玩转企业集群运维管理系列（十二）：Keepalived 双主、非抢占模式及脑裂问题详解

民工哥

发布于 2023-12-14 15:03:21

3950

发布于 2023-12-14 15:03:21

文章被收录于专栏：民工哥技术之路民工哥技术之路

Keepalived 双主模式

双机双主：正常情况下nginx是一台提供服务，另外一条备份。

互为主备要引入两个VIP，如mysql双主，nginx双主，这样要引入两个VIP，也就是还需要引入。

virtual_ipaddress {
    192.168.179.199
    192.1681.79.188
}

只是代表这个实例有两个VIP。

192.168.179.102

[root@localhost ~]# cat /etc/keepalived/keepalived.conf 
global_defs {
    router_id real-server1
    script_user root
    enable_script_security
 }
 
vrrp_script chk_nginx {
    script "/data/shell/check_nginx_status.sh"
    interval 2
}
  
vrrp_instance VI_1 {
     state MASTER
     interface ens32   
     virtual_router_id 151 
     priority 100 
     advert_int 5  
     authentication {
         auth_type  PASS
         auth_pass  1111
 
     }
     virtual_ipaddress {  
       192.168.179.199
     }
    
      track_script {                                                                                  
       chk_nginx
    }
 }
  
vrrp_instance VI_2 {
     state BACKUP
     interface ens32
     virtual_router_id 152
     priority 50
     advert_int 5
     authentication {
         auth_type  PASS
         auth_pass  2222
 
     }
     virtual_ipaddress {
         192.168.179.198
     }
 
    track_script {
    chk_nginx
    }
}

更多关于企业集群运维管理系列的学习文章，请参阅：玩转企业集群运维管理专栏，本系列持续更新中。

[root@localhost ~]# cat /data/shell/check_nginx_status.sh 
#!/bin/bash
nginx_status=`ps -ef | grep nginx | grep -v grep | grep -v check | wc -l`
 
if [ $nginx_status -eq 0 ];then
 systemctl restart keepalived.service
fi

先启动nginx然后keepalived

[root@localhost ~]# ip a
2: ens32: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
    link/ether 00:0c:29:0e:1a:bf brd ff:ff:ff:ff:ff:ff
    inet 192.168.179.102/24 brd 192.168.179.255 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.199/32 scope global ens32
       valid_lft forever preferred_lft forever
    inet6 fe80::2d42:a0b1:1cdc:74c0/64 scope link 
       valid_lft forever preferred_lft forever

192.168.179.103

[root@localhost ~]# cat /etc/keepalived/keepalived.conf 
global_defs {
    router_id real-server2
    script_user root
    enable_script_security
 }
 
vrrp_script chk_nginx {
    script "/data/shell/check_nginx_status.sh"
    interval 2
}
 
vrrp_instance VI_1 {
     state BACKUP
     interface ens32
     virtual_router_id 151
     priority 50
     advert_int 5
     authentication {
         auth_type  PASS
         auth_pass  1111
     }
     virtual_ipaddress {
       192.168.179.199
     }
    
      track_script {                                                                                  
       chk_nginx
    }
 }
 
vrrp_instance VI_2 {
     state MASTER
     interface ens32
     virtual_router_id 152
     priority 100
     advert_int 5
     authentication {
         auth_type  PASS
         auth_pass  2222
 
     }
     virtual_ipaddress {
         192.168.179.198
     }
 
    track_script {
    chk_nginx
    }
} 
 
[root@localhost ~]# ip a
2: ens32: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
    link/ether 00:0c:29:61:90:c1 brd ff:ff:ff:ff:ff:ff
    inet 192.168.179.103/24 brd 192.168.179.255 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.198/32 scope global ens32
       valid_lft forever preferred_lft forever
    inet6 fe80::f54d:5639:6237:2d0e/64 scope link 
       valid_lft forever preferred_lft forever

