0.7.4 开发版CPU使用率一直100%，即使没处理请求的时候也一样

[AndrewTsao]

请问V1模式下，使用FLASHMLA开启状态，有8个进程CPU100%正常吗？

程序版本:

Package                           Version
--------------------------------- ----------------------
aiohappyeyeballs                  2.6.1
aiohttp                           3.11.13
aiosignal                         1.3.2
airportsdata                      20250224
annotated-types                   0.7.0
anyio                             4.8.0
astor                             0.8.1
attrs                             25.2.0
blake3                            1.0.4
certifi                           2025.1.31
charset-normalizer                3.4.1
click                             8.1.8
cloudpickle                       3.1.1
compressed-tensors                0.9.2
cupy-cuda12x                      13.4.0
depyf                             0.18.0
dill                              0.3.9
diskcache                         5.6.3
distro                            1.9.0
dnspython                         2.7.0
einops                            0.8.1
email_validator                   2.2.0
fastapi                           0.115.11
fastapi-cli                       0.0.7
fastrlock                         0.8.3
filelock                          3.17.0
flashinfer-python                 0.2.2+cu124torch2.5
frozenlist                        1.5.0
fsspec                            2025.3.0
gguf                              0.10.0
h11                               0.14.0
httpcore                          1.0.7
httptools                         0.6.4
httpx                             0.28.1
huggingface-hub                   0.29.3
idna                              3.10
importlib_metadata                8.6.1
iniconfig                         2.0.0
interegular                       0.3.3
Jinja2                            3.1.6
jiter                             0.9.0
jsonschema                        4.23.0
jsonschema-specifications         2024.10.1
lark                              1.2.2
llvmlite                          0.43.0
lm-format-enforcer                0.10.11
markdown-it-py                    3.0.0
MarkupSafe                        3.0.2
mdurl                             0.1.2
mistral_common                    1.5.3
modelscope                        1.23.2
mpmath                            1.3.0
msgpack                           1.1.0
msgspec                           0.19.0
multidict                         6.1.0
nest-asyncio                      1.6.0
networkx                          3.4.2
numba                             0.60.0
numpy                             1.26.4
nvidia-cublas-cu12                12.4.5.8
nvidia-cuda-cupti-cu12            12.4.127
nvidia-cuda-nvrtc-cu12            12.4.127
nvidia-cuda-runtime-cu12          12.4.127
nvidia-cudnn-cu12                 9.1.0.70
nvidia-cufft-cu12                 11.2.1.3
nvidia-curand-cu12                10.3.5.147
nvidia-cusolver-cu12              11.6.1.9
nvidia-cusparse-cu12              12.3.1.170
nvidia-cusparselt-cu12            0.6.2
nvidia-nccl-cu12                  2.21.5
nvidia-nvjitlink-cu12             12.4.127
nvidia-nvtx-cu12                  12.4.127
openai                            1.66.3
opencv-python-headless            4.11.0.86
outlines                          0.1.11
outlines_core                     0.1.26
packaging                         24.2
partial-json-parser               0.2.1.1.post5
pillow                            11.1.0
pip                               25.0.1
pluggy                            1.5.0
prometheus_client                 0.21.1
prometheus-fastapi-instrumentator 7.0.2
propcache                         0.3.0
protobuf                          6.30.0
psutil                            7.0.0
py-cpuinfo                        9.0.0
pybind11                          2.13.6
pycountry                         24.6.1
pydantic                          2.10.6
pydantic_core                     2.27.2
Pygments                          2.19.1
pytest                            8.3.5
python-dotenv                     1.0.1
python-json-logger                3.3.0
python-multipart                  0.0.20
PyYAML                            6.0.2
pyzmq                             26.3.0
ray                               2.43.0
referencing                       0.36.2
regex                             2024.11.6
requests                          2.32.3
rich                              13.9.4
rich-toolkit                      0.13.2
rpds-py                           0.23.1
safetensors                       0.5.3
scipy                             1.15.2
sentencepiece                     0.2.0
setuptools                        75.8.2
shellingham                       1.5.4
sniffio                           1.3.1
starlette                         0.46.1
sympy                             1.13.1
tiktoken                          0.9.0
tokenizers                        0.21.0
torch                             2.5.1+cu124
torchaudio                        2.5.1+cu124
torchvision                       0.20.1+cu124
tqdm                              4.67.1
transformers                      4.49.0
triton                            3.1.0
typer                             0.15.2
typing_extensions                 4.12.2
urllib3                           2.3.0
uvicorn                           0.34.0
uvloop                            0.21.0
vllm                              0.7.4.dev413+gf53a0586
watchfiles                        1.0.4
websockets                        15.0.1
wheel                             0.45.1
xformers                          0.0.28.post3
xgrammar                          0.1.15
yarl                              1.18.3
zipp                              3.21.0

