[Bug]: high cpu utilization when there is no inference task

[liangjs]

Your current environment

I deploy models with the vllm/vllm-openai:v0.9.0.1 docker image.
The command line arguments: --tensor-parallel-size 2 and --enforce-eager.

🐛 Describe the bug

The CPU utlization is high even if no tasks are running.

This does not depend on models. I encountered the issue with both qwen3 and devstral models.

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.

[chaunceyjiang]

It has already been fixed, refer to #16226.

[njhill]

Resolved by #16226. @liangjs please build from main or use the next release, and set VLLM_SLEEP_WHEN_IDLE=1.