脚本 service_check_admin.py
import requests
import subprocess
import sys
import os
import logging
from time import sleep# 配置日志 - 记录到文件以便后台运行时查看
LOG_FILE = "service_monitor1.log"
logging.basicConfig(level=logging.INFO,format='%(asctime)s - %(levelname)s - %(message)s',handlers=[logging.FileHandler(LOG_FILE),logging.StreamHandler(sys.stdout) # 保留控制台输出（后台运行时不可见）]
)
logger = logging.getLogger(__name__)# 服务配置
SERVICE_URLS = ["http://127.0.0.1:8089/index.html"# "http://127.0.0.1:9999"
]
JAR_COMMAND = r"java -jar E:\qaml\datax\dataxweb\datax-admin-2.1.2.jar" # 修改为你的JAR路径
# JAR_COMMAND = r"java -jar E:\qaml\datax\dataxweb\datax-executor-2.1.2.jar"
TIMEOUT = 10 # 请求超时时间（秒）
MAX_RETRIES = 2 # 检测失败时的重试次数
RETRY_DELAY = 5 # 重试间隔（秒）def check_service(url):"""检查单个服务的可用性"""try:response = requests.get(url, timeout=TIMEOUT)if response.status_code == 200:logger.info(f"服务正常: {url}")return Trueelse:logger.warning(f"服务响应异常: {url} (状态码: {response.status_code})")return Falseexcept requests.exceptions.RequestException as e:logger.error(f"服务连接失败: {url} - {str(e)}")return Falsedef check_all_services():"""检查所有服务的可用性（带重试机制）"""logger.info("开始服务健康检查...")for attempt in range(1, MAX_RETRIES + 1):all_ok = Truefor url in SERVICE_URLS:if not check_service(url):all_ok = Falsebreak # 发现一个失败就停止检查其他URLif all_ok:logger.info(f"所有服务运行正常 (尝试 {attempt}/{MAX_RETRIES})")return Trueif attempt < MAX_RETRIES:logger.warning(f"检测失败，{RETRY_DELAY}秒后重试 ({attempt}/{MAX_RETRIES})")sleep(RETRY_DELAY)logger.error(f"所有 {MAX_RETRIES} 次尝试均失败")return Falsedef start_jar_service():"""在Windows后台启动JAR服务（无窗口）"""logger.info("尝试启动JAR服务...")try:# 使用Windows特有的CREATE_NO_WINDOW标志startupinfo = subprocess.STARTUPINFO()startupinfo.dwFlags |= subprocess.STARTF_USESHOWWINDOWstartupinfo.wShowWindow = 0 # 0 = SW_HIDE (隐藏窗口)process = subprocess.Popen(JAR_COMMAND,stdout=open("jar_output1.log", "w"),stderr=open("jar_error1.log", "w"),startupinfo=startupinfo,creationflags=subprocess.CREATE_NO_WINDOW)logger.info(f"JAR服务已启动 (PID: {process.pid})")return Trueexcept Exception as e:logger.error(f"启动JAR服务失败: {str(e)}")return Falsedef main():"""主函数：执行一次完整的健康检查和必要的重启"""if check_all_services():logger.info("服务状态正常，无需操作")sys.exit(0) # 正常退出else:logger.warning("检测到服务异常，尝试重启...")if start_jar_service():logger.info("服务重启成功")sys.exit(1) # 异常但已处理else:logger.error("服务重启失败")sys.exit(2) # 异常且处理失败if __name__ == "__main__":logger.info("====== 服务检测开始 ======")main()logger.info("====== 服务检测结束 ======")
bat执行文件：start_check_admin.bat
@echo off echo 进度 100% 已完成 >> status.log start /B pythonw "D:\datax\dataxweb\service_check_admin.py" exit
