|
| 1 | +# ────────────────────────────────────────────────────────────── |
| 2 | +# DME ↔ AMDSMI Integration CI |
| 3 | +# ────────────────────────────────────────────────────────────── |
| 4 | +# Architecture: |
| 5 | +# GPU-Agent (C++) ──gRPC──▶ DME (Go) ──▶ Prometheus /metrics |
| 6 | +# │ │ |
| 7 | +# └── libamdsmi.so ◀───────┘ (built from projects/amdsmi) |
| 8 | +# |
| 9 | +# This workflow builds AMDSMI from the super-repo, compiles |
| 10 | +# GPU-Agent and Device Metrics Exporter, then verifies that |
| 11 | +# Prometheus GPU metrics are exposed correctly. |
| 12 | +# ────────────────────────────────────────────────────────────── |
| 13 | + |
| 14 | +name: DME AMDSMI Integration CI |
| 15 | + |
| 16 | +on: |
| 17 | + pull_request: |
| 18 | + branches: [develop] |
| 19 | + paths: |
| 20 | + - 'projects/amdsmi/**' |
| 21 | + - '.github/workflows/dme-amdsmi-ci.yml' |
| 22 | + push: |
| 23 | + branches: [develop] |
| 24 | + paths: |
| 25 | + - 'projects/amdsmi/**' |
| 26 | + - '.github/workflows/dme-amdsmi-ci.yml' |
| 27 | + workflow_dispatch: |
| 28 | + inputs: |
| 29 | + dme_branch: |
| 30 | + description: 'DME branch/tag to test against' |
| 31 | + default: 'main' |
| 32 | + type: string |
| 33 | + gpu_agent_branch: |
| 34 | + description: 'GPU Agent branch/tag (fallback if submodule missing)' |
| 35 | + default: 'main' |
| 36 | + type: string |
| 37 | + |
| 38 | +permissions: |
| 39 | + contents: read |
| 40 | + |
| 41 | +concurrency: |
| 42 | + group: dme-amdsmi-integration-${{ github.ref }} |
| 43 | + cancel-in-progress: true |
| 44 | + |
| 45 | +env: |
| 46 | + DEBIAN_FRONTEND: noninteractive |
| 47 | + BUILD_TYPE: Release |
| 48 | + DME_REPO: https://github.com/ROCm/device-metrics-exporter.git |
| 49 | + GPU_AGENT_REPO: https://github.com/ROCm/gpu-agent.git |
| 50 | + DME_BRANCH: ${{ inputs.dme_branch || 'main' }} |
| 51 | + GPU_AGENT_BRANCH: ${{ inputs.gpu_agent_branch || 'main' }} |
| 52 | + GO_VERSION: '1.25.5' |
| 53 | + DME_METRICS_PORT: 5000 |
| 54 | + DME_DIR: /tmp/dme |
| 55 | + # GPU Agent Makefile hardcodes ABS_DIR to this path |
| 56 | + GPU_AGENT_WORKDIR: /usr/src/github.com/ROCm/gpu-agent |
| 57 | + |
| 58 | +jobs: |
| 59 | + integration-test: |
| 60 | + name: DME Integration • ${{ matrix.os }} |
| 61 | + runs-on: ${{ vars.RUNNER_TYPE }} |
| 62 | + continue-on-error: true |
| 63 | + timeout-minutes: 120 |
| 64 | + strategy: |
| 65 | + fail-fast: false |
| 66 | + matrix: |
| 67 | + os: [Ubuntu22] |
| 68 | + container: |
| 69 | + image: ${{ vars[format('{0}_DOCKER_IMAGE', matrix.os)] }} |
| 70 | + options: >- |
| 71 | + --rm --privileged |
| 72 | + --device=/dev/kfd |
| 73 | + --device=/dev/dri |
| 74 | + --group-add video |
| 75 | +
|
| 76 | + steps: |
| 77 | + # ──────────────────────────────────────────────────────── |
