[ci] add cloudwatch metrics for scheduled workflow failures (#419)

.github/workflows/canary-model-zoo.yml

@@ -63,3 +63,10 @@ jobs:
         if: ${{github.event.inputs.download}} == 'huggingface' || ${{github.event.inputs.download}} == 'all'
         working-directory: canary
         run: DJL_ENGINE=tokenizers ./gradlew run -Dmain="ai.djl.canary.ModelZooTest"
+
+  publish-success-metric:
+    needs: [ canary-modelzoo ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJLDemo-CanaryModelZoo-Failure

.github/workflows/canary.yml

@@ -67,3 +67,10 @@ jobs:
       repo-id: ${{ github.events.inputs.repo-id }}
       djl-version: ${{ github.event.inputs.djl-version }}
       pt-version: ${{ github.event.inputs.pt-version }}
+
+  publish-success-metric:
+    needs: [ aarch64-linux, centos, gpu, m1-mac, mac, ubuntu, windows ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJLDemo-Canary-Failure

.github/workflows/nightly.yml

@@ -40,3 +40,9 @@ jobs:
           cd extension && mvn package install && cd ..
           cd extension-example && ./mvnw package
 
+  publish-success-metric:
+    needs: [ compile-test ]
+    if: always()
+    uses: ./.github/workflows/publish-job-success.yml
+    with:
+      metric-name: DJLDemo-NightlyIntegrationTest-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