adapt to latest workflows

rapidsai · Nov 1, 2024 · eeac5b0 · eeac5b0
1 parent 70c386e
commit eeac5b0
Showing 1 changed file with 142 additions and 113 deletions.
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
@@ -10,59 +10,52 @@ concurrency:
   cancel-in-progress: true
 
 env:
-  # TODO: put this in a shared org-wide secret?
   OTEL_SERVICE_NAME: 'pr-cuml'
-  # TODO: this should be set as an org-wide variable
   OTEL_EXPORTER_OTLP_ENDPOINT: https://tempo.gha-runners.nvidia.com:4318
-  # These are where the secrets in github env vars are written to files. These files don't
-  # exist unless you explicitly write them in a step.
-  # The purpose of setting the environment variable is to tell OpenTelemetry tools where to find them.
-  #     We abuse it a bit by also using it as the write destination for the certificate files.
-  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"
   OTEL_EXPORTER_OTLP_PROTOCOL: "http/protobuf"
-  OTEL_EXPORTER_OTLP_HEADERS: ${{ secrets.OTEL_EXPORTER_OTLP_HEADERS }}
+  OTEL_RESOURCE_ATTRIBUTES: "git.repository=${{github.repository}},git.ref=${{github.ref}},git.sha=${{github.sha}},git.job_url=${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+  # Set these to point the shared-actions clones at a branch on the rapidsai/shared-actions repo
+  SHARED_ACTIONS_REPO: "rapidsai/shared-actions"
+  SHARED_ACTIONS_REF: "main"
 
 jobs:
-  telemetry-setup:
+  reexports:
+    # Re-export secrets and env vars to make them available to shared actions and workflows
+    # You can't use env.SOMETHING in a "with:" section when calling a shared workflow.
     runs-on: ubuntu-latest
     outputs:
-      start_time: ${{ steps.timestamp.outputs.START_TIME }}
-      traceparent: ${{ steps.telemetry-setup.outputs.traceparent }}
-      endpoint: ${{ steps.var-reexports.outputs.endpoint }}
-      top_level_service_name: ${{ steps.var-reexports.outputs.service_name }}
+        endpoint: "${{steps.reexport.outputs.endpoint}}"
+        service_name: "${{steps.reexport.outputs.service_name}}"
+        otel_resource_attributes: "${{steps.reexport.outputs.otel_resource_attributes}}"
+        shared_actions_repo: "${{steps.reexport.outputs.shared_actions_repo}}"
+        shared_actions_ref: "${{steps.reexport.outputs.shared_actions_ref}}"
     steps:
-      - name: Get starting timestamp
-        id: timestamp
-        run:
-            echo "START_TIME=$(date +%s.%N)" >> ${GITHUB_OUTPUT}
-      - name: Echo endpoint to make it available to shared workflows
-        id: var-reexports
-        run: |
-          echo endpoint="${OTEL_EXPORTER_OTLP_ENDPOINT}" >> ${GITHUB_OUTPUT}
-          echo service_name="${OTEL_SERVICE_NAME}" >> ${GITHUB_OUTPUT}
-      - name: Write certificate files for mTLS
-        run: |
-          mkdir -p /tmp/certs
-          cat << EOF > "${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
-      - name: Telemetry setup
-        id: telemetry-setup
-        uses: rapidsai/shared-actions/telemetry-traceparent@add-telemetry
-      - name: Start root span
-        uses: rapidsai/shared-actions/telemetry-create-span@add-telemetry
-        with:
-            name: "root span"
-            traceparent: ${{steps.telemetry-setup.outputs.traceparent}}
-            start_time: ${{steps.timestamp.outputs.start_time}}
+        - id: reexport
+          # NOTE: certs are base-64 encoded so that they're easier to handle here
+          run: |
+            echo endpoint="${OTEL_EXPORTER_OTLP_ENDPOINT}" >> ${GITHUB_OUTPUT}
+            echo service_name="${OTEL_SERVICE_NAME}" >> ${GITHUB_OUTPUT}
+            echo otel_resource_attributes="${OTEL_RESOURCE_ATTRIBUTES}" >> ${GITHUB_OUTPUT}
+            echo shared_actions_ref="${SHARED_ACTIONS_REF}" >> ${GITHUB_OUTPUT}
+            echo shared_actions_repo="${SHARED_ACTIONS_REPO}" >> ${GITHUB_OUTPUT}
+
+  top-level-telemetry-traceparent:
+    runs-on: ubuntu-latest
+    needs:
+        - reexports
+    outputs:
+      traceparent: "${{steps.traceparent.outputs.traceparent}}"
+    env:
+        OTEL_SERVICE_NAME: ${{needs.reexports.outputs.service_name}}
+    steps:
+        - name: Clone shared-actions repo
+          uses: actions/checkout@v4
+          with:
+            repository: ${{env.SHARED_ACTIONS_REPO}}
+            ref: ${{env.SHARED_ACTIONS_REF}}
+            path: ./shared-actions