启动命令

VLLM_ATTENTION_BACKEND=FLASHMLA
VLLM_USE_V1=1
OMP_NUM_THREADS=12

  /opt/vllm-0.7.4-dev/bin/vllm serve DeepSeek-R1
      --max-model-len 131072
      --max-num-batched-tokens 8192
      --enable-reasoning --reasoning-parser deepseek_r1
      --api_key ${VLLM_API_KEY}
      --tensor-parallel-size 8
      --trust-remote-code
      --disable-log-requests
      --enable-prefix-caching
      --enable-chunked-prefill
      --gpu_memory_utilization=0.95
      -O3

硬件配置： 8 x H200

strace结果，

#0  0x00007f11a6f5fc9b in sched_yield () from target:/usr/lib/x86_64-linux-gnu/libc.so.6
#1  0x000056525e213103 in os_sched_yield_impl (module=<optimized out>) at /usr/local/src/conda/python-3.11.11/Modules/posixmodule.c:7198
#2  os_sched_yield (module=<optimized out>, _unused_ignored=<optimized out>) at /usr/local/src/conda/python-3.11.11/Modules/clinic/posixmodule.c.h:3017
#3  0x000056525e2483fe in cfunction_vectorcall_NOARGS (func=0x7f11a6daeca0, args=<optimized out>, nargsf=<optimized out>, kwnames=<optimized out>) at /usr/local/src/conda/python-3.11.11/Include/cpython/methodobject.h:52
#4  0x000056525e25d5ec in _PyObject_VectorcallTstate (kwnames=<optimized out>, nargsf=<optimized out>, args=<optimized out>, callable=0x7f11a6daeca0, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_call.h:92
#5  PyObject_Vectorcall (callable=0x7f11a6daeca0, args=<optimized out>, nargsf=<optimized out>, kwnames=<optimized out>) at /usr/local/src/conda/python-3.11.11/Objects/call.c:299
#6  0x000056525e250d19 in _PyEval_EvalFrameDefault (tstate=tstate@entry=0x56525e5dd9f8 <_PyRuntime+166328>, frame=<optimized out>, frame@entry=0x7f0ca5ce7e70, throwflag=throwflag@entry=0) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:4769
#7  0x000056525e2dfeb7 in _PyEval_EvalFrame (throwflag=0, frame=0x7f0ca5ce7e70, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_ceval.h:73
#8  gen_send_ex2 (closing=0, exc=0, presult=<synthetic pointer>, arg=0x0, gen=0x7f0ca5ce7e20) at /usr/local/src/conda/python-3.11.11/Objects/genobject.c:219
#9  gen_iternext (gen=0x7f0ca5ce7e20) at /usr/local/src/conda/python-3.11.11/Objects/genobject.c:594
#10 builtin_next (self=<optimized out>, args=0x7f11a6c97568, nargs=1) at /usr/local/src/conda/python-3.11.11/Python/bltinmodule.c:1477
#11 0x000056525e251b37 in _PyEval_EvalFrameDefault (tstate=tstate@entry=0x56525e5dd9f8 <_PyRuntime+166328>, frame=<optimized out>, frame@entry=0x7f11a6c97508, throwflag=throwflag@entry=0) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:5050
#12 0x000056525e294602 in _PyEval_EvalFrame (throwflag=0, frame=0x7f11a6c97508, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_ceval.h:73
#13 _PyEval_Vector (kwnames=<optimized out>, argcount=<optimized out>, args=0x7ffdb180bfe0, locals=0x0, func=0x7f11a6b16fc0, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:6434
#14 _PyFunction_Vectorcall (kwnames=<optimized out>, nargsf=<optimized out>, stack=0x7ffdb180bfe0, func=0x7f11a6b16fc0) at /usr/local/src/conda/python-3.11.11/Objects/call.c:393
#15 _PyObject_VectorcallTstate (tstate=0x56525e5dd9f8 <_PyRuntime+166328>, callable=0x7f11a6b16fc0, args=0x7ffdb180bfe0, nargsf=<optimized out>, kwnames=<optimized out>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_call.h:92
#16 0x000056525e2551fc in method_vectorcall (kwnames=0x0, nargsf=0, args=0x0, method=0x7f0c95006a80) at /usr/local/src/conda/python-3.11.11/Objects/classobject.c:67
#17 _PyObject_VectorcallTstate (args=0x0, nargsf=0, kwnames=0x0, callable=0x7f0c95006a80, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_call.h:92
#18 _PyObject_CallNoArgs (func=0x7f0c95006a80) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_call.h:107
#19 _PyEval_EvalFrameDefault (tstate=tstate@entry=0x56525e5dd9f8 <_PyRuntime+166328>, frame=<optimized out>, frame@entry=0x7f11a6c97320, throwflag=throwflag@entry=0) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:4422
