[Bug] The engine hangs after requesting health_generate 190 times.

[unix1986]

Checklist

• 1. I have searched related issues but cannot get the expected help.
• 2. The bug has not been fixed in the latest version.
• 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.
• 4. If the issue you raised is not a bug but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 5. Please use English, otherwise it will be closed.

Describe the bug

When only requesting /health_generate, the 190th request times out because the engine hangs. However, if I use --disable-cuda-graph, it works okay.

Reproduction

use the code for requesting:

#!/bin/bash
i=0
while true
do
	curl -m 60 "http://host:port/health_generate"
	i=$((i+1))
	echo "done $i"
done

Environment

I suppose it's irrelevant to the GPU environment and models we use. The following is the environment I use, for reference only：
OS：ubuntu22.04
GPU: A800
CUDA Driver: 530.30.02
CUDA Runtime: 12.1
Model: Qwen2-72b-instruct-gptq-int4
sglang：latest