skills.rest
DeepExperience
DeepExperienceOFFICIAL
Explore all skills by this author →
GitHub

debug-hang

Diagnose stalled Ray distributed training tasks by tracing blocked execution paths.

71|Updated May 9, 2026
Search Tags:#debugging#reinforcement learning#ray#distributed training#py-spy#hang diagnosis#actor analysis
One-click install
npx skills add https://github.com/DeepExperience/HyperEyes --skill debug-hang-deepexperience
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: debug-hang
Source: https://github.com/DeepExperience/HyperEyes/tree/main/RL/relax-hypereyes/Relax/skills/debug-hang
Command: npx skills add https://github.com/DeepExperience/HyperEyes --skill debug-hang-deepexperience

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps engineers diagnose stalled Ray distributed training jobs by tracing resource issues, blocked tasks, actor dependencies, and synchronization failures that prevent progress.

Core Features & Use Cases

  • Cluster Diagnostics: Collect Ray cluster status, running jobs, task details, and actor states to identify abnormal resource usage and execution bottlenecks.
  • Blocking Analysis: Gather process stacks with py-spy, trace dependency chains, and classify common hang patterns such as data waits, distributed synchronization, resource contention, and remote call blocking.
  • Use Case: When a large-scale reinforcement learning training run stops making progress with low GPU utilization, use this Skill to collect diagnostics and determine whether the issue comes from data pipelines, Ray actors, configuration mismatches, or deadlock conditions.

Quick Start

Use the debug-hang skill to investigate why my Ray distributed training job is hanging and generate a root cause diagnosis report.

Frequently Asked Questions about debug-hang

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I diagnose a stalled Ray distributed training job?▼

To diagnose a stalled Ray distributed training job, collect cluster state, task metadata, actor information, and process stack traces using py-spy to trace blocked execution paths and identify hang root causes.

Why does my Ray reinforcement learning training hang with low GPU utilization?▼

Ray reinforcement learning training hangs with low GPU utilization due to data pipeline waits, actor coordination issues, resource contention, configuration mismatches, or distributed synchronization deadlocks preventing progress.

How do I trace blocked execution paths in distributed training?▼

Trace blocked execution paths in distributed training by gathering process stacks with py-spy, tracing actor dependency chains, and classifying common hang patterns like remote call blocking and synchronization failures.

Can I use py-spy to analyze Ray actor dependencies and deadlocks?▼

Yes, you can use py-spy to gather process stack traces that analyze Ray actor dependencies, identify abnormal resource usage, and classify synchronization failures causing deadlock conditions in your cluster.

What is the best way to troubleshoot Ray cluster resource bottlenecks?▼

The best way to troubleshoot Ray cluster resource bottlenecks is to collect cluster status, running jobs, task details, and actor states to identify abnormal resource usage and execution bottlenecks.

Related Skills

Explore More Software Engineering Skills→
EnkeZEnkeZ
1

debug-hang

Diagnose Ray distributed training hang issues by analyzing cluster status and blocking patterns.

Community
Advanced
TumeloRamaphosaTumeloRamaphosa

systematic-debugging

Enforce root-cause investigation before proposing fixes in debugging workflows.

Community
Advanced
tuliopc23tuliopc23

axiom-hang-diagnostics

Diagnose application hangs by identifying main thread blocking with profiling tools.

Community
Advanced
arsityarsity

torchforge-rl-training

Orchestrate PyTorch-native RL experiments with distributed training and weight sync.

Community
Advanced