Windows 多服务强制重启监控

multi_service_watchdog.bat

@echo off
setlocal enabledelayedexpansion

:: ====================================================================
:: 多服务强制重启监控（增强版）
:: 兼容：Windows Server 2008 中文系统
:: 功能：
:: 1. Stop Pending 超时自动强杀
:: 2. 自动识别服务 EXE
:: 3. SYSTEM 权限 taskkill（需 PsExec）
:: ====================================================================

:: ====== 配置服务名列表 ======
set serviceList=Spooler RapService

:: ====== 参数 ======
set maxRetries=3
set retryDelay=10
set stopTimeout=30
set pendingThreshold=10

:: ====== 日志 ======
for %%F in ("%~nx0") do set scriptName=%%~nF
set log=%~dp0%scriptName%.log

call :log "多服务强制重启监控启动"

:: ====== 主循环 ======
for %%S in (%serviceList%) do (
    set svcname=%%S
    set tryCount=0

    call :log ""
    call :log "开始处理服务 [!svcname!]"

:retry_loop_%%S
    set /a tryCount+=1

    call :log "尝试停止服务 [!svcname!]..."
    net stop "!svcname!" >> %log% 2>&1

    :: ===== 等待停止（增强版）=====
    set waitTime=0
    set pendingTime=0

:wait_stop_%%S

    set state=
    for /f "tokens=3 delims=: " %%a in ('sc query "!svcname!" ^| findstr "STATE"') do set state=%%a

    if /i "!state!"=="STOPPED" (
        call :log "服务 [!svcname!] 成功停止"
        goto start_service_%%S
    )

    if /i "!state!"=="STOP_PENDING" (
        set /a pendingTime+=1
    ) else (
        set pendingTime=0
    )

    set /a waitTime+=1

    if !pendingTime! GEQ %pendingThreshold% (
        call :log "服务 [!svcname!] 长时间 STOP_PENDING，强制终止"
        goto force_kill_%%S
    )

    if !waitTime! GEQ %stopTimeout% (
        call :log "服务 [!svcname!] 停止超时"
        goto force_kill_%%S
    )

    timeout /t 1 /nobreak >nul
    goto wait_stop_%%S

:start_service_%%S
    call :log "尝试启动服务 [!svcname!]..."
    net start "!svcname!" >> %log% 2>&1

    timeout /t %retryDelay% /nobreak >nul

    set state=
    for /f "tokens=3 delims=: " %%a in ('sc query "!svcname!" ^| findstr "STATE"') do set state=%%a

    if /i not "!state!"=="RUNNING" (
        call :log "服务 [!svcname!] 第 !tryCount! 次启动失败"

        if !tryCount! GEQ %maxRetries% (
            call :log "服务 [!svcname!] 重启失败超过 %maxRetries% 次，准备重启服务器"
            goto reboot_server
        ) else (
            call :log "等待 %retryDelay% 秒后重试..."
            timeout /t %retryDelay% /nobreak >nul
            goto retry_loop_%%S
        )
    ) else (
        call :log "服务 [!svcname!] 启动成功"
    )

    call :log "服务 [!svcname!] 处理完成"
)

call :log "多服务强制重启监控结束"
goto end

:: ====== 强制杀进程（增强版）======
:force_kill_%%S

call :get_exe_name "!svcname!"

call :log "检测到服务进程名: !exeName!"

:: 杀 PID
for /f "tokens=2 delims=:" %%a in ('sc queryex "!svcname!" ^| findstr "PID"') do set pid=%%a
set pid=!pid: =!

if not "!pid!"=="" (
    call :log "SYSTEM 权限 taskkill PID !pid!"
    psexec -accepteula -s -d taskkill /PID !pid! /F >> %log% 2>&1
) else (
    call :log "未获取到 PID"
)

:: 按 EXE 名兜底
if not "!exeName!"=="" (
    call :log "SYSTEM 权限 taskkill !exeName!"
    psexec -accepteula -s -d taskkill /IM "!exeName!" /F >> %log% 2>&1
) else (
    call :log "未获取 EXE 名"
)

timeout /t 3 /nobreak >nul
goto retry_loop_%%S

:: ====== 获取 EXE 名 ======
:get_exe_name
setlocal enabledelayedexpansion

set exeName=
set fullpath=

for /f "tokens=2,* delims=:" %%a in ('sc qc "%~1" ^| findstr /i "BINARY_PATH_NAME"') do (
    set fullpath=%%b
)

if defined fullpath (
    set fullpath=!fullpath:~1!
    for %%f in ("!fullpath!") do set exeName=%%~nxf
)

endlocal & set exeName=%exeName%
goto :eof

:: ====== 重启服务器 ======
:reboot_server
call :log "等待 30 秒后重启服务器"
echo 30 秒后重启服务器
@rem timeout /t 30 /nobreak >nul
@rem shutdown /r /f /t 0
goto end

:: ====== 日志函数 ======
:log
setlocal enabledelayedexpansion
for /f "tokens=1-2 delims=." %%a in ("%time%") do set logtime=%%a
echo === [%date% !logtime!] %~1 >> %log%
endlocal & goto :eof

:end
endlocal