NV triton啟動方式說明

1. 地址：
   1. server：https://github.com/triton-inference-server/server
   2. client：https://github.com/triton-inference-server/client
2. 編譯部署方式：
   1. xx.yy-py3 包括server，可用於直接部署server鏡像
   2. xx.yy-py3-sdk 包括python、c++ client示例，用於直接使用client鏡像
   3. xx.yy-py3-min 基礎環境，用於編譯開發，教程：https://github.com/triton-inference-server/server/blob/main/docs/build.md#ubuntu-docker
   4. 非docker編譯開發，教程：https://github.com/triton-inference-server/server/blob/main/docs/build.md#ubuntu-without-docker
3. 以gpu部署方式為例：

   1. 部署server：

      docker pull nvcr.io/nvidia/tritonserver:21.05-py3
      git clone https://github.com/triton-inference-server/server.git
      cd server/docs/examples
      ./fetch_models.sh
      docker run --gpus=1 --rm -p8010:8000 -p8011:8001 -p8012:8002 -v/mnt/zhangliang35/code/github/triton/triton-inference-server/server/docs/examples/model_repository:/models nvcr.io/nvidia/tritonserver:21.05-py3 tritonserver --model-repository=/models

       

      測試服務是否正常啟動：curl -v localhost:8010/v2/health/ready 返回200表明啟動正常。服務8000為rpc端口，8001為rpc端口，8002為Metrics端口

   2. 部署client：

      docker pull nvcr.io/nvidia/tritonserver:21.05-py3-sdk
      docker run -it --rm --net=host nvcr.io/nvidia/tritonserver:21.05-py3-sdk
      /workspace/install/bin/image_client -m densenet_onnx -u localhost:8010 -c 3 -s INCEPTION /workspace/images/mug.jpg
      返回：
      Request 0, batch size 1
      Image '/workspace/images/mug.jpg':
          15.349568 (504) = COFFEE MUG
          13.227468 (968) = CUP
          10.424896 (505) = COFFEEPOT

       

      其中，-i grpc -u localhost:8001 可以指定client請求grpc端口8001

4. 鏡像組成分析：
   1. client：
      1. /workspace/install/bin目錄下存放各類client c++ bin文件
      2. /workspace/client中存放client源碼
      3. /workspace/build中存放編譯產出
      4. /workspace/images中存放測試圖片
   2. server（工作目錄為/opt/tritonserver）：
      1. /bin/tritonserver
      2. backends 存放各類依賴
      3. include、lib存放頭文件及so庫

      4. nvidia_entrypoint.sh  配置環境，然后透傳啟動命令

         #!/bin/bash
         # Copyright (c) 2019-2021, NVIDIA CORPORATION. All rights reserved.
         #
         # Redistribution and use in source and binary forms, with or without
         # modification, are permitted provided that the following conditions
         # are met:
         #  * Redistributions of source code must retain the above copyright
         #    notice, this list of conditions and the following disclaimer.
         #  * Redistributions in binary form must reproduce the above copyright
         #    notice, this list of conditions and the following disclaimer in the
         #    documentation and/or other materials provided with the distribution.
         #  * Neither the name of NVIDIA CORPORATION nor the names of its
         #    contributors may be used to endorse or promote products derived
         #    from this software without specific prior written permission.
         #
         # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
         # EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
         # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
         # PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR
         # CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
         # EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
         # PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
         # PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
         # OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
         # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
         # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
          
         set -e
         cat <<EOF
          
         =============================
         == Triton Inference Server ==
         =============================
          
         NVIDIA Release ${NVIDIA_TRITON_SERVER_VERSION} (build ${NVIDIA_BUILD_ID})
          
         Copyright (c) 2018-2021, NVIDIA CORPORATION.  All rights reserved.
          
         Various files include modifications (c) NVIDIA CORPORATION.  All rights reserved.
          
         This container image and its contents are governed by the NVIDIA Deep Learning Container License.
         By pulling and using the container, you accept the terms and conditions of this license:
         https://developer.nvidia.com/ngc/nvidia-deep-learning-container-license
         EOF
          
         if [[ "$(find -L /usr -name libcuda.so.1 | grep -v "compat") " == " " || "$(ls /dev/nvidiactl 2>/dev/null) " == " " ]]; then
           echo
           echo "WARNING: The NVIDIA Driver was not detected.  GPU functionality will not be available."
           echo "   Use Docker with NVIDIA Container Toolkit to start this container; see"
           echo "   https://github.com/NVIDIA/nvidia-docker."
           ln -s `find / -name libnvidia-ml.so -print -quit` /opt/tritonserver/lib/libnvidia-ml.so.1
           export TRITON_SERVER_CPU_ONLY=1
         else
           ( /usr/local/bin/checkSMVER.sh )
           DRIVER_VERSION=$(sed -n 's/^NVRM.*Kernel Module *\([0-9.]*\).*$/\1/p' /proc/driver/nvidia/version 2>/dev/null || true)
           if [[ ! "$DRIVER_VERSION" =~ ^[0-9]*.[0-9]*(.[0-9]*)?$ ]]; then
             echo "Failed to detect NVIDIA driver version."
           elif [[ "${DRIVER_VERSION%%.*}" -lt "${CUDA_DRIVER_VERSION%%.*}" ]]; then
             if [[ "${_CUDA_COMPAT_STATUS}" == "CUDA Driver OK" ]]; then
               echo
               echo "NOTE: Legacy NVIDIA Driver detected.  Compatibility mode ENABLED."
             else
               echo
               echo "ERROR: This container was built for NVIDIA Driver Release ${CUDA_DRIVER_VERSION%.*} or later, but"
               echo "       version ${DRIVER_VERSION} was detected and compatibility mode is UNAVAILABLE."
               echo
               echo "       [[${_CUDA_COMPAT_STATUS}]]"
               sleep 2
             fi
           fi
         fi
          
         if ! cat /proc/cpuinfo | grep flags | sort -u | grep avx >& /dev/null; then
           echo
           echo "ERROR: This container was built for CPUs supporting at least the AVX instruction set, but"
           echo "       the CPU detected was $(cat /proc/cpuinfo |grep "model name" | sed 's/^.*: //' | sort -u), which does not report"
           echo "       support for AVX.  An Illegal Instrution exception at runtime is likely to result."
           echo "       See https://en.wikipedia.org/wiki/Advanced_Vector_Extensions#CPUs_with_AVX ."
           sleep 2
         fi
          
         echo
          
         if [[ $# -eq 0 ]]; then
           exec "/bin/bash"
         else
           exec "$@"
         fi