name: Repair Disk (safe cleanup)

# 根分区 100% 满导致 MongoDB 写卡死。只清「可再生垃圾」：失控日志/pm2日志/journal/npm缓存/旧nginx日志。
# 绝不触碰：数据库文件、用户媒体、/var/www 站点、/var/lib/docker、/data。清完测 DB 恢复，必要时只重启本产品 API。
on:
  workflow_dispatch:

jobs:
  repair:
    runs-on: ubuntu-latest
    steps:
      - name: SSH safe disk cleanup + verify recovery
        uses: appleboy/ssh-action@v1
        with:
          host: ${{ secrets.ECS_HOST }}
          username: ${{ secrets.ECS_USER }}
          password: ${{ secrets.ECS_PASSWORD }}
          command_timeout: 10m
          script: |
            echo "========== 清理前 df =========="; df -h / | tail -1
            echo "========== 1) 截断失控的超大 pm2 stdout 日志(>500M，只清空不删文件) =========="
            for f in ~/.pm2/logs/*.log; do
              [ -f "$f" ] || continue
              sz=$(stat -c%s "$f" 2>/dev/null || echo 0)
              if [ "$sz" -gt 524288000 ]; then echo "  清空 $(du -h "$f" | cut -f1)  $f"; : > "$f"; fi
            done
            echo "========== 2) pm2 flush(清其余各产品日志) =========="
            pm2 flush 2>/dev/null && echo "  pm2 flush done" || echo "  (pm2 flush 跳过)"
            echo "========== 3) journal 收缩到 100M =========="
            journalctl --vacuum-size=100M 2>&1 | tail -2 || echo "  (无 journalctl)"
            echo "========== 4) npm 缓存清理(可再生) =========="
            npm cache clean --force 2>/dev/null && echo "  npm cache cleaned" || echo "  (npm cache 跳过)"
            echo "========== 5) 旧的 rotated nginx 日志(*.gz / *.N) =========="
            find /var/log/nginx -type f \( -name "*.gz" -o -regex ".*\.[0-9]+" \) -delete 2>/dev/null && echo "  旧 nginx 日志已清" || echo "  (无)"
            echo "========== 清理后 df =========="; df -h / | tail -1
            echo "========== 6) MongoDB ping =========="
            if command -v mongosh >/dev/null; then timeout 10 mongosh --quiet --eval "db.adminCommand('ping')" 2>&1 | head -4; \
            elif command -v mongo >/dev/null; then timeout 10 mongo --quiet --eval "db.adminCommand('ping')" 2>&1 | head -4; \
            else echo "  (无 mongo 客户端，跳过 ping)"; fi
            echo "========== 7) 本地读库接口测试(10s) =========="
            curl -s -m 10 -o /dev/null -w "  plaza -> code=%{http_code} time=%{time_total}s\n" http://localhost:3603/api/plaza 2>&1
            CODE=$(curl -s -m 10 -o /dev/null -w "%{http_code}" http://localhost:3603/api/plaza 2>/dev/null)
            if [ "$CODE" != "200" ]; then
              echo "========== 8) plaza 仍非200，重启本产品 API 清连接池 =========="
              pm2 restart digitalme-api --update-env 2>/dev/null | tail -3; sleep 7
              curl -s -m 15 -o /dev/null -w "  plaza(重启后) -> code=%{http_code} time=%{time_total}s\n" http://localhost:3603/api/plaza 2>&1
            else
              echo "  ✓ 读库接口已恢复 200，无需重启"
            fi
            echo "========== 完成 =========="