#20 0x000056525e27577f in _PyEval_EvalFrame (throwflag=0, frame=0x7f11a6c97320, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_ceval.h:73
#21 _PyEval_Vector (kwnames=<optimized out>, argcount=0, args=0x7f11a6c49138, locals=0x0, func=0x7f1075a0b560, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:6434
#22 _PyFunction_Vectorcall (func=0x7f1075a0b560, stack=0x7f11a6c49138, nargsf=<optimized out>, kwnames=<optimized out>) at /usr/local/src/conda/python-3.11.11/Objects/call.c:393
#23 0x000056525e27f574 in _PyVectorcall_Call (kwargs=<optimized out>, tuple=<optimized out>, callable=0x7f1075a0b560, func=0x56525e275600 <_PyFunction_Vectorcall>, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Objects/call.c:257
#24 _PyObject_Call (kwargs=<optimized out>, args=<optimized out>, callable=0x7f1075a0b560, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Objects/call.c:328
#25 PyObject_Call (callable=0x7f1075a0b560, args=<optimized out>, kwargs=<optimized out>) at /usr/local/src/conda/python-3.11.11/Objects/call.c:355
#26 0x000056525e25500a in do_call_core (use_tracing=<optimized out>, kwdict=0x7f11a6c33b80, callargs=0x56525e5c3658 <_PyRuntime+58904>, func=0x7f1075a0b560, tstate=<optimized out>) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:7349
#27 _PyEval_EvalFrameDefault (tstate=tstate@entry=0x56525e5dd9f8 <_PyRuntime+166328>, frame=<optimized out>, frame@entry=0x7f11a6c97020, throwflag=throwflag@entry=0) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:5376
#28 0x000056525e307c5d in _PyEval_EvalFrame (throwflag=0, frame=0x7f11a6c97020, tstate=0x56525e5dd9f8 <_PyRuntime+166328>) at /usr/local/src/conda/python-3.11.11/Include/internal/pycore_ceval.h:73
#29 _PyEval_Vector (tstate=tstate@entry=0x56525e5dd9f8 <_PyRuntime+166328>, func=func@entry=0x7f11a6dcdf80, locals=locals@entry=0x7f11a6df6b40, args=args@entry=0x0, argcount=argcount@entry=0, kwnames=kwnames@entry=0x0) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:6434
#30 0x000056525e30739f in PyEval_EvalCode (co=co@entry=0x7f11a6d80e40, globals=globals@entry=0x7f11a6df6b40, locals=locals@entry=0x7f11a6df6b40) at /usr/local/src/conda/python-3.11.11/Python/ceval.c:1148
#31 0x000056525e32530a in run_eval_code_obj (tstate=tstate@entry=0x56525e5dd9f8 <_PyRuntime+166328>, co=co@entry=0x7f11a6d80e40, globals=globals@entry=0x7f11a6df6b40, locals=locals@entry=0x7f11a6df6b40) at /usr/local/src/conda/python-3.11.11/Python/pythonrun.c:1741
#32 0x000056525e320f93 in run_mod (mod=mod@entry=0x565260144508, filename=filename@entry=0x56525e5ba550 <_PyRuntime+21776>, globals=globals@entry=0x7f11a6df6b40, locals=locals@entry=0x7f11a6df6b40, flags=flags@entry=0x7ffdb180c488, arena=arena@entry=0x7f11a6d1b650) at /usr/local/src/conda/python-3.11.11/Python/pythonrun.c:1762
#33 0x000056525e3159a2 in PyRun_StringFlags (str=<optimized out>, start=<optimized out>, globals=0x7f11a6df6b40, locals=0x7f11a6df6b40, flags=0x7ffdb180c488) at /usr/local/src/conda/python-3.11.11/Python/pythonrun.c:1632
#34 0x000056525e31575c in PyRun_SimpleStringFlags (command=0x7f11a6d962d0 "from multiprocessing.spawn import spawn_main; spawn_main(tracker_fd=53, pipe_handle=22)\n", flags=flags@entry=0x7ffdb180c488) at /usr/local/src/conda/python-3.11.11/Python/pythonrun.c:487
#35 0x000056525e3300ec in pymain_run_command (command=<optimized out>) at /usr/local/src/conda/python-3.11.11/Modules/main.c:255
#36 pymain_run_python (exitcode=0x7ffdb180c480) at /usr/local/src/conda/python-3.11.11/Modules/main.c:596
#37 Py_RunMain () at /usr/local/src/conda/python-3.11.11/Modules/main.c:684
#38 0x000056525e2f7617 in Py_BytesMain (argc=<optimized out>, argv=<optimized out>) at /usr/local/src/conda/python-3.11.11/Modules/main.c:738
#39 0x00007f11a6e80d90 in ?? () from target:/usr/lib/x86_64-linux-gnu/libc.so.6
#40 0x00007f11a6e80e40 in __libc_start_main () from target:/usr/lib/x86_64-linux-gnu/libc.so.6
#41 0x000056525e2f74ca in _start ()