debug-hang

What problem does it solve?

This Skill helps users automatically diagnose and troubleshoot issues causing training tasks to hang in distributed Ray environments, reducing downtime and manual effort.

Core Features & Use Cases

• Cluster Status Monitoring: Checks the health and resource usage of Ray clusters to identify potential bottlenecks.
• Task and Actor Inspection: Lists running tasks and actors, providing insights into their states and resource assignments.
• Stack Trace Collection: Gathers call stacks from processes to pinpoint blocking code paths and thread states.
• Root Cause Analysis: Analyzes dependency chains and common hang patterns to identify root causes like deadlocks, resource starvation, or misconfigurations.
• Use Case: When a training job stalls unexpectedly, users can run this Skill to gather comprehensive diagnostics and pinpoint the root of the issue.

Quick Start

Run the provided diagnosis script with your Ray cluster address to automatically gather and analyze cluster diagnostics.