102上干掉Nginx

102操作以及日志

[root@localhost ~]# pkill nginx
[root@localhost ~]# 
 
Nov 21 19:12:49 localhost Keepalived_vrrp[52348]: VRRP sockpool: [ifindex(2), proto(112), unicast(0), fd(10,11)]
Nov 21 19:12:49 localhost Keepalived[52346]: Stopping
Nov 21 19:12:49 localhost systemd: Stopping LVS and VRRP High Availability Monitor...
Nov 21 19:12:49 localhost Keepalived_healthcheckers[52347]: Stopped
Nov 21 19:12:50 localhost Keepalived_vrrp[52348]: Stopped
Nov 21 19:12:50 localhost Keepalived[52346]: Stopped Keepalived v1.3.5 (03/19,2017), git commit v1.3.5-6-g6fa32f2
Nov 21 19:12:50 localhost systemd: start request repeated too quickly for keepalived.service
Nov 21 19:12:50 localhost systemd: Failed to start LVS and VRRP High Availability Monitor.
Nov 21 19:12:50 localhost systemd: Unit keepalived.service entered failed state.
Nov 21 19:12:50 localhost systemd: keepalived.service failed.

103上查看日志，可以看到199跑到103上了，下面是103日志和ip

Nov 21 19:12:45 localhost Keepalived_vrrp[7582]: VRRP_Instance(VI_1) Transition to MASTER STATE
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: VRRP_Instance(VI_1) Entering MASTER STATE
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: VRRP_Instance(VI_1) setting protocol VIPs.
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: Sending gratuitous ARP on ens32 for 192.168.179.199
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: VRRP_Instance(VI_1) Sending/queueing gratuitous ARPs on ens32 for 192.168.179.199
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: Sending gratuitous ARP on ens32 for 192.168.179.199
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: Sending gratuitous ARP on ens32 for 192.168.179.199
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: Sending gratuitous ARP on ens32 for 192.168.179.199
Nov 21 19:12:50 localhost Keepalived_vrrp[7582]: Sending gratuitous ARP on ens32 for 192.168.179.199
Nov 21 19:12:55 localhost Keepalived_vrrp[7582]: Sending gratuitous ARP on ens32 for 192.168.179.199
Nov 21 19:12:55 localhost Keepalived_vrrp[7582]: VRRP_Instance(VI_1) Sending/queueing gratuitous ARPs on ens32 for 192.168.179.199
[root@localhost ~]# ip a
2: ens32: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
    link/ether 00:0c:29:61:90:c1 brd ff:ff:ff:ff:ff:ff
    inet 192.168.179.103/24 brd 192.168.179.255 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.198/32 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.199/32 scope global ens32
       valid_lft forever preferred_lft forever
    inet6 fe80::f54d:5639:6237:2d0e/64 scope link 
       valid_lft forever preferred_lft forever

更多关于企业集群运维管理系列的学习文章，请参阅：玩转企业集群运维管理专栏，本系列持续更新中。

非抢占模式 Nginx+keepalived实战

简单点说抢占模式就是，当master宕机后，backup 接管服务。后续当master恢复后，vip漂移到master上，master重新接管服务，多了一次多余的vip切换，而在实际生产中是不需要这样。实际生产中是，当原先的master恢复后，状态变为backup，不接管服务，这是非抢占模式。

重点：非抢占式俩节点state必须为bakcup,且必须配置nopreempt。

注意：这样配置后，我们要注意启动服务的顺序，优先启动的获取master权限，与优先级没有关系了。

总结：抢占模式即MASTER从故障中恢复后，会将VIP从BACKUP节点中抢占过来。非抢占模式即MASTER恢复后不抢占BACKUP升级为MASTER后的VIP。

1、两个节点的state都必须配置为BACKUP
2、两个节点都必须加上配置 nopreempt
3、其中一个节点的优先级必须要高于另外一个节点的优先级。

master节点keepalived配置如下（不抢占机制）

[root@real-server1 ~]# cat /etc/keepalived/keepalived.conf 
global_defs {
    router_id real-server1   #运行keepalived机器的标识
    script_user root
    enable_script_security
 }
 
vrrp_script chk_nginx {
    script "/data/shell/check_nginx_status.sh"   #监控服务脚本，脚本记得授予x执行权限；
    interval 2     #指定脚本执行的间隔。单位是秒。默认为1s。
}
 
vrrp_instance VI_1 {
     state BACKUP
     interface ens32         #绑定虚拟机的IP
     virtual_router_id 151   #虚拟路由id，和从机保持一致
     priority 100
     nopreempt               #设置为不抢占
     advert_int 5            #查间隔，默认1秒，VRRP心跳包的发送周期，单位为s 组播信息发送间隔，两个节点设置必须一样
     authentication {
         auth_type  PASS    #主辅认证密码（生产环境介意修改），最长支持八位
         auth_pass  1111
 