| 78 | + # Setup |
| 79 | + # ──────────────────────────────────────────────────────── |
| 80 | + - name: Checkout Super-repo |
| 81 | + uses: actions/checkout@v4 |
| 82 | + |
| 83 | + - name: Set Project Directory |
| 84 | + run: | |
| 85 | + PROJECT_DIR=$(find "$(pwd)" -maxdepth 3 -name "CMakeLists.txt" \ |
| 86 | + -path "*/amdsmi/*" -exec dirname {} \; | head -1) |
| 87 | + echo "AMDSMI_DIR=${PROJECT_DIR}" >> "$GITHUB_ENV" |
| 88 | + echo "AMDSMI project: ${PROJECT_DIR}" |
| 89 | +
|
| 90 | + - name: Install System Dependencies |
| 91 | + run: | |
| 92 | + max_retries=3 |
| 93 | + for i in $(seq 1 $max_retries); do |
| 94 | + apt-get update && break || sleep 5 |
| 95 | + done |
| 96 | + apt-get install -y --no-install-recommends \ |
| 97 | + build-essential cmake git curl wget ca-certificates \ |
| 98 | + pkg-config libdrm-dev libpci-dev \ |
| 99 | + autoconf automake libtool unzip |
| 100 | +
|
| 101 | + - name: Install Go ${{ env.GO_VERSION }} |
| 102 | + run: | |
| 103 | + wget -q "https://go.dev/dl/go${{ env.GO_VERSION }}.linux-amd64.tar.gz" \ |
| 104 | + -O /tmp/go.tar.gz |
| 105 | + rm -rf /usr/local/go |
| 106 | + tar -C /usr/local -xzf /tmp/go.tar.gz |
| 107 | + rm -f /tmp/go.tar.gz |
| 108 | + echo "/usr/local/go/bin" >> "$GITHUB_PATH" |
| 109 | + echo "${HOME}/go/bin" >> "$GITHUB_PATH" |
| 110 | + /usr/local/go/bin/go version |
| 111 | +
|
| 112 | + # ──────────────────────────────────────────────────────── |
| 113 | + # Phase 1 — Build AMDSMI from the super-repo |
| 114 | + # ──────────────────────────────────────────────────────── |
| 115 | + - name: Build and Install AMDSMI |
| 116 | + run: | |
| 117 | + echo "::group::cmake configure" |
| 118 | + cd "${{ env.AMDSMI_DIR }}" |
| 119 | + mkdir -p build && cd build |
| 120 | + cmake .. \ |
| 121 | + -DBUILD_TESTS=ON \ |
| 122 | + -DENABLE_ESMI_LIB=ON \ |
| 123 | + -DCMAKE_INSTALL_PREFIX=/opt/rocm \ |
| 124 | + -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} |
| 125 | + echo "::endgroup::" |
| 126 | +
|
| 127 | + echo "::group::make" |
| 128 | + make -j "$(nproc)" |
| 129 | + echo "::endgroup::" |
| 130 | +
|
| 131 | + make install |
| 132 | + ldconfig |
| 133 | +
|
| 134 | + echo "=== Verify AMDSMI installation ===" |
| 135 | + ls -la /opt/rocm/lib/libamd_smi.so* |
| 136 | + ls /opt/rocm/include/amd_smi/ | head -10 |
| 137 | +
|
| 138 | + # ──────────────────────────────────────────────────────── |
| 139 | + # Phase 2 — Clone and prepare build environment |
| 140 | + # ──────────────────────────────────────────────────────── |
| 141 | + - name: Clone Device Metrics Exporter |
| 142 | + run: | |
| 143 | + git clone --recurse-submodules --depth 1 \ |
| 144 | + -b "${{ env.DME_BRANCH }}" \ |
| 145 | + "${{ env.DME_REPO }}" "${{ env.DME_DIR }}" |
| 146 | +
