[ci] add cloudwatch metrics for scheduled workflow failures (#2966)

.github/workflows/codeql-analysis-java.yml

@@ -73,3 +73,10 @@ jobs:
 
       - name: Perform CodeQL Analysis
         uses: github/codeql-action/analyze@v2
+
+  publish-success-metric:
+    needs: [ analyze ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-CodeQL-Failure

.github/workflows/docker_publish.yml

@@ -56,4 +56,11 @@ jobs:
           context: .
           file: docker/spark/Dockerfile
           build-args: DJL_VERSION=${DJL_VERSION}
-          tags: deepjavalibrary/djl-spark:${{ env.DJL_VERSION }}-cpu
+          tags: deepjavalibrary/djl-spark:${{ env.DJL_VERSION }}-cpu
+
+  publish-success-metric:
+    needs: [ publish ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-SparkDockerPublish-Failure

.github/workflows/docs.yml

@@ -65,3 +65,10 @@ jobs:
         run: |
           aws s3 sync ../site s3://djl-ai/documentation/nightly --delete
           aws cloudfront create-invalidation --distribution-id E733IIDCG0G5U --paths "/*"
+
+  publish-success-metric:
+    needs: [ documentation ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-DocumentationPublish-Failure

.github/workflows/native_jni_s3_paddle.yml

@@ -108,3 +108,10 @@ jobs:
           PADDLE_VERSION="$(cat gradle.properties | awk -F '=' '/paddlepaddle_version/ {print $2}')"
           aws s3 sync jnilib s3://djl-ai/publish/paddlepaddle-${PADDLE_VERSION}/jnilib
           aws cloudfront create-invalidation --distribution-id E371VB8JQ6NRVY --paths "/paddlepaddle-${PADDLE_VERSION}/jnilib*"
+
+  publish-success-metric:
+    needs: [ publish ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-NativeJNIPaddleS3Publish-Failure

.github/workflows/native_jni_s3_pytorch.yml

@@ -306,3 +306,10 @@ jobs:
           cd /home/ubuntu/djl_benchmark_script/scripts
           instance_id=${{ needs.create-aarch64-runner.outputs.aarch64_instance_id }}
           ./stop_instance.sh $instance_id
+
+  publish-success-metric:
+    needs: [ build-pytorch-jni-macos, build-pytorch-jni-linux, build-pytorch-jni-precxx11, build-pytorch-jni-windows, build-pytorch-jni-arm64-macos, stop-runners ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-NativeJNIPytorchS3Publish-Failure

.github/workflows/native_jni_s3_pytorch_android.yml

@@ -43,3 +43,10 @@ jobs:
           PYTORCH_VERSION=${PYTORCH_VERSION:-$(cat gradle.properties | awk -F '=' '/pytorch_version/ {print $2}')}
           aws s3 sync engines/pytorch/pytorch-native/jnilib s3://djl-ai/publish/pytorch/${PYTORCH_VERSION}/jnilib
           aws cloudfront create-invalidation --distribution-id E371VB8JQ6NRVY --paths "/pytorch/${PYTORCH_VERSION}/jnilib*"
+
+  publish-success-metric:
+    needs: [ build-pytorch-jni-android ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-NativeJNIPytorchAndroidS3Publish-Failure

.github/workflows/nightly_android.yml

@@ -31,3 +31,10 @@ jobs:
           emulator-options: -no-snapshot-save -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none
           disable-animations: true
           script: cd android/core && ./gradlew cAT
+
+  publish-success-metric:
+    needs: [ build ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-AndroidIntegrationTests-Failure

.github/workflows/nightly_publish.yml

@@ -255,3 +255,10 @@ jobs:
           cd /home/ubuntu/djl_benchmark_script/scripts
           instance_id=${{ needs.create-runners.outputs.gpu_1_instance_id }}
           ./stop_instance.sh $instance_id
+
+  publish-success-metric:
+    needs: [ publish, stop-runners ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-NightlyIntegrationTestsPublish-Failure

.github/workflows/publish-job-success.yml

@@ -0,0 +1,27 @@
+name: Publish Job Success Metric to CloudWatch
+
+on:
+  workflow_call:
+    inputs:
+      metric-name:
+        description: "The name of the job to publish a metric for"
+        type: string
+        required: true
+
+jobs:
+  publish-job-success-to-cloudwatch:
+    if: ${{ github.event_name == 'schedule' }}
+    runs-on: [ self-hosted, scheduler ]
+    steps:
+      - name: Configure AWS Credentials
+        uses: aws-actions/configure-aws-credentials@v2
+        with:
+          aws-region: us-west-2
+      - name: Publish Job Success Metric
+        run: |
+          [[ ${{ job.status }} == "success" ]]
+          failedBuild=$?
+          aws cloudwatch put-metric-data --namespace GithubCI \
+            --metric-name ${{ inputs.metric-name }} \
+            --value $failedBuild \
+            --unit Count

.github/workflows/publish_android_packages.yml

@@ -50,3 +50,10 @@ jobs:
           ORG_GRADLE_PROJECT_signingPassword: ${{ secrets.ORG_GRADLE_PROJECT_signingPassword }}
           ORG_GRADLE_PROJECT_ossrhUsername: ${{ secrets.ORG_GRADLE_PROJECT_ossrhUsername }}
           ORG_GRADLE_PROJECT_ossrhPassword: ${{ secrets.ORG_GRADLE_PROJECT_ossrhPassword }}
+
+  publish-success-metric:
+    needs: [ release-android ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-AndroidPublish-Failurei

.github/workflows/serving_publish.yml

@@ -105,3 +105,10 @@ jobs:
           ORG_GRADLE_PROJECT_ossrhUsername: ${{ secrets.ORG_GRADLE_PROJECT_ossrhUsername }}
           ORG_GRADLE_PROJECT_ossrhPassword: ${{ secrets.ORG_GRADLE_PROJECT_ossrhPassword }}
           DJL_STAGING: ${{ github.event.inputs.repo-id }}
+
+  publish-success-metric:
+    needs: [ publish ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJL-DJLServingPublish-Failure