     }
     virtual_ipaddress {     #虚拟IP地址
       192.168.179.199
     }
    
      track_script {                                                                                  
       chk_nginx
    }
 }
 
#这个脚本你可以测试一下，将keepalived正常启动，然后你pkill nginx，执行这个脚本看看后面会发生什么，通过/var/log/messages来验证这个脚本是否正确，或者systemctl status keepalived查看状态，正确之后就可以配置在你的keepalived的配置文件当中。
[root@real-server1 ~]# cat /data/shell/check_nginx_status.sh
#!/bin/bash
nginx_status=$(ps -ef | grep nginx | grep -v grep | grep -v check | wc -l)
 
if [ $nginx_status -eq 0 ];then
  systemctl stop keepalived.service
fi
[root@real-server1 ~]# chmod o+x /data/shell/check_nginx_status.sh 
 
#用脚本实现健康检查，如果nginx进程为0就要发生keepalived切换，实现VIP漂移。当你的nginx挂掉了，那么你的keepalived永远都启动不了，因为下面脚本定义了systemctl stop keepalived.service，nginx没有起来那么keepalived起来会自动关闭。
 
#advert_int 5  检查间隔，默认1秒，VRRP心跳包的发送周期，单位为s，组播信息发送间隔，可以看到组播包里面的信息包含了virtual_router_id 151 虚拟路由ID和优先级priority 100。
[root@localhost shell]# tcpdump -i ens32 -nn net 224.0.0.18
tcpdump: verbose output suppressed, use -v or -vv for full protocol decode
listening on ens32, link-type EN10MB (Ethernet), capture size 262144 bytes
10:26:42.910170 IP 192.168.179.102 > 224.0.0.18: VRRPv2, Advertisement, vrid 151, prio 100, authtype simple, intvl 5s, length 20
10:26:47.911831 IP 192.168.179.102 > 224.0.0.18: VRRPv2, Advertisement, vrid 151, prio 100, authtype simple, intvl 5s, length 20
10:26:52.915502 IP 192.168.179.102 > 224.0.0.18: VRRPv2, Advertisement, vrid 151, prio 100, authtype simple, intvl 5s, length 20

测试配置是否正确

[root@real-server1 ~]# ps -ef | grep keepalived | grep -v grep
root      70592      1  0 20:05 ?        00:00:00 /usr/sbin/keepalived -D
root      70593  70592  0 20:05 ?        00:00:00 /usr/sbin/keepalived -D
root      70594  70592  0 20:05 ?        00:00:00 /usr/sbin/keepalived -D
#keepalived正常启动的时候，共启动3个进程，一个是父进程，负责监控其子进程，一个是vrrp子进程，另外一个是checkers子进程。