|
| 147 | + # Verify gpu-agent submodule was cloned |
| 148 | + if [ ! -d "${{ env.DME_DIR }}/gpuagent/sw" ]; then |
| 149 | + echo "GPU Agent submodule not populated — cloning separately" |
| 150 | + rm -rf "${{ env.DME_DIR }}/gpuagent" |
| 151 | + git clone --recurse-submodules --depth 1 \ |
| 152 | + -b "${{ env.GPU_AGENT_BRANCH }}" \ |
| 153 | + "${{ env.GPU_AGENT_REPO }}" "${{ env.DME_DIR }}/gpuagent" |
| 154 | + fi |
| 155 | +
|
| 156 | + echo "=== DME directory ===" |
| 157 | + ls "${{ env.DME_DIR }}/" |
| 158 | +
|
| 159 | + - name: Prepare GPU Agent Build Environment |
| 160 | + run: | |
| 161 | + GPUAGENT_SRC="${{ env.DME_DIR }}/gpuagent" |
| 162 | +
|
| 163 | + # GPU Agent Makefile hardcodes ABS_DIR — symlink to match |
| 164 | + mkdir -p "$(dirname "${{ env.GPU_AGENT_WORKDIR }}")" |
| 165 | + ln -sfn "${GPUAGENT_SRC}" "${{ env.GPU_AGENT_WORKDIR }}" |
| 166 | +
|
| 167 | + # Copy locally-built AMDSMI to gpu-agent's expected paths |
| 168 | + AMDSMI_TP="${GPUAGENT_SRC}/sw/nic/third-party/rocm/amd_smi_lib" |
| 169 | + mkdir -p "${AMDSMI_TP}/include" "${AMDSMI_TP}/x86_64/lib" |
| 170 | + cp -v /opt/rocm/lib/libamd_smi.so* "${AMDSMI_TP}/x86_64/lib/" |
| 171 | + cp -rv /opt/rocm/include/amd_smi/* "${AMDSMI_TP}/include/" |
| 172 | +
|
| 173 | + # Place libs in the runtime lib dir as well |
| 174 | + BLD_LIB="${GPUAGENT_SRC}/sw/nic/build/x86_64/sim/lib" |
| 175 | + mkdir -p "${BLD_LIB}" |
| 176 | + cp -v /opt/rocm/lib/libamd_smi.so* "${BLD_LIB}/" |
| 177 | +
|
| 178 | + echo "=== GPU Agent build environment ready ===" |
| 179 | +
|
| 180 | + # ──────────────────────────────────────────────────────── |
| 181 | + # Phase 3 — Build GPU Agent (C++ binary) |
| 182 | + # ──────────────────────────────────────────────────────── |
| 183 | + - name: Build GPU Agent Third-party Libraries |
| 184 | + timeout-minutes: 60 |
| 185 | + run: | |
| 186 | + export PATH="/usr/local/go/bin:${HOME}/go/bin:${PATH}" |
| 187 | + cd "${{ env.GPU_AGENT_WORKDIR }}/sw/nic/gpuagent" |
| 188 | +
|
| 189 | + echo "Building third-party C++ libraries (protobuf, gRPC, abseil, …)" |
| 190 | + echo "::group::build-libs" |
| 191 | + make build-libs 2>&1 | tail -50 |
| 192 | + echo "::endgroup::" |
| 193 | +
|
| 194 | + - name: Build GPU Agent Binary |
| 195 | + timeout-minutes: 30 |
| 196 | + run: | |
| 197 | + export PATH="/usr/local/go/bin:${HOME}/go/bin:${PATH}" |
| 198 | + cd "${{ env.GPU_AGENT_WORKDIR }}/sw/nic/gpuagent" |
| 199 | +
|
| 200 | + go mod vendor |
| 201 | +
|
| 202 | + echo "::group::make gpuagent" |
| 203 | + make gpuagent 2>&1 | tail -40 |
| 204 | + echo "::endgroup::" |
| 205 | +
|
| 206 | + BLD_BIN="${{ env.GPU_AGENT_WORKDIR }}/sw/nic/build/x86_64/sim/bin" |
| 207 | + echo "=== GPU Agent build artifacts ===" |
| 208 | + ls -la "${BLD_BIN}/" |
