NV triton启动方式说明

地址：
1. server：https://github.com/triton-inference-server/server
2. client：https://github.com/triton-inference-server/client
编译部署方式：
1. xx.yy-py3 包括server，可用于直接部署server镜像
2. xx.yy-py3-sdk 包括python、c++ client示例，用于直接使用client镜像
3. xx.yy-py3-min 基础环境，用于编译开发，教程：https://github.com/triton-inference-server/server/blob/main/docs/build.md#ubuntu-docker
4. 非docker编译开发，教程：https://github.com/triton-inference-server/server/blob/main/docs/build.md#ubuntu-without-docker

以gpu部署方式为例：

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

测试服务是否正常启动：curl -v localhost:8010/v2/health/ready 返回200表明启动正常。服务8000为rpc端口，8001为rpc端口，8002为Metrics端口

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请求grpc端口8001

镜像组成分析：

client：
1. /workspace/install/bin目录下存放各类client c++ bin文件
2. /workspace/client中存放client源码
3. /workspace/build中存放编译产出
4. /workspace/images中存放测试图片

server（工作目录为/opt/tritonserver）：

/bin/tritonserver
backends 存放各类依赖
include、lib存放头文件及so库

nvidia_entrypoint.sh 配置环境，然后透传启动命令

#!/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