migrate: SQA bug-fix bundle from mono-repo (nvds_logger, tracker_ReID, build.sh, BUILD.md, MR !89 Triton samples)

BUILD.md

@@ -0,0 +1,173 @@
+# DeepStream Mono-Repo — Build & Run Guide
+
+## Prerequisites
+
+### NVIDIA Compute Stack
+
+Before building from source, the NVIDIA compute stack must be installed on the host. Required versions:
+
+| Platform | OS | Driver | CUDA | cuDNN | TensorRT |
+|---|---|---|---|---|---|
+| x86 dGPU | Ubuntu 24.04 | 590.48.01+ | 13.1 | 9.18.0 | 10.14.x |
+| Jetson (aarch64) | JetPack 7.1 GA | — (bundled) | 13.0 | 9.12.0 | 10.13.x |
+| SBSA / DGX Spark | Ubuntu 24.04 | 590.48.01+ | 13.0 | 9.18.0 | 10.14.x |
+
+For installation instructions, refer to the official [DeepStream SDK Installation Guide](https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Installation.html).
+
+> **Docker users:** NVIDIA NGC provides pre-built DeepStream 9.0 containers with the full compute stack pre-installed. If using a container, skip this section and proceed to Git LFS setup below.
+
+---
+
+### Clone the Repository
+
+```bash
+sudo apt-get install cmake
+```
+
+### x86 (dGPU)
+
+1. Download the DS 9.0 public released Debian package, install it, and add the
+   DeepStream lib directory to `LD_LIBRARY_PATH` so the runtime can locate the
+   shared libraries:
+   ```bash
+   sudo apt install ./deepstream-9.0_9.0.0-1_amd64.deb
+   sudo /opt/nvidia/deepstream/deepstream-9.0/install.sh
+   export LD_LIBRARY_PATH=/opt/nvidia/deepstream/deepstream-9.0/lib:$LD_LIBRARY_PATH
+   ```
+   Append the `export` line to `~/.bashrc` to persist it across shells.
+
+2. Install system dependencies:
+   ```bash
+   sudo apt install \
+     libssl3 \
+     libssl-dev \
+     libcurl4-openssl-dev \
+     libgles2-mesa-dev \
+     libgstreamer1.0-0 \
+     gstreamer1.0-tools \
+     gstreamer1.0-plugins-good \
+     gstreamer1.0-plugins-bad \
+     gstreamer1.0-plugins-ugly \
+     gstreamer1.0-libav \
+     libgstreamer-plugins-base1.0-dev \
+     libgstrtspserver-1.0-0 \
+     libjansson4 \
+     libjsoncpp-dev \
+     libyaml-cpp-dev \
+     libmosquitto1
+   ```
+
+   See also the [DeepStream Installation Guide — Install Prerequisite Packages](https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Installation.html#id2).
+
+### aarch64 (Jetson)
+
+1. Download the DS 9.0 public released Debian package, install it, and add the
+   DeepStream lib directory to `LD_LIBRARY_PATH` so the runtime can locate the
+   shared libraries:
+   ```bash
+   sudo apt install ./deepstream-9.0_9.0.0-1_arm64.deb
+   sudo /opt/nvidia/deepstream/deepstream-9.0/install.sh
+   export LD_LIBRARY_PATH=/opt/nvidia/deepstream/deepstream-9.0/lib:$LD_LIBRARY_PATH
+   ```
+   Append the `export` line to `~/.bashrc` to persist it across shells.
+
+2. Install system dependencies:
+   ```bash
+   sudo apt install \
+     libssl3 \
+     libssl-dev \
+     libcurl4-openssl-dev \
+     libgstreamer1.0-0 \
+     gstreamer1.0-tools \
+     gstreamer1.0-plugins-good \
+     gstreamer1.0-plugins-bad \
+     gstreamer1.0-plugins-ugly \
+     gstreamer1.0-libav \
+     libgstreamer-plugins-base1.0-dev \
+     libgstrtspserver-1.0-0 \