两个子进程都被系统watchlog看管，两个子进程各自负责复杂自己的事。Healthcheck子进程检查各自服务器的健康状况，例如http,lvs。如果healthchecks进程检查到master上服务不可用了，就会通知本机上的vrrp子进程，让他删除通告，并且去掉虚拟IP，转换为BACKUP状态。
Jul 31 20:05:38 real-server1 Keepalived[70591]: Starting Keepalived v1.3.5 (03/19,2017), git commit v1.3.5-6-g6fa32f2
Jul 31 20:05:38 real-server1 Keepalived[70591]: Opening file '/etc/keepalived/keepalived.conf'.
Jul 31 20:05:38 real-server1 systemd: PID file /var/run/keepalived.pid not readable (yet?) after start.
Jul 31 20:05:38 real-server1 Keepalived[70592]: Starting Healthcheck child process, pid=70593
Jul 31 20:05:38 real-server1 Keepalived[70592]: Starting VRRP child process, pid=70594
Jul 31 20:05:38 real-server1 systemd: Started LVS and VRRP High Availability Monitor.
Jul 31 20:05:38 real-server1 Keepalived_healthcheckers[70593]: Opening file '/etc/keepalived/keepalived.conf'.
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: Registering Kernel netlink reflector
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: Registering Kernel netlink command channel
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: Registering gratuitous ARP shared channel
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: Opening file '/etc/keepalived/keepalived.conf'.
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: VRRP_Instance(VI_1) removing protocol VIPs.
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: Using LinkWatch kernel netlink reflector...
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: VRRP_Instance(VI_1) Entering BACKUP STATE
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: VRRP sockpool: [ifindex(2), proto(112), unicast(0), fd(10,11)]
Jul 31 20:05:38 real-server1 Keepalived_vrrp[70594]: VRRP_Script(chk_nginx) succeeded
 
#测试健康检查脚本是否有用，可以看到脚本没问题
[root@real-server2 ~]# pkill nginx
Jul 31 20:06:50 real-server1 Keepalived[70592]: Stopping
Jul 31 20:06:50 real-server1 systemd: Stopping LVS and VRRP High Availability Monitor...
Jul 31 20:06:50 real-server1 Keepalived_vrrp[70594]: VRRP_Instance(VI_1) sent 0 priority
Jul 31 20:06:50 real-server1 Keepalived_vrrp[70594]: VRRP_Instance(VI_1) removing protocol VIPs.
Jul 31 20:06:50 real-server1 Keepalived_healthcheckers[70593]: Stopped
Jul 31 20:06:51 real-server1 Keepalived_vrrp[70594]: Stopped
Jul 31 20:06:51 real-server1 systemd: Stopped LVS and VRRP High Availability Monitor.
Jul 31 20:06:51 real-server1 Keepalived[70592]: Stopped Keepalived v1.3.5 (03/19,2017), git commit v1.3.5-6-g6fa32f2

backup节点keepalived配置如下

[root@real-server2 ~]# cat /etc/keepalived/keepalived.conf 
global_defs {
    router_id real-server2
    script_user root
    enable_script_security
 }
 
vrrp_script chk_nginx {
    script "/data/shell/check_nginx_status.sh"
    interval 2
}
 
vrrp_instance VI_1 {
     state BACKUP
     interface ens32
     virtual_router_id 151
     priority 50
     advert_int 5
     authentication {
         auth_type  PASS
         auth_pass  1111
     }
     virtual_ipaddress {
       192.168.179.199
     }
    
      track_script {                                                                                  
       chk_nginx
    }
 }

更多关于企业集群运维管理系列的学习文章，请参阅：玩转企业集群运维管理专栏，本系列持续更新中。

不抢占机制演示如下

#Master节点测试，将nginx进程杀掉，根据健康检查脚本会实现vip漂移到备节点
[root@real-server1 ~]# ip a
2: ens32: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
    link/ether 00:0c:29:61:90:c1 brd ff:ff:ff:ff:ff:ff
    inet 192.168.179.103/24 brd 192.168.179.255 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.199/32 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.199/24 brd 192.168.179.255 scope global secondary ens32:1
       valid_lft forever preferred_lft forever
    inet6 fe80::f54d:5639:6237:2d0e/64 scope link 
       valid_lft forever preferred_lft forever
[root@real-server1 ~]# pkill nginx
 
#keepalived主节点日志如下
Jul 31 20:27:45 real-server1 Keepalived[72926]: Stopping
Jul 31 20:27:45 real-server1 systemd: Stopping LVS and VRRP High Availability Monitor...
Jul 31 20:27:45 real-server1 Keepalived_vrrp[72928]: VRRP_Instance(VI_1) sent 0 priority
Jul 31 20:27:45 real-server1 Keepalived_vrrp[72928]: VRRP_Instance(VI_1) removing protocol VIPs.
Jul 31 20:27:45 real-server1 Keepalived_healthcheckers[72927]: Stopped
Jul 31 20:27:46 real-server1 Keepalived_vrrp[72928]: Stopped
Jul 31 20:27:46 real-server1 Keepalived[72926]: Stopped Keepalived v1.3.5 (03/19,2017), git commit v1.3.5-6-g6fa32f2
Jul 31 20:27:46 real-server1 systemd: Stopped LVS and VRRP High Availability Monitor. 
 
