name: Bootstrap nodes health check

# Scheduled liveness check for the preset bootstrap nodes. Runs on a
# GitHub-hosted runner (public internet) so nodes advertising private/cloud
# internal IPs are correctly seen as unreachable. On any unreachable node it
# fails the job and opens/updates a tracking issue labelled `bootstrap-health`.

on:
  schedule:
    - cron: "0 6 * * *" # daily 06:00 UTC
  workflow_dispatch:

env:
  nim_version: v2.2.10

concurrency:
  group: ${{ github.workflow }}
  cancel-in-progress: false

permissions:
  contents: read
  issues: write

jobs:
  ping:
    name: Ping preset bootstrap nodes
    runs-on: ubuntu-latest
    steps:
      - name: Checkout sources
        uses: actions/checkout@v6
        with:
          submodules: recursive

      - name: Setup Nimbus Build System
        uses: ./.github/actions/nimbus-build-system
        with:
          os: linux
          nim_version: ${{ env.nim_version }}

      - name: Ping bootstrap nodes
        id: ping
        continue-on-error: true
        run: make CI=true bootstrapHealthCheck
        shell: bash

      - name: Build report
        id: report
        if: always()
        run: |
          json=build/bootstrap-health-report.json
          if [ ! -f "$json" ]; then
            echo "no_output=true" >> "$GITHUB_OUTPUT"
            echo "::error::check_spr produced no output file"
            exit 0
          fi
          dead=$(jq '[.[] | select(.alive==false)] | length' "$json")
          total=$(jq 'length' "$json")
          echo "dead=$dead" >> "$GITHUB_OUTPUT"
          {
            echo "## Bootstrap node liveness ($((total - dead))/$total reachable)"
            echo
            echo "| Network | Result | Address | Reason |"
            echo "|---|---|---|---|"
            jq -r '.[] | "| \(.network) | \(if .alive then "✅ ALIVE" else "❌ DEAD" end) | \(.address) | \(.reason) |"' "$json"
          } | tee report.md >> "$GITHUB_STEP_SUMMARY"
        shell: bash

      - name: Open or update tracking issue
        if: always() && steps.report.outputs.dead != '0' && steps.report.outputs.dead != ''
        env:
          GH_TOKEN: ${{ github.token }}
          DEAD: ${{ steps.report.outputs.dead }}
          RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
        run: |
          gh label create bootstrap-health --color B60205 \
            --description "Automated bootstrap-node liveness alerts" 2>/dev/null || true
          {
            echo "Scheduled bootstrap-node liveness check found **${DEAD}** unreachable node(s)."
            echo
            echo "Run: ${RUN_URL}"
            echo
            cat report.md
          } > issue-body.md
          existing=$(gh issue list --label bootstrap-health --state open --json number --jq '.[0].number')
          if [ -n "$existing" ]; then
            gh issue comment "$existing" --body-file issue-body.md
          else
            gh issue create --title "Bootstrap nodes unreachable" \
              --label bootstrap-health --body-file issue-body.md
          fi
        shell: bash

      - name: Fail if any node is unreachable
        if: always() && ((steps.report.outputs.dead != '0' && steps.report.outputs.dead != '') || steps.report.outputs.no_output == 'true')
        env:
          DEAD: ${{ steps.report.outputs.dead }}
        run: |
          echo "Bootstrap liveness check failed: ${DEAD} unreachable node(s)."
          exit 1
        shell: bash