vllm-project/vllm

[Feature]: Priority scheduling supports preemption of requests in the running queue by requests in the waiting queue

Open

#40,004 创建于 2026年4月16日

在 GitHub 查看
 (5 评论) (0 反应) (0 负责人)Python (80,034 star) (16,816 fork)batch import
feature requesthelp wanted

描述

🚀 The feature, motivation and pitch

The current priority scheduling only supports evicting low-priority requests from the running queue when resources are insufficient. However, when scheduling the waiting queue, if pending requests cannot be scheduled (for example, when the number of requests in the running queue has reached max_num_seqs), even high-priority requests cannot preempt requests in the running queue.

Alternatives

No response

Additional context

No response

Before submitting a new issue...

  • Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

贡献者指南

技术栈
python
领域
backendperformanceai
议题类型
feature
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。
4
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。
1-2 days
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。
active
清晰度议题是否清楚说明期望改动、验收标准和下一步。
mostly clear
前置要求
Understanding of vLLM schedulerPython programmingConcurrency concepts
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。
25
研究方向
The issue involves modifying the priority scheduling logic in the vLLM scheduler. Key files to examine include vllm/v1/core/scheduler.py and vllm/v1/core/schedule.py. The current implementation only preempts low priority requests from the running queue when resources are insufficient. To allow high priority waiting requests to preempt running requests, you'll need to add logic to check priority when scheduling the waiting queue and potentially evict lower priority running requests. Review the existing preemption mechanism and the scheduling loop. Start by understanding the data structures for request queues and priority handling.
[Feature]: Priority scheduling supports preemption of requests in the running queue by requests in the waiting queue · vllm-project/vllm#40004 | Good First Issue