架构说明架构GTID MySQL 一主一从 Keepalived VRRP VIP组件分工MySQL 主从实现数据复制冗余check_mysql.sh健康探测脚本MySQL 异常释放 VIPkeepalived vrrp实现 VIP 虚拟 IP 漂移notify 系列回调脚本VIP 切换自动修改 MySQLread_only完成数据库角色切换⚠️重要局限 Keepalived 只是网络组件脚本只开关只读参数不会自动重建旧节点主从复制双节点存在脑裂风险异步复制切换存在数据丢失风险。环境规划节点IP角色VIPNodeA192.168.1.10MySQL Master192.168.1.100虚 IPNodeB192.168.1.11MySQL Slave192.168.1.100虚 IP步骤 1部署 MySQL GTID 主从复制1.1 node‑A主库my.cnf[mysqld] server-id1 datadir/var/lib/mysql socket/tmp/mysql.sock log_binmysql-bin binlog_formatROW gtid_modeON enforce_gtid_consistencyON log_slave_updatesON # 建议开启半同步降低切换丢数据概率 plugin-load-addrpl_semi_sync_master.so rpl_semi_sync_master_enabled11.2 node‑B从库my.cnf[mysqld] server-id2 datadir/var/lib/mysql socket/tmp/mysql.sock log_binmysql-bin relay_logrelay-bin gtid_modeON enforce_gtid_consistencyON log_slave_updatesON read_onlyON super_read_onlyON plugin-load-addrpl_semi_sync_slave.so rpl_semi_sync_slave_enabled1两台重启 mysqlsystemctl restart mysqld1.3 主库创建复制账号 node‑A 执行CREATE USER repl% IDENTIFIED BY Repl123456; GRANT REPLICATION SLAVE ON *.* TO repl%; flush privileges;1.4 从库搭建复制链路 node‑B 执行CHANGE MASTER TO MASTER_HOST192.168.1.10, MASTER_USERrepl, MASTER_PASSWORDRepl123456, MASTER_AUTO_POSITION1; start slave; show slave status\G校验Slave_IO_Running: Yes、Slave_SQL_Running: Yes。步骤 2安装 Keepalivednode‑A、node‑B 两台都执行yum install keepalived -y systemctl enable keepalived # VRRP协议号112防火墙放行 firewall-cmd --add-protocolvrrp --permanent firewall-cmd --reload步骤 3全套切换脚本编写两台节点全部部署脚本存放路径/etc/keepalived/全部使用本地 socket/tmp/mysql.sock不走 TCP、不受 VIP 影响。脚本 1健康检测脚本 check_mysql.shvrrp_script 周期性调用MySQL 检测失败停止本机 keepalived 释放 VIP#!/bin/bash # /etc/keepalived/check_mysql.sh MYSQL_USERroot MYSQL_PASSRoot123456 MYSQL_SOCK/tmp/mysql.sock mysqladmin -S ${MYSQL_SOCK} -u${MYSQL_USER} -p${MYSQL_PASS} ping /dev/null 21 if [ $? -eq 0 ];then exit 0 else # mysql异常停止keepalived释放VIP systemctl stop keepalived exit 1 fi脚本 2notify_master.sh本机接管 VIP成为主节点触发关闭只读允许业务写入#!/bin/bash # /etc/keepalived/notify_master.sh MYSQL_USERroot MYSQL_PASSRoot123456 MYSQL_SOCK/tmp/mysql.sock mysql -S ${MYSQL_SOCK} -u${MYSQL_USER} -p${MYSQL_PASS} EOF set global read_onlyOFF; set global super_read_onlyOFF; EOF脚本 3notify_backup.sh本机失去 VIP降级备机触发开启只读禁止写入#!/bin/bash # /etc/keepalived/notify_backup.sh MYSQL_USERroot MYSQL_PASSRoot123456 MYSQL_SOCK/tmp/mysql.sock mysql -S ${MYSQL_SOCK} -u${MYSQL_USER} -p${MYSQL_PASS} EOF set global read_onlyON; set global super_read_onlyON; EOF脚本 4notify_fault.shVRRP 故障状态兜底强制开启只读#!/bin/bash # /etc/keepalived/notify_fault.sh MYSQL_USERroot MYSQL_PASSRoot123456 MYSQL_SOCK/tmp/mysql.sock mysql -S ${MYSQL_SOCK} -u${MYSQL_USER} -p${MYSQL_PASS} EOF set global read_onlyON; set global super_read_onlyON; EOF脚本赋权重要权限不足脚本不执行chmod x /etc/keepalived/*.sh chmod 600 /etc/keepalived/*.shchmod 600防止普通用户读取脚本内数据库明文密码。