已关闭
[Bug-Report|缺陷反馈]: A3单边通信CQ overflow #822
hu-yiliang11创建于 28 天前关闭于 27 天前
Leewis
28 天前 评论：
28 天前 评论：
/assign @gcw_NcEfY7mt
28 天前 将 gcw_NcEfY7mt 设为负责人
Leewis
27 天前 评论：
27 天前 评论：
针对上述问题，对应PR已经合入，相关链接：https://gitcode.com/cann/hcomm/pull/4953
27 天前 添加了label：resolved
Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.
Describe the current behavior / 问题描述 (Mandatory / 必填)
使用CANN9.1.0版本，单边通信的时候CQ overflow，报错流同步超时
Environment / 环境信息 (Mandatory / 必填)
A3
海思内部版本：CANN 9.1.0
Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)
deeps v4 flash+vllm-ascend 0.26.0版本 + pd分离+hixl
Describe the expected behavior / 预期结果 (Mandatory / 必填)
正常运行，CQ不会被打爆
Related log / screenshot / 日志 / 截图 (Mandatory / 必填)
[ERROR] RUNTIME(188164,):2026-08-25-13:13:47.019.026 [stream.cc:1878]195997 SynchronizeExecutedTask:Stream synchronize timeout, device_id=2, stream_id=216, timeout=100000ms.
[ERROR] RUNTIME(188164,):2026-08-25-13:13:47.019.062 [stream.cc:1944]195997 SynchronizeImpl:failed, stream_id=216, error=0x7030010
[ERROR] RUNTIME(188164,):2026-08-25-13:13:47.023.593 [api_impl.cc:1664]195997 StreamSynchronize:Stream synchronize timeout, the current task is type_name=NOTIFY_WAIT, device_id=2, stream_id=216, task_id=679, task_type=14
[ERROR] RUNTIME(188164,):2026-08-25-13:13:47.023.628 [api_error.cc:1101]195997 StreamSynchronize:Stream synchronize failed, stream_id=216, timeout=100000ms.
[ERROR] RUNTIME(188164,):2026-08-25-13:13:47.023.641 [api_c_stream.cc:180]195997 rtStreamSynchronizeWithTimeout:ErrCode=507046, desc=[stream sync timeout], InnerCode=0x7030010
[ERROR] RUNTIME(188164,):2026-08-25-13:13:47.023.643 [error_message_manage.cc:66]195997 FuncErrorReason:rtStreamSynchronizeWithTimeout execution failed.
[ERROR] GE(188164,):2026-08-25-13:13:47.036.220 [comm_channel.cc:516][HIXL] 195997 operator(): ErrorNo: 103901 Call aclrt api failed, ret: 0x7BCA6
[ERROR] GE(188164,):2026-08-25-13:13:47.053.759 [comm_channel.cc:533][HIXL] 195997 TransferSync: ErrorNo: 103901 TransferSync failed.
[ERROR] GE(188164,):2026-08-25-13:13:47.053.764 [adxl_inner_engine.cc:506][HIXL] 195997 TransferSync: ErrorNo: 103901 Failed to transfer sync,
黑匣子日志：
exceptionId 0xa4302003
dev 0x0
arg 0x0
core 0x18
eType 0x3a
module ABNM_CQ_CQE_Err
desc Get Abnormal CQ Error.
[ERROR] KERNEL(13095,sklogd):2026-08-25-21:12:07.010.139 [klogd.c:247][3981956.323356] hns3 0002:79:00.0 hns_0: local work queue 0x172 catast error, sub_event type is: 244 SUBSYSTEM=pci DEVICE=+pci:0002:79:00.0
network info: RoCE Get Abnormal CQ Error
Special notes for this issue/备注 (Optional / 选填)