#!/usr/bin/env bash
# =============================================================================
# 修复 scense 测试环境访问（9380 端口监听 / 进程存活 / MySQL 通道）
# 修复点（对应 Bug fK06kZRswv12f0_-xSr_1 步骤[2] 自杀退出255）：
#   - 原脚本 set -e：步骤[2] 端口检测失败即自杀退出 255，部署链中断。
#   - 现改为：检测失败 → 记录原因 → 自动重启（systemd 优先，兜底 nohup）→
#     再等 30s 复检；仍失败才输出诊断日志并以非零退出（已完成一次兜底重启）。
# =============================================================================
set -uo pipefail

APP_NAME="scense"
APP_PORT="${SERVER_PORT:-9380}"
WORKDIR="$(cd "$(dirname "$0")/.." && pwd)"
VENV_DIR="$WORKDIR/venv"
ENV_FILE="$WORKDIR/.env"
LOG="$WORKDIR/logs/scense.log"

log() { echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*"; }

[ -f "$ENV_FILE" ] && { set -a; . "$ENV_FILE"; set +a; }

# 步骤[1] 端口监听检测（/dev/tcp 无需额外工具）
check_port() {
  (exec 3<>"/dev/tcp/127.0.0.1/$APP_PORT") 2>/dev/null && { exec 3>&- 3<&-; return 0; }
  return 1
}

# 步骤[2] 进程存活检测
check_proc() {
  pgrep -f "app/$APP_NAME.py" >/dev/null 2>&1
}

# 步骤[3] MySQL 通道检测
check_mysql() {
  command -v mysql >/dev/null 2>&1 || return 1
  mysql --host="${MYSQL_HOST:-127.0.0.1}" --port="${MYSQL_PORT:-3306}" \
    --user="${MYSQL_USER:-root}" --password="${MYSQL_PASSWORD:-}" \
    -e "SELECT 1" "${MYSQL_DATABASE:-scense}" >/dev/null 2>&1
}

# 重启兜底：systemd 优先，失败回退 nohup 直接拉起
restart_app() {
  log "restart $APP_NAME ..."
  if systemctl restart "$APP_NAME" 2>/dev/null; then
    return 0
  fi
  pkill -f "app/$APP_NAME.py" 2>/dev/null || true
  sleep 1
  nohup "$VENV_DIR/bin/python" "$WORKDIR/app/$APP_NAME.py" >> "$LOG" 2>&1 &
  return 0
}

main() {
  local fix_count=0

  if check_port; then
    log "[1] port $APP_PORT listening"
  else
    log "[1] port $APP_PORT NOT listening -> need fix"
    fix_count=$((fix_count+1))
  fi

  if check_proc; then
    log "[2] process app/$APP_NAME.py running"
  else
    log "[2] process app/$APP_NAME.py NOT found -> need fix"
    fix_count=$((fix_count+1))
  fi

  if check_mysql; then
    log "[3] MySQL channel (${MYSQL_DATABASE:-scense}@${MYSQL_HOST:-127.0.0.1}) OK"
  else
    log "[3] MySQL channel unreachable -> need fix"
    fix_count=$((fix_count+1))
  fi

  if [ "$fix_count" -gt 0 ]; then
    log "detected $fix_count issue(s), restart app (bootstrap, no suicide exit)"
    restart_app
    for _ in $(seq 1 30); do
      if check_port && check_proc; then
        log "OK: $APP_NAME back on port $APP_PORT after restart"
        exit 0
      fi
      sleep 1
    done
    log "WARN: $APP_NAME still not listening after restart; last log lines:"
    tail -n 30 "$LOG" 2>/dev/null || true
    exit 1
  fi

  log "OK: scense test env accessible (port=$APP_PORT)"
  exit 0
}

main "$@"