+        - id: traceparent
+          uses: ./shared-actions/telemetry-traceparent
   pr-builder:
     needs:
       - changed-files
@@ -76,10 +69,11 @@ jobs:
       - conda-python-tests-dask
       - conda-notebook-tests
       - docs-build
-      - telemetry-setup
       - wheel-build-cuml
       - wheel-tests-cuml
       - devcontainer
+      - reexports
+      - top-level-telemetry-traceparent
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@add-telemetry
     if: always()
@@ -120,17 +114,20 @@ jobs:
           - '!thirdparty/LICENSES/**'
   checks:
     secrets: inherit
-    needs: telemetry-setup
+    needs: [reexports, top-level-telemetry-traceparent]
     uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@add-telemetry
     with:
       enable_check_generated_files: false
       ignored_pr_jobs: >-
         optional-job-conda-python-tests-cudf-pandas-integration
         final-telemetry-update
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   clang-tidy:
-    needs: [checks, telemetry-setup]
+    needs: [checks, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@add-telemetry
     with:
@@ -139,74 +136,95 @@ jobs:
       arch: "amd64"
       container_image: "rapidsai/ci-conda:cuda11.8.0-ubuntu22.04-py3.10"
       run_script: "ci/run_clang_tidy.sh"
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   conda-cpp-build:
-    needs: [checks, telemetry-setup]
+    needs: [checks, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-build.yaml@add-telemetry
     with:
       build_type: pull-request
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   conda-cpp-tests:
-    needs: [conda-cpp-build, changed-files, telemetry-setup]
+    needs: [conda-cpp-build, changed-files, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-tests.yaml@add-telemetry
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_cpp
     with:
       build_type: pull-request
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   conda-cpp-checks:
-    needs: [conda-cpp-build, telemetry-setup]
+    needs: [conda-cpp-build, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-post-build-checks.yaml@add-telemetry
     with:
       build_type: pull-request
       enable_check_symbols: true
       symbol_exclusions: raft_cutlass
   conda-python-build:
-    needs: [conda-cpp-build, telemetry-setup]
+    needs: [conda-cpp-build, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-python-build.yaml@add-telemetry
     with:
       build_type: pull-request
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   conda-python-tests-singlegpu:
-    needs: [conda-python-build, changed-files, telemetry-setup]
+    needs: [conda-python-build, changed-files, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@add-telemetry
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python
     with:
       build_type: pull-request
       script: "ci/test_python_singlegpu.sh"
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   optional-job-conda-python-tests-cudf-pandas-integration:
-    needs: [conda-python-build, changed-files, telemetry-setup]
+    needs: [conda-python-build, changed-files, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@add-telemetry
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python
     with:
       matrix_filter: map(select(.ARCH == "amd64"))
       build_type: pull-request
       script: "ci/test_python_integration.sh"
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   conda-python-tests-dask:
-    needs: [conda-python-build, changed-files, telemetry-setup]
+    needs: [conda-python-build, changed-files, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@add-telemetry
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python
     with:
       build_type: pull-request
       script: "ci/test_python_dask.sh"
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   conda-notebook-tests:
-    needs: [conda-python-build, changed-files, telemetry-setup]
+    needs: [conda-python-build, changed-files, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@add-telemetry
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_notebooks
@@ -216,10 +234,13 @@ jobs:
       arch: "amd64"
       container_image: "rapidsai/ci-conda:latest"
       run_script: "ci/test_notebooks.sh"
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   docs-build:
-    needs: [conda-python-build, telemetry-setup]
+    needs: [conda-python-build, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@add-telemetry
     with:
@@ -228,10 +249,13 @@ jobs:
       arch: "amd64"
       container_image: "rapidsai/ci-conda:latest"
       run_script: "ci/build_docs.sh"
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   wheel-build-cuml:
-    needs: [checks, telemetry-setup]
+    needs: [checks, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@add-telemetry
     with:
@@ -240,59 +264,64 @@ jobs:
       extra-repo: rapidsai/cumlprims_mg
       extra-repo-sha: branch-24.12
       extra-repo-deploy-key: CUMLPRIMS_SSH_PRIVATE_DEPLOY_KEY
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   wheel-tests-cuml:
-    needs: [wheel-build-cuml, changed-files, telemetry-setup]
+    needs: [wheel-build-cuml, changed-files, reexports, top-level-telemetry-traceparent]
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@add-telemetry
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python
     with:
       build_type: pull-request
       script: ci/test_wheel.sh
-      default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-      traceparent: ${{ needs.telemetry-setup.outputs.traceparent }}
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
   devcontainer:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/build-in-devcontainer.yaml@add-telemetry
     with:
       arch: '["amd64"]'
       cuda: '["12.5"]'
       extra-repo-deploy-key: CUMLPRIMS_SSH_PRIVATE_DEPLOY_KEY
+      default_endpoint: "${{needs.reexports.outputs.endpoint}}"
+      otel_resource_attributes: "${{needs.reexports.outputs.otel_resource_attributes}}"
+      traceparent: ${{ needs.top-level-telemetry-traceparent.outputs.traceparent }}
+      shared_actions_repo: ${{needs.reexports.outputs.shared_actions_repo}}
+      shared_actions_ref: ${{needs.reexports.outputs.shared_actions_ref}}
       build_command: |
         sccache -z;
         build-all --verbose;
         sccache -s;
   final-telemetry-update:
     runs-on: ubuntu-latest
-    needs: [pr-builder, telemetry-setup]
+    needs:
+        - reexports
+        - top-level-telemetry-traceparent
+        - pr-builder
     steps:
-      - name: Get final timestamp
-        id: timestamp
-        run:
-            echo "FINAL_TIME=$(date +%s.%N)" >> ${GITHUB_OUTPUT}
-        # Main purpose of this traceparent line here is to ensure that otel-cli is installed.
-      - name: Get job traceparent
-        uses: rapidsai/shared-actions/telemetry-traceparent@add-telemetry
-      - 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
-      - name: Update root span with final completion time
-        if: always()
-        uses: rapidsai/shared-actions/telemetry-create-span@add-telemetry
-        with:
-            service: ${{needs.telemetry-setup.outputs.top_level_service_name}}
-            name: "end-of-job update"
-            default_endpoint: "${{needs.telemetry-setup.outputs.endpoint}}"
-            traceparent: ${{needs.telemetry-setup.outputs.traceparent}}
-            start_time: ${{needs.telemetry-setup.outputs.start_time}}
-            end_time: ${{steps.timestamp.outputs.FINAL_TIME}}
+        - name: Clone shared-actions repo
+          uses: actions/checkout@v4
+          with:
+            repository: rapidsai/shared-actions
+            ref: ${{needs.reexports.outputs.shared_actions_ref}}
+            path: ./shared-actions
+        - name: Send summary
+          uses: ./shared-actions/telemetry-summarize
+          with:
+            traceparent: ${{needs.top-level-telemetry-traceparent.outputs.traceparent}}
+            ca_cert: "${{secrets.OTEL_EXPORTER_OTLP_CA_CERTIFICATE}}"
+            client_cert: "${{secrets.OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE}}"
+            client_key: "${{secrets.OTEL_EXPORTER_OTLP_CLIENT_KEY}}"
+        - name: Send root span with final time
+          uses: ./shared-actions/telemetry-finalize-root-span
+          with:
+            traceparent: ${{needs.top-level-telemetry-traceparent.outputs.traceparent}}
+            ca_cert: "${{secrets.OTEL_EXPORTER_OTLP_CA_CERTIFICATE}}"
+            client_cert: "${{secrets.OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE}}"
+            client_key: "${{secrets.OTEL_EXPORTER_OTLP_CLIENT_KEY}}"