mainnet-monitor.yml

name: Mainnet Health Monitor

on:
  schedule:
    - cron: "30 */6 * * *"
  workflow_dispatch:
    inputs:
      contracts:
        description: 'Optional: JSON array of {id, name, category, expected_admin} to monitor. Overrides LIVE_MAINNET.md / .deployment-manifest.json.'
        required: false
        default: ""

permissions:
  contents: write
  issues: write

env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true

jobs:
  monitor:
    name: Check deployed mainnet contracts
    runs-on: ubuntu-latest
    env:
      XDG_CONFIG_HOME: /tmp/stellar-xdg
      MAINNET_MONITOR_CONTRACTS: ${{ github.event.inputs.contracts || '' }}

    steps:
      - name: Checkout repository
        uses: actions/checkout@v6
        with:
          fetch-depth: 0

      - name: Install Stellar CLI
        uses: stellar/stellar-cli@v23.0.1

      - name: Register mainnet network + monitor identity
        run: |
          stellar network add mainnet \
            --rpc-url https://mainnet.sorobanrpc.com \
            --network-passphrase "Public Global Stellar Network ; September 2015"
          stellar keys generate mainnet-monitor --network mainnet --overwrite
          stellar keys public-key mainnet-monitor

      - name: Run contract health checks
        id: health
        run: node .github/scripts/mainnet-monitor.js

      - name: Open tracking issues for failures
        if: steps.health.outputs.healthy != 'true'
        uses: actions/github-script@v8
        with:
          script: |
            const fs = require("node:fs");
            const failures = JSON.parse(fs.readFileSync("monitor/failures.json", "utf8"));

            for (const failure of failures) {
              const query = `repo:${context.repo.owner}/${context.repo.repo} is:issue is:open label:mainnet label:bug "${failure.id}"`;
              const existing = await github.rest.search.issuesAndPullRequests({ q: query, per_page: 1 });
              const title = `Mainnet health check failed: ${failure.name}`;
              const body = [
                "The scheduled mainnet health monitor could not verify a deployed contract.",
                "",
                `- Contract: ${failure.name}`,
                `- Address: \`${failure.id}\``,
                `- Health: ${failure.health}`,
                `- Paused: ${failure.paused}`,
                `- Checked at: ${failure.checkedAt}`,
                `- Workflow run: ${context.serverUrl}/${context.repo.owner}/${context.repo.repo}/actions/runs/${context.runId}`,
                "",
                "```text",
                failure.output || "no diagnostics captured",
                "```",
                "",
                failure.paused
                  ? "## Triage\nA circuit breaker (`is_paused`) reported paused. Follow ROLLBACK_PROCEDURE.md immediately and confirm no guarded execution path is active."
                  : "",
              ].join("\n");

              if (existing.data.items.length === 0) {
                await github.rest.issues.create({
                  owner: context.repo.owner,
                  repo: context.repo.repo,
                  title,
                  body,
                  labels: ["mainnet", "bug"],
                });
              } else {
                await github.rest.issues.createComment({
                  owner: context.repo.owner,
                  repo: context.repo.repo,
                  issue_number: existing.data.items[0].number,
                  body,
                });
              }
            }

      - name: Send alert notification (health failure)
        if: steps.health.outputs.healthy != 'true'
        continue-on-error: true
        env:
          ALERT_WEBHOOK_URLS: ${{ secrets.ALERT_WEBHOOK_URLS }}
          ALERT_WEBHOOK_SECRET: ${{ secrets.ALERT_WEBHOOK_SECRET }}
        run: |
          node .github/scripts/alert.js \
            --event "mainnet.health.failed" \
            --title "Mainnet health check failed" \
            --message "One or more deployed mainnet contracts failed their scheduled health check. Tracking issues have been opened." \
            --severity high \
            --field "failed=${{ steps.health.outputs.failures }}/${{ steps.health.outputs.total }}" \
            --field "degraded=${{ steps.health.outputs.degraded }}" \
            --field "paused=${{ steps.health.outputs.paused }}" \
            --field "trigger=mainnet-monitor" \
            --run-url "${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"

      - name: Send alert notification (circuit breaker paused)
        if: steps.health.outputs.paused == 'true'
        continue-on-error: true
        env:
          ALERT_WEBHOOK_URLS: ${{ secrets.ALERT_WEBHOOK_URLS }}
          ALERT_WEBHOOK_SECRET: ${{ secrets.ALERT_WEBHOOK_SECRET }}
        run: |
          node .github/scripts/alert.js \
            --event "mainnet.breaker.paused" \
            --title "Mainnet circuit breaker PAUSED" \
            --message "At least one monitored mainnet contract reports is_paused=true (circuit breaker tripped). Treat as an incident — follow ROLLBACK_PROCEDURE.md and do NOT resume traffic before the pause is resolved." \
            --severity critical \
            --field "paused=${{ steps.health.outputs.paused }}" \
            --field "trigger=mainnet-monitor" \
            --run-url "${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"

      - name: Publish status badge JSON
        if: always() && steps.health.outputs.total != ''
        run: |
          set -euo pipefail

          git config user.name "github-actions[bot]"
          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"

          worktree="$(mktemp -d)"
          if git ls-remote --exit-code --heads origin monitor >/dev/null 2>&1; then
            git fetch origin monitor
            git worktree add "$worktree" origin/monitor
          else
            git worktree add --detach "$worktree"
            git -C "$worktree" checkout --orphan monitor
            git -C "$worktree" rm -rf . >/dev/null 2>&1 || true
          fi

          mkdir -p "$worktree/badges"
          cp monitor/status.json "$worktree/badges/mainnet-monitor.json"

          git -C "$worktree" add badges/mainnet-monitor.json
          if git -C "$worktree" diff --cached --quiet; then
            echo "Badge JSON already up to date."
            exit 0
          fi

          git -C "$worktree" commit -m "chore: update mainnet monitor badge"
          git -C "$worktree" push origin HEAD:monitor

      - name: Report unhealthy contracts
        if: steps.health.outputs.healthy != 'true'
        continue-on-error: true
        run: |
          echo "::warning::One or more mainnet contracts failed their health check. See the opened GitHub issues and alerts for details."
          exit 1