verl-project/verl

[Bug] workflow `e2e_prime` is sometimes stuck

Open

#1,344 创建于 2025年4月30日

在 GitHub 查看
 (1 评论) (0 反应) (0 负责人)Python (21,533 star) (3,940 fork)auto 404
bugcall for contributiongood first issue

描述

Motivation

The workflow e2e_prime is sometimes stuck and reaches timeout, while sometimes finishes normally. See https://github.com/volcengine/verl/actions/workflows/e2e_prime.yml for details.

Plan

  • Reproduce the issue locally with the container and commands used in the workflow.
  • Check where the root cause is.

贡献者指南

技术栈
pythonshellgithub actions
领域
ci cdtesting
议题类型
bug
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
3
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
half day
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
fresh
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
Familiarity with GitHub ActionsDocker experiencePython basicsDebugging CI pipelines
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
40
研究方向
First, examine the workflow file at .github/workflows/e2e prime.yml to understand the commands and environment. Then, attempt to replicate the issue locally using the same container image as in the workflow. Analyze the logs of both failed and successful runs (available at https://github.com/volcengine/verl/actions/workflows/e2e prime.yml) to identify where the process hangs. Look for common causes such as resource exhaustion, network timeouts, or race conditions in test setup.