galera_cluster_tool/monitor.py

import os
import sys
import time

import psutil


# 定义状态
NORMAL = 'NORMAL'
ABNORMAL = 'ABNORMAL'

# 定义状态持续时间阈值（秒）
NORMAL_THRESHOLD = 10 * 60
ABNORMAL_THRESHOLD = 30


def get_process(name="mysqld"):

    for proc in psutil.process_iter():
        if proc.name() == name:
            return proc
    return None


# 定义状态机类
class ProcessMonitorStateMachine:
    def __init__(self, process_name):
        self.process_name = process_name
        self.state = NORMAL
        self.last_state_change_time = time.time()

    def check_process(self):
        """检查进程是否存在，并返回其状态"""
        process = get_process(name=self.process_name)
        if process:
            return NORMAL
        else:
            return ABNORMAL

    def process_event(self):
        """处理事件，根据事件类型切换状态"""
        current_time = time.time()

        process_state = self.check_process()
        print(f"当前状态：{process_state} {current_time}")
        if process_state != self.state:
            self.state = process_state
            self.last_state_change_time = current_time
            print(f"状态变更为: {self.state}")
            if self.state == ABNORMAL:
                self.handle_abnormal_timeout()

        else:
            if self.state == NORMAL and (current_time - self.last_state_change_time) > NORMAL_THRESHOLD:
                self.handle_normal_timeout()
                self.last_state_change_time = current_time

            elif self.state == ABNORMAL and (current_time - self.last_state_change_time) > ABNORMAL_THRESHOLD:
                self.handle_abnormal_timeout()
                self.last_state_change_time = current_time

    def handle_normal_timeout(self):
        """处理正常状态超时的指令"""
        print("程序已正常，退出监控")
        sys.exit(0)

    def handle_abnormal_timeout(self):
        """处理异常状态超时的指令"""
        print("异常状态持续时间过长，执行特定指令。")
        os.system("/usr/bin/mysqld_bootstrap")


# 使用状态机监控进程
def main():
    process_name = "mysqld"  # 指定要监控的进程名称
    monitor = ProcessMonitorStateMachine(process_name)
    print("start...")
    # 模拟定时检查进程状态
    while True:
        monitor.process_event()
        time.sleep(10)  # 每10秒检查一次


if __name__ == '__main__':
    main()