add telemetry stuff for conda-cpp-build workflow

.github/workflows/conda-cpp-build.yaml

@@ -21,6 +21,40 @@ on:
  matrix_filter:
  type: string
  default: "."
+ # telemetry settings: Destination to send telemetry to
+ default_endpoint:
+ type: string
+ # defaults to <default_endpoint>/v1/traces
+ # Change it if you want to send to a different host or port number
+ traces_endpoint:
+ type: string
+ # defaults to <default_endpoint>/v1/metrics
+ # Change it if you want to send to a different host or port number
+ metrics_endpoint:
+ type: string
+ # defaults to <default_endpoint>/v1/logs
+ # Change it if you want to send to a different host or port number
+ logs_endpoint:
+ type: string
+ traceparent:
+ type: string
+ description: |
+ Opentelemetry traceparent. Format is described in https://medium.com/@mesutatasoy/understanding-traceparent-and-microservices-in-opentelemetry-notepad-series-7-de5c16bf6462
+ Generally, 00-<trace_id 32 chars>-<span_id 16 chars>-01
+ parent_span:
+ type: string
+ traces_exporters:
+ type: string
+ description: Exporter name to send data to. May use 'console' for debugging. Can be multiple, comma-separated, 'console,otlp'
+ default: "otlp"
+ metrics_exporters:
+ type: string
+ description: Exporter name to send data to. May use 'console' for debugging. Can be multiple, comma-separated, 'console,otlp'
+ default: "otlp"
+ logs_exporters:
+ type: string
+ description: Exporter name to send data to. May use 'console' for debugging. Can be multiple, comma-separated, 'console,otlp'
+ default: "otlp"
 
 defaults:
  run:
@@ -41,6 +75,21 @@ permissions:
  security-events: none
  statuses: none
 
+env:
+ TOP_LEVEL_TRACEPARENT: ${{ inputs.traceparent }}
+ OTEL_EXPORTER_OTLP_ENDPOINT: "${{ inputs.default-endpoint }}"
+ OTEL_EXPORTER_OTLP_TRACES_ENDPOINT: "${{ inputs.traces-endpoint }}"
+ OTEL_EXPORTER_OTLP_METRICS_ENDPOINT: "${{ inputs.metrics-endpoint }}"
+ OTEL_EXPORTER_OTLP_LOGS_ENDPOINT: "${{ inputs.logs-endpoint }}"
+ OTEL_EXPORTER_OTLP_PROTOCOL: "http/protobuf"
+ OTEL_EXPORTER_OTLP_HEADERS: ${{ secrets.OTEL_EXPORTER_OTLP_HEADERS }}
+ OTEL_EXPORTER_OTLP_CERTIFICATE: "/tmp/certs/ca.crt"
+ OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE: "/tmp/certs/client.crt"
+ OTEL_EXPORTER_OTLP_CLIENT_KEY: "/tmp/certs/client.key"
+ RAPIDS_OTEL_TRACES_EXPORTER: "${{ inputs.traces-exporters || 'otlp' }}"
+ RAPIDS_OTEL_METRICS_EXPORTER: "${{ inputs.metrics-exporters || 'otlp' }}"
+ RAPIDS_OTEL_LOGS_EXPORTER: "${{ inputs.logs-exporters || 'otlp' }}"
+
 jobs:
  compute-matrix:
  runs-on: ubuntu-latest
@@ -59,10 +108,10 @@ jobs:
  export MATRIX="
  # amd64
  - { ARCH: 'amd64', PY_VER: '3.10', CUDA_VER: '11.8.0', LINUX_VER: 'ubuntu22.04' }
- - { ARCH: 'amd64', PY_VER: '3.10', CUDA_VER: '12.5.1', LINUX_VER: 'ubuntu22.04' }
+ #- { ARCH: 'amd64', PY_VER: '3.10', CUDA_VER: '12.5.1', LINUX_VER: 'ubuntu22.04' }
  # arm64
- - { ARCH: 'arm64', PY_VER: '3.10', CUDA_VER: '11.8.0', LINUX_VER: 'ubuntu22.04' }
- - { ARCH: 'arm64', PY_VER: '3.10', CUDA_VER: '12.5.1', LINUX_VER: 'ubuntu22.04' }
+ #- { ARCH: 'arm64', PY_VER: '3.10', CUDA_VER: '11.8.0', LINUX_VER: 'ubuntu22.04' }
+ #- { ARCH: 'arm64', PY_VER: '3.10', CUDA_VER: '12.5.1', LINUX_VER: 'ubuntu22.04' }
  "
 