+     libjansson4 \
+     libjsoncpp-dev \
+     libyaml-cpp-dev \
+     libmosquitto1
+   ```
+
+   See also the [DeepStream Installation Guide — Install Prerequisite Packages](https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Installation.html#install-prerequisite-packages).
+
+### SBSA / DGX Spark (Server Base System Architecture)
+
+SBSA and DGX Spark share the same code path. There is no DS tar/deb package for this platform. The validated environment is the SBSA Docker image.
+Build and validation must be done **inside the Docker container**.
+
+1. Start the SBSA Docker container (with GPU access):
+   ```bash
+   docker run -it --gpus all <sbsa-docker-image> bash
+   ```
+
+2. Install build prerequisites inside the container:
+   ```bash
+   apt-get update && apt-get install -y git cmake make build-essential
+   ```
+
+3. `sudo` is not present in the container by default. Create a passthrough stub since you are already root:
+   ```bash
+   printf '#!/bin/sh\nexec env "$@"\n' > /usr/local/bin/sudo && chmod +x /usr/local/bin/sudo
+   ```
+
+4. Clone the mono-repo:
+   ```bash
+   git clone https://gitlab-master.nvidia.com/DeepStreamSDK/deepstream -b main deepstream
+   cd deepstream
+   ```
+
+---
+
+## Build and Install
+
+`build.sh` auto-detects the host platform and:
+- On SBSA / DGX Spark: passes `AARCH64_IS_SBSA=1` to all Makefiles (both `aarch64` and `sbsa` report `uname -m` as `aarch64`; SBSA is identified by the absence of `/etc/nv_tegra_release`)
+- Builds and installs open-source dependencies (opentelemetry, civetweb, prometheus-cpp, azure-iot-sdk)
+- Builds all source components and installs them directly to `/opt/nvidia/deepstream/deepstream-<NVDS_VERSION>/`
+
+```bash
+sudo bash build.sh
+```
+
+Default CUDA version is architecture-dependent (`13.1` for x86_64, `13.0` for aarch64/sbsa/DGX Spark). To override:
+
+```bash
+sudo CUDA_VER=13.1 bash build.sh   # x86
+sudo CUDA_VER=13.0 bash build.sh   # aarch64
+```
+
+To override the target DeepStream version (default: 9.0):
+
+```bash
+sudo NVDS_VERSION=9.0 bash build.sh
+```
+
+After first run, clear the GStreamer plugin cache so it picks up newly installed plugins:
+
+```bash
+rm -rf ~/.cache/gstreamer-1.0/
+```
+
+**Build outputs:**
+- GStreamer plugins       → `/opt/nvidia/deepstream/deepstream-<NVDS_VERSION>/lib/gst-plugins/`
+- Utility libs            → `/opt/nvidia/deepstream/deepstream-<NVDS_VERSION>/lib/`
+- Sample app binaries     → `/opt/nvidia/deepstream/deepstream-<NVDS_VERSION>/bin/`
+- service-maker app bins  → `/opt/nvidia/deepstream/deepstream-<NVDS_VERSION>/bin/service-maker-<app>`
+- service-maker modules   → `/opt/nvidia/deepstream/deepstream-<NVDS_VERSION>/service-maker/modules/`
+
+---
+
+## service-maker — Local Compilation
+
+Individual service-maker apps can be compiled locally without running `build.sh`.
+From within any app directory (e.g. `src/service-maker/sources/apps/cpp/deepstream_test1_app/`):
+
+```bash
+mkdir build && cd build && cmake .. && make
+```
+
+The binary is placed in the local `build/` directory and nothing is installed to `/opt`.
+See each app's `README` for usage instructions.