步骤 4Keepalived 配置文件node‑A keepalived.conf初始主priority100global_defs { router_id NODE_A } vrrp_script check_mysql { script /etc/keepalived/check_mysql.sh interval 2 weight -30 fall 2 rise 1 } vrrp_instance VI_MYSQL { state BACKUP interface ens33 # 修改为本机网卡 ip a查看 virtual_router_id 51 priority 100 advert_int 1 nopreempt # 非抢占故障恢复不抢回VIP避免业务抖动 authentication { auth_type PASS auth_pass 123456 } virtual_ipaddress { 192.168.1.100/24 dev ens33 label ens33:0 } track_script { check_mysql } # 切换回调脚本高可用核心 notify_master /etc/keepalived/notify_master.sh notify_backup /etc/keepalived/notify_backup.sh notify_fault /etc/keepalived/notify_fault.sh }node‑B keepalived.conf初始从 priority90仅修改 router_id、priority其余全部参数必须和 A 保持一致global_defs { router_id NODE_B } vrrp_script check_mysql { script /etc/keepalived/check_mysql.sh interval 2 weight -30 fall 2 rise 1 } vrrp_instance VI_MYSQL { state BACKUP interface ens33 virtual_router_id 51 priority 90 advert_int 1 nopreempt authentication { auth_type PASS auth_pass 123456 } virtual_ipaddress { 192.168.1.100/24 dev ens33 label ens33:0 } track_script { check_mysql } notify_master /etc/keepalived/notify_master.sh notify_backup /etc/keepalived/notify_backup.sh notify_fault /etc/keepalived/notify_fault.sh }步骤 5启动服务验证初始状态systemctl start keepalived systemctl enable keepalived # 查看VIP在哪台机器 ip a预期VIP192.168.1.100在 node‑A。 业务测试mysql -h 192.168.1.100 -uroot -p可以读写。步骤 6故障切换完整流程说明故障场景node‑A MySQL 宕机node‑Acheck_mysql.shmysqladmin ping 检测失败 → 执行systemctl stop keepalived本机 VIP 释放。node‑B 收不到 VRRP 心跳包接管 VIP192.168.1.100。node‑B 获取 VIP 触发notify_master.sh自动执行set global read_onlyOFF; set global super_read_onlyOFF;✅业务访问 VIP直接读写 node‑B无需人工干预完成故障转移。故障恢复node‑A 修复完成node‑A 启动 MySQL、keepalived。由于配置nopreempt不会抢占 VIPVIP 仍然留在 node‑B。⚠️此时 node‑A 是独立 MySQL 实例脚本不会自动重建主从复制。人工操作将旧主 node‑A 配置为新主 node‑B 的从库stop slave; reset slave all; CHANGE MASTER TO MASTER_HOST192.168.1.11, MASTER_USERrepl, MASTER_PASSWORDRepl123456, MASTER_AUTO_POSITION1; start slave; show slave status\G步骤 7脚本调试排错手段# 手动执行脚本看脚本是否报错 bash -x /etc/keepalived/check_mysql.sh bash -x /etc/keepalived/notify_master.sh # keepalived日志查看脚本执行记录 tail -f /var/log/messages # 查看keepalived状态 systemctl status keepalived关键风险 局限脑裂风险主备之间网络断开两边 MySQL 都正常运行健康检测脚本不会停止 keepalived两台机器同时持有 VIP双写导致数据错乱。双节点 keepalived 架构无法彻底消除脑裂。notify 脚本只开关read_only参数不会自动重建旧节点主从复制故障节点恢复需要人工处理复制链路。脚本本身属于单点故障账号密码错误、socket 路径错误、权限不足都会导致切换失效。异步复制切换存在丢失 binlog 风险半同步可以降低概率但不能 100% 避免。脚本存在明文密码安全风险生产建议使用 mysql login‑path 避免明文。精简版MySQL 主从 Keepalived 实现完整高可用分为搭建 GTID MySQL 一主一从复制建议开启半同步复制。两台节点部署 Keepalived全部配置 BACKUP 模式加 nopreempt 非抢占防止故障恢复抢占 VIP 抖动。编写check_mysql.sh健康检测脚本MySQL 异常停止本机 keepalived 释放 VIP。配置 notify_master、notify_backup、notify_fault 回调脚本VIP 漂移自动开关 read_only 参数完成数据库读写角色切换。 故障发生时故障节点释放 VIP备节点接管 VIP 并且自动放开写权限业务通过 VIP 继续访问。 该方案存在脑裂风险故障节点恢复后脚本不会自动重建主从复制需要人工介入。