design with prioritized (fault_type, rank, iter, nodes) entries and automated submission.
  • Deterministic batching: two simultaneous jobs per round with SBATCH and waiting until both finish.
  • Automatic post-run analysis: log-analysis and FR-analysis, then attribution scoring and gap aggregation to guide improvements.

Quick Start

Configure your environment and run the batch workflow with bash scripts/prepare_node_alloc.sh to start submitting fault-injection experiments.

Dependency Matrix

Required Modules

None required

Components

scripts

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: fault-injection-loop
Download link: https://github.com/NVIDIA/nvidia-resiliency-ext/archive/main.zip#fault-injection-loop

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.