ci: Send team alerts on specific keywords (#10986)

* ci: Send team alerts on specific keywords Signed-off-by: Oliver Koenig <[email protected]> * f Signed-off-by: Oliver Koenig <[email protected]> --------- Signed-off-by: Oliver Koenig <[email protected]>
NVIDIA · Oct 27, 2024 · 0c5f5fb · 0c5f5fb
1 parent 8fcd5cb
commit 0c5f5fb
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 5 deletions.
diff --git a/.github/workflows/_test_template.yml b/.github/workflows/_test_template.yml
@@ -33,13 +33,17 @@ on:
       log:
         description: Last 2000 characters of the test step's log
         value: ${{ jobs.main.outputs.log }} 
+      potential_infra_failure:
+        description: Boolean flag when infra-related keyword spotted in logs.
+        value: ${{ jobs.main.outputs.potential_infra_failure }}
 jobs:
 
   main:
     runs-on: ${{ inputs.RUNNER }} 
     outputs:
       conclusion: ${{ steps.main.conclusion }}
       log: ${{ steps.main.outputs.log }}
+      potential_infra_failure: ${{ steps.main.outputs.potential_infra_failure }}
     steps:
         - name: Docker system cleanup
           run: |
@@ -75,6 +79,9 @@ jobs:
             
             echo "log=$(tail -c 2000 err.log |  base64 -w 0)" >> "$GITHUB_OUTPUT"
             
+            potential_infra_failure=$(cat err.log | grep -Eqi "gpu|cuda|device" && echo true || echo false)
+            echo "potential_infra_failure=$potential_infra_failure" >> "$GITHUB_OUTPUT"
+
             exit $EXIT_CODE
             
         - uses: "NVIDIA/NeMo/.github/actions/cancel-workflow@main"

diff --git a/.github/workflows/cicd-main.yml b/.github/workflows/cicd-main.yml
@@ -4515,7 +4515,10 @@ jobs:
         if: ${{ always() && steps.pipeline-conclusion.outputs.FAILED == 'true' && env.SLACK_WEBHOOK != '' }}
         env: 
           SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+          SLACK_WEBHOOK_ADMIN: <!subteam^${{ secrets.SLACK_WEBHOOK_ADMIN }}>
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_ACTOR: ${{ github.actor }}
+          BRANCH: ${{ github.head_ref || github.ref_name }}
           REPOSITORY: ${{ github.repository }}
           RUN_ID: ${{ github.run_id }}
           PR_NUMBER: ${{ github.event.number }}
@@ -4571,13 +4574,15 @@ jobs:
             echo "* [$JOB_NAME]($JOB_URL)" | tee -a $GITHUB_STEP_SUMMARY
 
             LOGS=$(echo $JOB | yq '(.value.outputs.log | @base64d)' | tr -d '"')
+            LOGS=$([[ $(echo $LOGS | wc -c) -gt 0 ]] && echo -E "\`\`\`\n$LOGS\n\`\`\`" || echo "")
+            LOGS=$([[ $(echo $JOB | yq '.value.outputs.potential_infra_failure') == "true" ]] && echo -E "$LOGS\n\ncc: $SLACK_WEBHOOK_ADMIN" || echo -E "$LOGS")
             
             SUMMARY=$(echo "$SUMMARY" | jq \
               --arg pr "<$PR_URL|$PR_TITLE>" \
               --arg job "<$JOB_URL|$JOB_NAME>" \
-              --arg logs "$LOGS" \
-              --arg author "<https://github.com/${{ github.actor }}|${{ github.actor }}>" \
-              --arg branch "<https://github.com/$REPOSITORY/tree/${{ github.head_ref || github.ref_name }}|${{ github.head_ref || github.ref_name }}>"\
+              --arg logs "$(echo -e "$LOGS")" \
+              --arg author "<https://github.com/$GITHUB_ACTOR|$GITHUB_ACTOR>" \
+              --arg branch "<https://github.com/$REPOSITORY/tree/$BRANCH|$BRANCH>"\
               '. += [
               {
                 "type": "section",
@@ -4588,8 +4593,7 @@ jobs:
                     + "\nJob: " + $job
                     + "\nAuthor: " + $author
                     + "\nBranch: " + $branch
-                    + "\nLogs:" 
-                    + "```\n" + $logs + "\n```" 
+                    + "\nLogs:" + $logs
                   )
                 }
               }