build.sh

@@ -0,0 +1,138 @@
+#!/usr/bin/env bash
+# Build and install all DeepStream mono-repo components.
+# Usage: bash build.sh [CUDA_VER=<ver>]
+set -e
+
+ARCH=$(uname -m)
+case "$ARCH" in
+  x86_64)  PLATFORM=x86;     CUDA_VER_DEFAULT=13.1 ;;
+  aarch64)
+    # Distinguish Jetson (has /etc/nv_tegra_release) from SBSA server hardware
+    if [ -f /etc/nv_tegra_release ]; then
+      PLATFORM=aarch64
+    else
+      PLATFORM=sbsa
+    fi
+    CUDA_VER_DEFAULT=13.0
+    ;;
+  *)        echo "Unsupported architecture: $ARCH"; exit 1 ;;
+esac
+
+CUDA_VER=${CUDA_VER:-$CUDA_VER_DEFAULT}
+NVDS_VERSION=${NVDS_VERSION:-9.0}
+SM_APP_BUILD=/tmp/ds-sm-apps
+SM_MOD_BUILD=/tmp/ds-sm-modules
+SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+
+# Extra make flags for SBSA — passes AARCH64_IS_SBSA=1 to all Makefiles
+PLATFORM_MAKE_FLAGS=
+[ "$PLATFORM" = "sbsa" ] && PLATFORM_MAKE_FLAGS="AARCH64_IS_SBSA=1"
+
+# ---------------------------------------------------------------------------
+# Prerequisites: install open-source dependencies
+# (Proprietary prebuilt libs are provided by the DS 9.0 deb install at
+#  /opt/nvidia/deepstream/deepstream-${NVDS_VERSION}/lib/.)
+# ---------------------------------------------------------------------------
+echo "==> Installing open-source dependencies"
+sudo NVDS_VERSION=$NVDS_VERSION bash "$SCRIPT_DIR/scripts/install_opensource_deps.sh"
+
+echo "==> Building PLATFORM=$PLATFORM CUDA_VER=$CUDA_VER (outputs directly to /opt/nvidia/deepstream/deepstream-${NVDS_VERSION}/)"
+
+MK="CUDA_VER=$CUDA_VER NVDS_VERSION=$NVDS_VERSION $PLATFORM_MAKE_FLAGS"
+
+# nvds_rest_server must be built before gstnvdscustomhelper (it links against it)
+make -C src/utils/nvds_rest_server $MK
+
+# gst-utils (order matters — gstnvcustomhelper must come first)
+make -C src/gst-utils/gstnvcustomhelper $MK
+make -C src/gst-utils/gst-nvdssr $MK
+make -C src/gst-utils/gstnvdscustomhelper $MK
+
+# Utils (no top-level Makefile — build each individually)
+for dir in src/utils/*/; do
+  make -C "$dir" $MK 2>/dev/null || true
+done
+
+# nvds_msgapi message broker adaptors (Makefiles live one level deeper than
+# the rest of src/utils, and azure_protocol_adaptor splits further into
+# device_client/ and module_client/).
+for dir in src/utils/nvds_msgapi/*/; do
+  if [ -f "$dir/Makefile" ]; then
+    make -C "$dir" $MK 2>/dev/null || true
+  else
+    for sub in "$dir"*/; do
+      [ -f "$sub/Makefile" ] && make -C "$sub" $MK 2>/dev/null || true
+    done
+  fi
+done
+
+# GStreamer plugins (no top-level Makefile — build each individually)
+for dir in src/gst-plugins/*/; do
+  make -C "$dir" $MK 2>/dev/null || true
+done
+
+# Sample apps (no top-level Makefile — build each individually)
+# Build to a staging dir first so a failed link step does not delete the
+# deb-installed binary from the DS bin dir (make removes its output on error).
+SA_BIN_STAGE=/tmp/ds-sample-apps-bin
+DS_BIN=/opt/nvidia/deepstream/deepstream-${NVDS_VERSION}/bin
+mkdir -p "$SA_BIN_STAGE"
+for dir in src/apps/sample_apps/*/; do
+  make -C "$dir" $MK BIN_DIR="$SA_BIN_STAGE" 2>/dev/null || true
+  for bin in "$SA_BIN_STAGE"/deepstream-*; do
+    [ -f "$bin" ] && [ -x "$bin" ] && cp -v "$bin" "$DS_BIN/" || true
+  done
+  rm -f "$SA_BIN_STAGE"/deepstream-*
+done
+rm -rf "$SA_BIN_STAGE"
+
+# Prepare Eigen lib
+sudo apt update || true
+sudo apt install -y libeigen3-dev
+sudo ln -sf /usr/include/eigen3/Eigen /usr/include/Eigen
+
+#Customized Yolo postprocessing lib
+make -C tools/yolo_deepstream/deepstream_yolo/nvdsinfer_custom_impl_Yolo/ $MK 2>/dev/null || true
+
+# tao-apps — build and install via top-level Makefile
+git submodule update --init --recursive || true
+make -C src/apps/tao_apps $MK 2>/dev/null || true
+
+# Reference apps (no top-level Makefile — build each individually)
+RA_BIN_STAGE=/tmp/ds-reference-apps-bin
+mkdir -p "$RA_BIN_STAGE"
+for dir in src/apps/reference_apps/*/; do
+  make -C "$dir" $MK BIN_DIR="$RA_BIN_STAGE" 2>/dev/null || true
+  for bin in "$RA_BIN_STAGE"/deepstream-*; do
+    [ -f "$bin" ] && [ -x "$bin" ] && cp -v "$bin" "$DS_BIN/" || true
+  done
+  rm -f "$RA_BIN_STAGE"/deepstream-*
+done
+rm -rf "$RA_BIN_STAGE"
+
+# service-maker apps (cmake-based) — build and install
+SM_BIN_INSTALL=/opt/nvidia/deepstream/deepstream-${NVDS_VERSION}/bin
+for dir in src/service-maker/sources/apps/cpp/*/; do
+  app=$(basename "$dir")
+  cmake -S "$dir" -B "$SM_APP_BUILD/$app" -DCMAKE_BUILD_TYPE=Release
+  cmake --build "$SM_APP_BUILD/$app" -j$(nproc)
+  for bin in "$SM_APP_BUILD/$app"/deepstream-*; do
+    [ -f "$bin" ] && [ -x "$bin" ] || continue
+    name=$(basename "$bin" | sed 's/^deepstream-/service-maker-/')
+    cp -v "$bin" "$SM_BIN_INSTALL/$name"
+  done
+done
+rm -rf "$SM_APP_BUILD"
+
+# service-maker modules (cmake-based) — build and install
+SM_MODULES_INSTALL=/opt/nvidia/deepstream/deepstream-${NVDS_VERSION}/service-maker/modules
+mkdir -p "$SM_MODULES_INSTALL"
+for dir in src/service-maker/sources/modules/*/; do
+  mod=$(basename "$dir")
+  cmake -S "$dir" -B "$SM_MOD_BUILD/$mod" -DCMAKE_BUILD_TYPE=Release
+  cmake --build "$SM_MOD_BUILD/$mod" -j$(nproc)
+  cp -v "$SM_MOD_BUILD/$mod/lib${mod}.so" "$SM_MODULES_INSTALL/"
+done
+rm -rf "$SM_MOD_BUILD"
+
+echo "==> Done"