| 209 | +
|
| 210 | + # ──────────────────────────────────────────────────────── |
| 211 | + # Phase 4 — Build Device Metrics Exporter (Go binary) |
| 212 | + # ──────────────────────────────────────────────────────── |
| 213 | + - name: Install Go Protobuf Tools |
| 214 | + run: | |
| 215 | + export PATH="/usr/local/go/bin:${HOME}/go/bin:${PATH}" |
| 216 | + go install google.golang.org/protobuf/cmd/protoc-gen-go@latest |
| 217 | + go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@latest |
| 218 | +
|
| 219 | + - name: Build Device Metrics Exporter |
| 220 | + timeout-minutes: 15 |
| 221 | + run: | |
| 222 | + export PATH="/usr/local/go/bin:${HOME}/go/bin:${PATH}" |
| 223 | + cd "${{ env.DME_DIR }}" |
| 224 | +
|
| 225 | + mkdir -p bin |
| 226 | + CGO_ENABLED=0 go build -C cmd/exporter \ |
| 227 | + -o "$(pwd)/bin/amd-metrics-exporter" |
| 228 | +
|
| 229 | + ls -la bin/amd-metrics-exporter |
| 230 | + echo "=== DME build complete ===" |
| 231 | +
|
| 232 | + # ──────────────────────────────────────────────────────── |
| 233 | + # Phase 5 — Integration Test |
| 234 | + # ──────────────────────────────────────────────────────── |
| 235 | + - name: Start GPU Agent |
| 236 | + run: | |
| 237 | + BLD_DIR="${{ env.GPU_AGENT_WORKDIR }}/sw/nic/build/x86_64/sim" |
| 238 | + export LD_LIBRARY_PATH="${BLD_DIR}/lib:/opt/rocm/lib:${LD_LIBRARY_PATH:-}" |
| 239 | +
|
| 240 | + echo "=== Starting GPU Agent ===" |
| 241 | + nohup "${BLD_DIR}/bin/gpuagent" > /tmp/gpuagent.log 2>&1 & |
| 242 | + GPUAGENT_PID=$! |
| 243 | + echo "GPUAGENT_PID=${GPUAGENT_PID}" >> "$GITHUB_ENV" |
| 244 | +
|
| 245 | + sleep 5 |
| 246 | +
|
| 247 | + if kill -0 "${GPUAGENT_PID}" 2>/dev/null; then |
| 248 | + echo "GPU Agent running (PID ${GPUAGENT_PID})" |
| 249 | + else |
| 250 | + echo "::error::GPU Agent failed to start" |
| 251 | + cat /tmp/gpuagent.log |
| 252 | + exit 1 |
| 253 | + fi |
| 254 | +
|
| 255 | + - name: Start Device Metrics Exporter |
| 256 | + run: | |
| 257 | + export LD_LIBRARY_PATH="/opt/rocm/lib:${LD_LIBRARY_PATH:-}" |
| 258 | +
|
| 259 | + echo "=== Starting DME ===" |
| 260 | + nohup "${{ env.DME_DIR }}/bin/amd-metrics-exporter" \ |
| 261 | + > /tmp/dme.log 2>&1 & |
| 262 | + DME_PID=$! |
| 263 | + echo "DME_PID=${DME_PID}" >> "$GITHUB_ENV" |
| 264 | +
|
| 265 | + sleep 5 |
| 266 | +
|
| 267 | + if kill -0 "${DME_PID}" 2>/dev/null; then |
| 268 | + echo "DME running (PID ${DME_PID})" |
| 269 | + else |
| 270 | + echo "::error::DME failed to start" |
| 271 | + cat /tmp/dme.log |
| 272 | + exit 1 |
| 273 | + fi |
| 274 | +
|
| 275 | + - name: Verify Prometheus Metrics Endpoint |
| 276 | + run: | |
| 277 | + MAX_RETRIES=5 |
| 278 | + RETRY_DELAY=3 |
| 279 | +
|
| 280 | + for i in $(seq 1 ${MAX_RETRIES}); do |
| 281 | + echo "Attempt ${i}/${MAX_RETRIES}: querying metrics …" |
