watchdog.py

# -*- encoding: utf-8 -*-
"""V1.02 监控看门狗 (watchdog)
- 每 30 秒检查 V1.02 监控进程 + ffmpeg 子进程
- 如果 V1.02 退出 → 弹 Windows 原生弹窗提醒（MB_TOPMOST）
- 如果 ffmpeg 全挂 → 弹窗提醒
- 如果 status 文件超过 120 秒没更新 → 弹窗（ffmpeg 可能僵死）
- 自动发现 V1.02 PID（通过命令行过滤）—— 不用硬编码
"""
import sys, io
sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8', errors='replace', line_buffering=True)

import os
import time
import ctypes
import subprocess
import threading

BASE_DIR = os.path.dirname(os.path.abspath(__file__))
STATUS_FILE = os.path.join(BASE_DIR, "recording_status.txt")
DOWNLOADS_DIR = os.path.join(BASE_DIR, "downloads")
V1_02_MARKER = "main_V1.02.py"

CHECK_INTERVAL = 30          # 检查间隔（秒）
STALE_THRESHOLD = 120        # status 文件超过这么多秒没更新 → 警告
FFMPEG_MUST_HAVE = 1          # 至少要有 1 个 ffmpeg 在跑（如果 status 标记 RECORDING）
ALERTED = set()              # 避免重复弹窗（每种异常只弹一次）


def find_v102_pid():
    """自动发现 V1.02 监控进程 PID（命令行含 main_V1.02.py）"""
    try:
        out = subprocess.run(
            ["powershell", "-NoProfile", "-Command",
             "Get-CimInstance Win32_Process -Filter \"Name='python.exe'\" "
             "| Where-Object { $_.CommandLine -like '*main_V1.02*' } "
             "| Select-Object -ExpandProperty ProcessId"],
            capture_output=True, text=True, timeout=10
        )
        for line in out.stdout.strip().splitlines():
            line = line.strip()
            if line.isdigit():
                return int(line)
    except Exception as e:
        print(f"[warn] find_v102_pid: {e}")
    return None


def is_pid_alive(pid):
    """检查 PID 是否还活着"""
    if not pid:
        return False
    try:
        kernel32 = ctypes.windll.kernel32
        PROCESS_QUERY_LIMITED_INFORMATION = 0x1000
        STILL_ACTIVE = 259
        handle = kernel32.OpenProcess(PROCESS_QUERY_LIMITED_INFORMATION, False, pid)
        if not handle:
            return False
        exit_code = ctypes.c_ulong()
        kernel32.GetExitCodeProcess(handle, ctypes.byref(exit_code))
        kernel32.CloseHandle(handle)
        return exit_code.value == STILL_ACTIVE
    except Exception:
        return False


def count_ffmpeg():
    """统计当前在跑的 ffmpeg 进程数"""
    try:
        out = subprocess.run(
            ["tasklist", "/FI", "IMAGENAME eq ffmpeg.exe"],
            capture_output=True, text=True, timeout=10
        )
        return out.stdout.lower().count("ffmpeg.exe")
    except Exception:
        return -1


def get_status_age():
    """返回 status 文件最后修改到现在过了多少秒；不存在返回 -1"""
    if not os.path.exists(STATUS_FILE):
        return -1
    return int(time.time() - os.path.getmtime(STATUS_FILE))


def is_recording_status():
    """status 文件内容是否标记 RECORDING"""
    if not os.path.exists(STATUS_FILE):
        return False
    try:
        with open(STATUS_FILE, "r", encoding="utf-8") as f:
            return f.read().strip().startswith("RECORDING|")
    except Exception:
        return False


def alert_popup(title, message, level="warning"):
    """弹 Windows 原生弹窗（MB_TOPMOST 置顶）"""
    icon = 0x30  # MB_ICONWARNING
    if level == "error":
        icon = 0x10  # MB_ICONERROR
    elif level == "info":
        icon = 0x40  # MB_ICONINFORMATION
    # 0x1000 = MB_TOPMOST
    def _show():
        try:
            ctypes.windll.user32.MessageBoxW(0, message, title, icon | 0x1000)
        except Exception as e:
            print(f"[warn] MessageBox failed: {e}")
    threading.Thread(target=_show, daemon=True).start()
    print(f"\n{'='*60}")
    print(f"  ⚠️  {title}")
    print(f"  {message}")
    print(f"{'='*60}\n")


def fmt_duration(sec):
    sec = max(0, int(sec))
    h, m, s = sec // 3600, (sec % 3600) // 60, sec % 60
    if h:
        return f"{h}h {m}m {s}s"
    return f"{m}m {s}s"


def main():
    print("=" * 60)
    print("  V1.02 Watchdog - 监控看门狗")
    print("=" * 60)
    print(f"  Project: {BASE_DIR}")
    print(f"  Status file: {STATUS_FILE}")
    print(f"  Check interval: {CHECK_INTERVAL}s")
    print(f"  Stale threshold: {STALE_THRESHOLD}s")
    print(f"  Marker: {V1_02_MARKER}")
    print("=" * 60)
    print("  按 Ctrl+C 停止 watchdog")
    print()

    v1_pid = None
    last_pid_check = 0

    try:
        while True:
            now = time.time()
            ts_short = time.strftime("%H:%M:%S")

