Diagnose distributed training hangs in Ray clusters by analyzing call stacks and task states.
npx skills add https://github.com/redai-infra/Relax --skill debug-hang
Please help me install this Agent Skill. Skill: debug-hang Source: https://github.com/redai-infra/Relax/tree/main/skills/debug-hang Command: npx skills add https://github.com/redai-infra/Relax --skill debug-hang
This Skill helps users automatically diagnose and troubleshoot issues causing training tasks to hang in distributed Ray environments, reducing downtime and manual effort.
Run the provided diagnosis script with your Ray cluster address to automatically gather and analyze cluster diagnostics.
High-intent search queries and answers about installing and using this skill.
To diagnose a distributed training hang, you can run the diagnosis script with your Ray cluster address to collect task states, monitor resource usage, and analyze blocking call stacks.
Distributed training hangs are caused by deadlocks, resource contention, resource starvation, or misconfigurations, requiring cluster health monitoring and stack trace analysis to pinpoint.
Checking Ray cluster health involves monitoring resource assignments and listing running tasks and actors to identify potential bottlenecks causing training jobs to stall.
Yes, collecting call stacks gathers thread states and blocking code paths from processes, enabling root cause analysis to identify deadlocks and resource starvation efficiently.
No dependencies are required, but you need a running Ray cluster address to execute the provided diagnosis script for automated detection of distributed training issues.
Troubleshooting resource contention involves inspecting task and actor states, analyzing dependency chains, and monitoring cluster health to resolve misconfigurations causing job stalls.
EnkeZDiagnose Ray distributed training hang issues by analyzing cluster status and blocking patterns.
DeepExperienceDiagnose stalled Ray distributed training tasks by tracing blocked execution paths.
ovachieverOrchestrate distributed training across clusters with Ray Train and Ray Tune.
kevinnftDrive Node.js debugging via CDP to set breakpoints and inspect runtime state.