| 282 | + HTTP_CODE=$(curl -s -o /tmp/metrics_output.txt \ |
| 283 | + -w "%{http_code}" \ |
| 284 | + "http://localhost:${{ env.DME_METRICS_PORT }}/metrics" \ |
| 285 | + || echo "000") |
| 286 | +
|
| 287 | + if [ "${HTTP_CODE}" = "200" ]; then |
| 288 | + echo "=== Metrics endpoint returned HTTP 200 ===" |
| 289 | + head -50 /tmp/metrics_output.txt |
| 290 | + echo "---" |
| 291 | +
|
| 292 | + if grep -q "^# HELP" /tmp/metrics_output.txt && |
| 293 | + grep -q "^# TYPE" /tmp/metrics_output.txt; then |
| 294 | + echo "=== Prometheus format verified ===" |
| 295 | + else |
| 296 | + echo "::warning::Response may not be standard Prometheus format" |
| 297 | + fi |
| 298 | + exit 0 |
| 299 | + fi |
| 300 | +
|
| 301 | + echo "HTTP ${HTTP_CODE} — retrying in ${RETRY_DELAY}s …" |
| 302 | + sleep "${RETRY_DELAY}" |
| 303 | + done |
| 304 | +
|
| 305 | + echo "::error::Metrics endpoint unreachable after ${MAX_RETRIES} attempts" |
| 306 | + echo "--- gpu-agent log ---" |
| 307 | + tail -100 /tmp/gpuagent.log 2>/dev/null || true |
| 308 | + echo "--- dme log ---" |
| 309 | + tail -100 /tmp/dme.log 2>/dev/null || true |
| 310 | + exit 1 |
| 311 | +
|
| 312 | + # ──────────────────────────────────────────────────────── |
| 313 | + # Cleanup & artifacts |
| 314 | + # ──────────────────────────────────────────────────────── |
| 315 | + - name: Collect Logs |
| 316 | + if: always() |
| 317 | + run: | |
| 318 | + mkdir -p /tmp/integration-logs |
| 319 | + cp /tmp/gpuagent.log /tmp/integration-logs/ 2>/dev/null || true |
| 320 | + cp /tmp/dme.log /tmp/integration-logs/ 2>/dev/null || true |
| 321 | + cp /tmp/metrics_output.txt /tmp/integration-logs/ 2>/dev/null || true |
| 322 | +
|
| 323 | + { |
| 324 | + echo "=== System ===" |
| 325 | + uname -a |
| 326 | + echo "=== GPU devices ===" |
| 327 | + ls -la /dev/kfd /dev/dri/ 2>&1 || true |
| 328 | + echo "=== AMDSMI libs ===" |
| 329 | + ldconfig -p | grep amd_smi || true |
| 330 | + } > /tmp/integration-logs/system-info.txt 2>&1 |
| 331 | +
|
| 332 | + - name: Upload Integration Test Logs |
| 333 | + if: always() |
| 334 | + uses: actions/upload-artifact@v4 |
| 335 | + with: |
| 336 | + name: dme-integration-logs-${{ matrix.os }} |
| 337 | + path: /tmp/integration-logs/ |
| 338 | + retention-days: 7 |
| 339 | + |
| 340 | + - name: Cleanup |
| 341 | + if: always() |
| 342 | + run: | |
| 343 | + for pid in "${{ env.DME_PID }}" "${{ env.GPUAGENT_PID }}"; do |
| 344 | + [ -n "${pid}" ] || continue |
| 345 | + kill "${pid}" 2>/dev/null || true |
| 346 | + done |
| 347 | + sleep 2 |
| 348 | + for pid in "${{ env.DME_PID }}" "${{ env.GPUAGENT_PID }}"; do |
| 349 | + [ -n "${pid}" ] || continue |
| 350 | + kill -9 "${pid}" 2>/dev/null || true |
| 351 | + done |
0 commit comments