            # 每 60 秒重新发现 V1.02 PID（V1.02 重启后 PID 会变）
            if now - last_pid_check > 60 or v1_pid is None:
                v1_pid = find_v102_pid()
                last_pid_check = now
                if v1_pid:
                    print(f"[{ts_short}] 发现 V1.02 监控 PID = {v1_pid}")
                else:
                    print(f"[{ts_short}] ⚠️ 找不到 V1.02 监控进程")

            v1_alive = is_pid_alive(v1_pid) if v1_pid else False
            ffmpeg_n = count_ffmpeg()
            status_age = get_status_age()
            recording_now = is_recording_status()

            print(f"[{ts_short}] V1.02={v1_alive}(PID {v1_pid})  ffmpeg={ffmpeg_n}  "
                  f"status_age={status_age}s  recording={recording_now}")

            # 检查 1: V1.02 监控进程退出
            if v1_pid and not v1_alive:
                key = f"v102_dead_{v1_pid}"
                if key not in ALERTED:
                    ALERTED.add(key)
                    alert_popup(
                        "[V1.02] ⚠️ 监控进程已退出",
                        f"V1.02 监控进程 (PID {v1_pid}) 已退出！\n\n"
                        f"录制已中断。\n\n"
                        f"处理建议：\n"
                        f"1. 检查 {STATUS_FILE} 看最后状态\n"
                        f"2. 重新双击 start.bat 启动录制服务\n"
                        f"3. 检查录制文件是否完整 (downloads\\)",
                        level="error"
                    )

            # 检查 2: status 标记 RECORDING 但 ffmpeg 全挂
            if recording_now and ffmpeg_n == 0:
                key = f"ffmpeg_dead_{status_age}"
                if key not in ALERTED:
                    ALERTED.add(key)
                    alert_popup(
                        "[V1.02] ⚠️ 录制异常中断",
                        f"状态文件显示 REC 中，但无 ffmpeg 进程！\n\n"
                        f"可能 ffmpeg crash 或被误杀。\n\n"
                        f"处理建议：\n"
                        f"1. 查看 V1.02 监控窗口日志\n"
                        f"2. 检查 downloads\\ 是否有 .aac 在写入\n"
                        f"3. 必要时重启 start.bat",
                        level="error"
                    )

            # 检查 3: status 文件超过 STALE_THRESHOLD 秒没更新（且 V1.02 活着）
            if v1_alive and status_age > STALE_THRESHOLD and status_age > 0:
                key = f"stale_{int(status_age / 60)}"
                if key not in ALERTED:
                    ALERTED.add(key)
                    alert_popup(
                        "[V1.02] ⚠️ 状态文件过期",
                        f"recording_status.txt 已经 {fmt_duration(status_age)} 没更新了！\n\n"
                        f"V1.02 监控还活着但可能卡住。\n\n"
                        f"处理建议：\n"
                        f"1. 看 V1.02 监控窗口是否有响应\n"
                        f"2. 必要时手动 Ctrl+C 停止 V1.02 然后重启",
                        level="warning"
                    )

            # 检查 4: 完全没找到 V1.02 进程（持续 2 分钟）
            if v1_pid is None and now - last_pid_check > 120:
                key = f"v102_missing_{int(now / 120)}"
                if key not in ALERTED:
                    ALERTED.add(key)
                    alert_popup(
                        "[V1.02] ❌ 监控未启动",
                        f"超过 2 分钟没找到 V1.02 监控进程！\n\n"
                        f"可能 start.bat 没运行，或 V1.02 启动失败。\n\n"
                        f"处理建议：\n"
                        f"1. 打开任务管理器看 python.exe 进程\n"
                        f"2. 双击 start.bat 启动 V1.02",
                        level="error"
                    )

            time.sleep(CHECK_INTERVAL)

    except KeyboardInterrupt:
        print("\n[watchdog] 收到 Ctrl+C，停止")
        return


if __name__ == "__main__":
    main()