fastdeploy 部署paddleocr，压力测试异常，进程退出

温馨提示：根据社区不完全统计，按照模板提问，可以加快回复和解决问题的速度

环境

【FastDeploy版本】： docker镜像 registry.baidubce.com/paddlepaddle/fastdeploy:x.y.z-gpu-cuda11.4-trt8.4-21.10
【系统平台】: Linux x64(Ubuntu 22.04)
【硬件】： Nvidia GPU P40， CUDA 11.7 CUDNN 8.4
【编译语言】：Python3.8

问题日志及出现问题的操作流程

• 附上详细的问题日志有助于快速定位分析
  call ocr
  0# 0x00005650067998A9 in fastdeployserver
  1# 0x00007FAF577AF210 in /usr/lib/x86_64-linux-gnu/libc.so.6
  2# void phi::fusion::ConvFusionKernel<float, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optionalphi::DenseTensor const&, std::vector<int, std::allocator > const&, std::vector<int, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::vector<int, std::allocator > const&, int, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, bool, std::vector<int, std::allocator > const&, int, phi::DenseTensor*, std::vector<phi::DenseTensor*, std::allocatorphi::DenseTensor* >) in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  3# void phi::KernelImpl<void ()(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optionalphi::DenseTensor const&, std::vector<int, std::allocator > const&, std::vector<int, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::vector<int, std::allocator > const&, int, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, bool, std::vector<int, std::allocator > const&, int, phi::DenseTensor, std::vector<phi::DenseTensor*, std::allocatorphi::DenseTensor* >), &(void phi::fusion::ConvFusionKernel<float, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, paddle::optionalphi::DenseTensor const&, std::vector<int, std::allocator > const&, std::vector<int, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::vector<int, std::allocator > const&, int, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, bool, std::vector<int, std::allocator > const&, int, phi::DenseTensor*, std::vector<phi::DenseTensor*, std::allocatorphi::DenseTensor* >))>::KernelCallHelper<paddle::optionalphi::DenseTensor const&, std::vector<int, std::allocator > const&, std::vector<int, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::vector<int, std::allocator > const&, int, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, bool, std::vector<int, std::allocator > const&, int, phi::DenseTensor*, std::vector<phi::DenseTensor*, std::allocatorphi::DenseTensor* >, phi::TypeTag >::Compute<1, 3, 0, 0, phi::GPUContext const, phi::DenseTensor const, phi::DenseTensor const, phi::DenseTensor const>(phi::KernelContext*, phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&) in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  4# paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, phi::Place const&, paddle::framework::RuntimeContext*) const in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  5# paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, phi::Place const&) const in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  6# paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, phi::Place const&) in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  7# 0x00007FAE3D72085D in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  8# paddle::AnalysisPredictor::ZeroCopyRun() in /opt/fastdeploy/third_libs/install/paddle_inference/paddle/lib/libpaddle_inference.so
  9# fastdeploy::PaddleBackend::Infer(std::vector<fastdeploy::FDTensor, std::allocatorfastdeploy::FDTensor >&, std::vector<fastdeploy::FDTensor, std::allocatorfastdeploy::FDTensor >*, bool) in /opt/fastdeploy/lib/libfastdeploy_runtime.so.0.0.0
  10# fastdeploy::Runtime::Infer() in /opt/fastdeploy/lib/libfastdeploy_runtime.so.0.0.0
  11# 0x00007FAEFC1B0FA4 in /opt/tritonserver/backends/fastdeploy/libtriton_fastdeploy.so
  12# 0x00007FAEFC1B4826 in /opt/tritonserver/backends/fastdeploy/libtriton_fastdeploy.so
  13# TRITONBACKEND_ModelInstanceExecute in /opt/tritonserver/backends/fastdeploy/libtriton_fastdeploy.so
  14# 0x00007FAF5833B83A in /opt/tritonserver/bin/../lib/libtritonserver.so
  15# 0x00007FAF5833C04D in /opt/tritonserver/bin/../lib/libtritonserver.so
  16# 0x00007FAF581F0801 in /opt/tritonserver/bin/../lib/libtritonserver.so
  17# 0x00007FAF58335DC7 in /opt/tritonserver/bin/../lib/libtritonserver.so
  18# 0x00007FAF57B9DDE4 in /usr/lib/x86_64-linux-gnu/libstdc++.so.6
  19# 0x00007FAF5801B609 in /usr/lib/x86_64-linux-gnu/libpthread.so.0
  20# clone in /usr/lib/x86_64-linux-gnu/libc.so.6

[StatusCode.UNAVAILABLE] recvmsg:Connection reset by peer
[StatusCode.UNAVAILABLE] recvmsg:Connection reset by peer
[StatusCode.UNAVAILABLE] recvmsg:Connection reset by peer
[StatusCode.UNAVAILABLE] recvmsg:Connection reset by peer
[StatusCode.UNAVAILABLE] recvmsg:Connection reset by peer

Reproduce the OCR call under the reported pressure test in the FastDeploy Triton backend, then inspect the stack from TRITONBACKEND_ModelInstanceExecute through PaddleBackend::Infer and Runtime::Infer. Done means the process remains available during the test and the reported connection resets are isolated to a specific component.