check-drivers-failures.yml

name: Check driver failures

on:
  workflow_call:
    inputs:
      logs-artifact:
        required: true
        type: string
        description: |
          Artifact name where driver build logs have been stored.

jobs:
  check-failures:
    runs-on: ubuntu-24.04
    steps:
      - uses: actions/checkout@v4

      - name: Restore failure files
        uses: actions/download-artifact@v4
        with:
          pattern: ${{ inputs.logs-artifact }}
          merge-multiple: true
          path: /tmp/FAILURES

      - name: Check build failures
        run: |
          set -eo pipefail

          FAILURES_DIR=${FAILURES_DIR:-/tmp/FAILURES}

          shopt -s nullglob
          cd "$FAILURES_DIR"
          failure_files=(*/*/*.log)

          for failure_file in "${failure_files[@]}"; do
              if [[ "$failure_file" =~ ^([^/]+)/([^/]+)/([^/]+)\.log$ ]]; then
                  # If the file is empty, there's nothing to alert on. Sometimes kernels
                  # that don't support eBPF leave the error file hanging around for no
                  # good reason (I suspect some shenanigan's with tee spawning after we
                  # mark the file for deletion).
                  if [[ ! -s "$failure_file" ]]; then
                      rm -f "$failure_file"
                      continue
                  fi

                  kernel_version="${BASH_REMATCH[1]}"
                  module_version="${BASH_REMATCH[2]}"
                  probe_type="${BASH_REMATCH[3]}"
                  echo >&2 "============================================================================"
                  echo >&2 "Failed to build ${probe_type} probe"
                  echo >&2 "Module version: ${module_version}"
                  echo >&2 "Kernel version: ${kernel_version}"
                  echo >&2
                  cat >&2 "$failure_file"
                  echo >&2
                  echo >&2
              fi
          done

          # We expand it again to ignore empty files.
          failure_files=(*/*/*.log)
          [[ "${#failure_files[@]}" == 0 ]]

      - name: Slack notification
        if: failure() && github.event_name != 'pull_request'
        uses: rtCamp/action-slack-notify@v2
        env:
          SLACK_WEBHOOK: ${{ secrets.SLACK_COLLECTOR_ONCALL_WEBHOOK }}
          SLACK_CHANNEL: team-acs-collector-oncall
          SLACK_COLOR: ${{ job.status }}
          SLACK_LINK_NAMES: true
          SLACK_TITLE: Driver build check failed
          MSG_MINIMAL: actions url,commit
          SLACK_MESSAGE: |
            @acs-collector-oncall