peptide-e2e/.gitea/workflows/deploy-health.yml

name: Deploy health (daily drift check)

# Gitea rewrite of .github/workflows/deploy-health.yml. GitHub's version read
# GitHub Deployment objects, which Gitea does not have. This one compares what is
# ACTUALLY on the server: the deploy composite action stamps
# ~/.peptide-deploys/production/<target> with the deployed commit, and this job
# compares that stamp with each repo's main HEAD. Drift (or a missing stamp after
# cutover) opens one issue in peptide-e2e; it never opens duplicates.
on:
  schedule:
    - cron: '0 13 * * *'   # 21:00 SGT, same slot as the GitHub job
  workflow_dispatch: {}

jobs:
  drift:
    name: Compare deployed commit with main
    runs-on: peptide-vps
    steps:
      - uses: actions/checkout@v4
      - name: Check every deployable repo
        shell: bash
        env:
          TOKEN: ${{ secrets.GITHUB_TOKEN }}
          SSH_PRIVATE_KEY: ${{ secrets.SSH_PRIVATE_KEY }}
          SSH_HOST: ${{ secrets.SSH_HOST }}
          SSH_PORT: ${{ secrets.SSH_PORT }}
          SSH_USERNAME: ${{ secrets.SSH_USERNAME }}
          PROD_ENABLED: ${{ vars.PROD_DEPLOY_ENABLED }}
        run: |
          set -euo pipefail
          WORK=$(mktemp -d); trap 'rm -rf "$WORK"' EXIT
          printf '%s\n' "$SSH_PRIVATE_KEY" > "$WORK/key"; chmod 600 "$WORK/key"
          ssh-keyscan -p "$SSH_PORT" "$SSH_HOST" > "$WORK/kh" 2>/dev/null || true
          SSHOPT="-n -i $WORK/key -p $SSH_PORT -o UserKnownHostsFile=$WORK/kh -o StrictHostKeyChecking=accept-new -o ConnectTimeout=30 -o BatchMode=yes"
          API="$GITHUB_SERVER_URL/api/v1"; H="Authorization: token $TOKEN"
          problems=""
          for row in $(python3 -c 'import json; d=json.load(open("tools/gitea-pipeline/repos.json")); [print(k+"="+v["target_path"]) for k,v in d.items() if not k.startswith("_") and v.get("target_path")]'); do
            repo=${row%%=*}; tp=${row#*=}; slug=$(printf '%s' "$tp" | tr '/' '_')
            # The job token is scoped to peptide-e2e; the repos are public, so read
            # their branches anonymously. Unreadable main = a problem, not a crash.
            head=$(curl -s "$API/repos/peptide-repo/$repo/branches/main" | python3 -c 'import json,sys
          try: print(json.load(sys.stdin)["commit"]["id"])
          except Exception: print("")')
            if [ -z "$head" ]; then problems="$problems\n- $repo: could not read main from Gitea"; continue; fi
            stamp=$(ssh $SSHOPT "$SSH_USERNAME@$SSH_HOST" "cat ~/.peptide-deploys/production/$slug 2>/dev/null" || true)
            dep=${stamp%% *}
            if [ -z "$dep" ]; then
              msg="$repo: no Gitea production deploy recorded"
              if [ "$PROD_ENABLED" = "true" ]; then problems="$problems\n- $msg (main is ${head:0:7})"; else echo "::notice::$msg (expected before cutover)"; fi
            elif [ "$dep" != "$head" ]; then
              problems="$problems\n- $repo: production has ${dep:0:7}, main is ${head:0:7}"
            else
              echo "OK $repo ${head:0:7}"
            fi
          done
          [ -z "$problems" ] && { echo "No drift."; exit 0; }
          printf "Drift found:%b\n" "$problems"
          open=$(curl -s -H "$H" "$API/repos/$GITHUB_REPOSITORY/issues?state=open&type=issues&limit=50" | python3 -c 'import json,sys;print(sum(1 for i in json.load(sys.stdin) if i["title"]=="Deploy drift detected"))')
          if [ "$open" -gt 0 ]; then echo "An open drift issue already exists; not opening another."; exit 1; fi
          python3 -c 'import json,sys; print(json.dumps({"title": "Deploy drift detected", "body": "Daily deploy-health found production out of step with main:\n" + sys.argv[1].replace("\\n","\n") + "\n\nRe-run the repo pipeline on main (workflow_dispatch) or investigate the failed deploy. Opened by deploy-health.yml."}))' "$problems" > issue.json
          curl -s -o /dev/null -w "issue HTTP %{http_code}\n" -H "$H" -H 'Content-Type: application/json' -X POST "$API/repos/$GITHUB_REPOSITORY/issues" --data-binary @issue.json
          exit 1