#!/bin/bash
# VPS 常驻进程滚动重启（sanguo-live-supervisor + sanguo-shadow-desk）。
#
# 为什么需要：vps-deploy 只推代码+重启 sanguo-api；supervisor/影子柜台是常驻
# python 进程，不重启就一直跑旧代码（2026-08-15 实录：部署后 16h 旧进程原样跑）。
#
# 用法：
#   bash scripts/nas_sync/vps_restart_resident.sh          # 直接执行（Mac runner / 本机）
#
# 盘中保护：北京时间 工作日 09:15-15:30 跳过重启（盘中杀树丢撮合现场），
# 打印 SKIP_RESIDENT_RESTART 并 exit 0 —— 代码已就位，下次非盘中部署/手动补跑生效。
set -u
VPS=49.232.102.198
SSH="ssh -o ControlPath=none -o ConnectTimeout=20 $VPS"

# ---- 1) 盘中保护 ----
BJ_HM=$(TZ=Asia/Shanghai date +%H%M)
BJ_DOW=$(TZ=Asia/Shanghai date +%u)
if [ "$BJ_DOW" -le 5 ] && [ "$BJ_HM" -ge 0915 ] && [ "$BJ_HM" -le 1530 ]; then
  echo "SKIP_RESIDENT_RESTART 北京时间盘中(周$BJ_DOW $BJ_HM)——supervisor/影子保持运行,下次非盘中部署生效"
  exit 0
fi

# ---- 2) 查 supervisor/影子 python 进程树根（动态查 PID：账户号会变）----
PIDS=$($SSH "wmic process where \"name='python.exe' and (commandline like '%sanguo_live%supervisor%' or commandline like '%sanguo_trader.shadow%')\" get processid /format:csv" 2>/dev/null \
  | tr -d '\r' | grep -oE '[0-9]+$' | sort -u || true)
echo "重启前进程: $(echo $PIDS | tr '\n' ' ')"

# ---- 3) 杀全树（/T 带走 runner_live / shadow --account 子进程）----
for pid in $PIDS; do
  $SSH "taskkill /F /T /PID $pid" 2>/dev/null || echo "  (PID $pid 已自行退出)"
done
# schtask 收尸（壳可能残留 running 态，end 掉才好 /run）
$SSH "schtasks /end /tn sanguo-live-supervisor" 2>/dev/null || true
$SSH "schtasks /end /tn sanguo-shadow-desk" 2>/dev/null || true

# ---- 4) 重拉 ----
sleep 3
$SSH "schtasks /run /tn sanguo-live-supervisor" || { echo "❌ supervisor 拉起失败"; exit 1; }
$SSH "schtasks /run /tn sanguo-shadow-desk" || { echo "❌ shadow-desk 拉起失败"; exit 1; }

# ---- 5) 等进程回来（内存实测 40-60s 起齐，给 120s 余量）----
for i in $(seq 1 12); do
  sleep 10
  GOT=$($SSH "wmic process where \"name='python.exe' and (commandline like '%sanguo_live%supervisor%' or commandline like '%sanguo_trader.shadow%')\" get processid /format:csv" 2>/dev/null \
    | tr -d '\r' | grep -cE '[0-9]+$' || true)
  echo "  ${i}0s: 已起 $GOT 个进程"
  if [ "${GOT:-0}" -ge 2 ]; then
    echo "RESIDENT_RESTART_DONE supervisor+影子已重启并吃上新代码"
    exit 0
  fi
done
echo "❌ 120s 内常驻进程未起齐(期望≥2: supervisor+shadow auto)"
exit 1