  MATRIX="$(
@@ -86,11 +135,38 @@ jobs:
  env:
  RAPIDS_BUILD_TYPE: ${{ inputs.build_type }}
  steps:
+ - name: Install OpenTelemetry instrumentation
+ run: |
+ curl -L -o otel-cli-${{ matrix.ARCH }}.deb https://github.com/equinix-labs/otel-cli/releases/download/v0.4.5/otel-cli_0.4.5_linux_${{ matrix.ARCH }}.deb
+ dpkg -i otel-cli-${{ matrix.ARCH }}.deb
+ pip install opentelemetry-distro[otlp] \
+ opentelemetry-exporter-prometheus \
+ git+https://github.com/msarahan/opentelemetry-python-contrib.git@add-conda-build-instrumentation#subdirectory=instrumentation/opentelemetry-instrumentation-conda-build
+ opentelemetry-bootstrap -a install
+ # Replace gha-tools that comes with the image with our hacked one
+ # original install goes to /usr/local/bin
+ # https://github.com/rapidsai/ci-imgs/blob/b1cff14b6d36ab082538fd947ad08f34527a2986/ci-conda.Dockerfile#L178-L180
+ - name: Download gha-tools with git clone
+ run: |
+ git clone https://github.com/msarahan/gha-tools.git -b add-otel-wrapper /tmp/gha-tools
+ cp /tmp/gha-tools/tools/* /usr/local/bin/
  - uses: aws-actions/configure-aws-credentials@v4
  with:
  role-to-assume: ${{ vars.AWS_ROLE_ARN }}
  aws-region: ${{ vars.AWS_REGION }}
  role-duration-seconds: 43200 # 12h
+ - name: Write certificate files for mTLS
+ run: |
+ mkdir -p /tmp/certs
+ cat << EOF > ${OTEL_EXPORTER_OTLP_CERTIFICATE}
+ ${{ secrets.OTEL_EXPORTER_OTLP_CA_CERTIFICATE }}
+ EOF
+ cat << EOF > ${OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE}
+ ${{ secrets.OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE }}
+ EOF
+ cat << EOF > ${OTEL_EXPORTER_OTLP_CLIENT_KEY}
+ ${{ secrets.OTEL_EXPORTER_OTLP_CLIENT_KEY }}
+ EOF
  - uses: actions/checkout@v4
  with:
  repository: ${{ inputs.repo }}
@@ -102,10 +178,29 @@ jobs:
  echo "RAPIDS_SHA=$(git rev-parse HEAD)" >> "${GITHUB_ENV}"
  echo "RAPIDS_REF_NAME=${{ inputs.branch || github.ref_name }}" >> "${GITHUB_ENV}"
  echo "RAPIDS_NIGHTLY_DATE=${{ inputs.date }}" >> "${GITHUB_ENV}"
+ - name: Telemetry setup
+ id: job-traceparent
+ # TODO: This is why we need to merge https://github.com/rapidsai/shared-actions/pull/19 before work on shared-workflows can proceed
+ # The build job can't use actions that aren't approved, so we can't test against a personal branch.
+ uses: rapidsai/shared-actions/telemetry-setup@main
  - name: C++ build
  run: ${{ inputs.script }}
  env:
+ # used in telemetry, because step names are not programmatically accessible
+ OTEL_SERVICE_NAME: "C++ build"
  GH_TOKEN: ${{ github.token }}
+ # This traceparent is expected to be the value of a step from the calling workflow (e.g. rmm/.github/workflows/pr.yaml.
+ # Ideally, this would be something like conda-cpp-build, which is job that calls a shared workflow.)
+ TRACEPARENT: ${{ steps.job-traceparent.outputs.traceparent }}
  - name: Upload additional artifacts
  if: "!cancelled()"
  run: rapids-upload-artifacts-dir cuda${RAPIDS_CUDA_VERSION%%.*}_$(arch)
+ - name: Telemetry summary
+ id: telemetry-summary
+ if: "always()"
+ uses: rapidsai/shared-actions/telemetry-summarize@main
+ with:
+ endpoint: ${{ inputs.traces_endpoint || inputs.default_endpoint && format('{0}/v1/traces', inputs.default_endpoint) }}
+ # TODO: This is why we need to merge https://github.com/rapidsai/shared-actions/pull/19 before work on shared-workflows can proceed
+ # The build job can't use actions that aren't approved, so we can't test against a personal branch.
+ traceparent: "${{ env.TOP_LEVEL_TRACEPARENT }}"