[Bug] When using sgl-router, canceled requests keep running on workers.

[hjc-puro]

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

I started sglang router with one worker under it, sent a /chat/completions request, canceled it before response, and it continued running on the worker (checked via nvidia-smi and sglang logs). In fact I believe it was stuck running - didn't finish after 5 minutes.

Reproduction

python -m sglang.launch_server \\
    --model-path ${MODEL_PATH} \\
    --port ${WORKER_PORT} \\
    --host 0.0.0.0 \\
    --tp ${TP_SIZE} \\
    --dist-init-addr ${HEAD_NODE_IP}:${DIST_INIT_PORT} \\
    --nnodes ${NNODES} \\
    --node-rank \$RANK \\
    --log-level info \\
    --trust-remote-code \\
    --grammar-backend xgrammar
python -m sglang_router.launch_router --worker-urls ${LOCAL_RANK0_URL} --port ${ROUTER_PORT} --host 0.0.0.0

Environment

Version: 0.4.4.post1