src/apps/sample_apps/README.md

@@ -106,3 +106,22 @@ Refer to the `README` in each app's directory for app-specific prerequisites, ru
 | App | Binary | Description |
 |---|---|---|
 | `deepstream-opencv-test` | `deepstream-opencv-test` | Access frame data in CPU memory for OpenCV processing. |
+
+### Triton Inference Server
+
+These samples do not produce a standalone binary. Each builds an `nvinferserver`
+custom-process library (`.so`) and is driven by the deb-installed `deepstream-app`
+with the included config file. See each sample's `README` for the prerequisite
+Triton container setup and model download steps.
+
+| Sample | Library | Description |
+|---|---|---|
+| `TritonBackendEnsemble` | `libnvdstriton_custom_impl_ensemble.so` | Triton ensemble model used as SGIE plus a custom Triton C++ backend that reads DeepStream stream IDs. |
+| `TritonOnnxYolo` | `libnvdstriton_custom_impl_yolo.so` | DS-Triton custom-lib for ONNX YoloV3 with dynamic-sized output tensors and multi-input postprocessing. |
+
+Run flow:
+
+```bash
+cd src/apps/sample_apps/<TritonSample>
+deepstream-app -c <included-config>.txt
+```

src/apps/sample_apps/TritonBackendEnsemble/Makefile

@@ -0,0 +1,22 @@
+################################################################################
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES.
+# All rights reserved.
+# SPDX-License-Identifier: LicenseRef-NvidiaProprietary
+################################################################################
+
+# Delegates to the nvdsinferserver custom-process library so the mono-repo
+# build.sh sample_apps loop picks this sample up. The Triton C++ custom
+# backend under nvds_triton_cpp_custom_backend/ is built separately via cmake
+# (see README) and is not wired into build.sh.
+
+SUBDIRS := nvdsinferserver_custom_impl_ensemble
+
+.PHONY: all clean $(SUBDIRS)
+
+all: $(SUBDIRS)
+
+$(SUBDIRS):
+	$(MAKE) -C $@
+
+clean:
+	$(MAKE) -C nvdsinferserver_custom_impl_ensemble clean