#来到备节点查看
[root@real-server2 ~]# ip a
2: ens32: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
    link/ether 00:0c:29:a7:ff:f7 brd ff:ff:ff:ff:ff:ff
    inet 192.168.179.104/24 brd 192.168.179.255 scope global ens32
       valid_lft forever preferred_lft forever
    inet 192.168.179.199/32 scope global ens32
       valid_lft forever preferred_lft forever
    inet6 fe80::831c:6df1:a633:742a/64 scope link 
       valid_lft forever preferred_lft forever
 
#备节点的keepalived日志，可以看到故障转移成功
Jul 31 08:27:45 real-server2 Keepalived_vrrp[8961]: VRRP_Instance(VI_1) Transition to MASTER STATE
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: VRRP_Instance(VI_1) Entering MASTER STATE
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: VRRP_Instance(VI_1) setting protocol VIPs.
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: Sending gratuitous ARP on ens32 for 192.168.179.199
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: VRRP_Instance(VI_1) Sending/queueing gratuitous ARPs on ens32 for 192.168.179.199
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: Sending gratuitous ARP on ens32 for 192.168.179.199
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: Sending gratuitous ARP on ens32 for 192.168.179.199
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: Sending gratuitous ARP on ens32 for 192.168.179.199
Jul 31 08:27:50 real-server2 Keepalived_vrrp[8961]: Sending gratuitous ARP on ens32 for 192.168.179.199
Jul 31 08:27:55 real-server2 Keepalived_vrrp[8961]: Sending gratuitous ARP on ens32 for 192.168.179.199
Jul 31 08:27:55 real-server2 Keepalived_vrrp[8961]: VRRP_Instance(VI_1) Sending/queueing gratuitous ARPs on ens32 for 192.168.179.199
 
#Master主节点将nginx启动，并且启动keepalived,可以看到现在进入了backup状态，没有抢占
[root@real-server1 ~]# /usr/local/nginx/sbin/nginx
[root@real-server1 ~]# systemctl start keepalived
Jul 31 20:43:33 real-server1 Keepalived[75196]: Starting Healthcheck child process, pid=75197
Jul 31 20:43:33 real-server1 Keepalived[75196]: Starting VRRP child process, pid=75198
Jul 31 20:43:33 real-server1 systemd: Started LVS and VRRP High Availability Monitor.
Jul 31 20:43:33 real-server1 Keepalived_healthcheckers[75197]: Opening file '/etc/keepalived/keepalived.conf'.
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: Registering Kernel netlink reflector
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: Registering Kernel netlink command channel
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: Registering gratuitous ARP shared channel
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: Opening file '/etc/keepalived/keepalived.conf'.
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: VRRP_Instance(VI_1) removing protocol VIPs.
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: Using LinkWatch kernel netlink reflector...
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: VRRP_Instance(VI_1) Entering BACKUP STATE
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: VRRP sockpool: [ifindex(2), proto(112), unicast(0), fd(10,11)]
Jul 31 20:43:33 real-server1 Keepalived_vrrp[75198]: VRRP_Script(chk_nginx) succeeded

最后总结上面简化配置为

#A调度机器设置为
vrrp_instance VI_feng
{
    ....
     
    state backup
    priority 100
    nopreempt
   ....
 
  }
 
#B调度机器设置为
vrrp_instance VI_feng
{
    ....
     
    state backup
    priority 70
    nopreempt
   ....
 
  }

不抢占是配置在优先级高的机器上面，同时状态要是backup，即集群内部要想实现不抢占，状态都需要设置为backup，优先级还是正常有高有低。谁优先级高配置一个不抢占参数nopreempt（因为优先级高的会抢占VIP）。每次抢占就需要发生切换和漂移，来回切换漂移影响业务访问，服务要中断！！！！更多关于企业集群运维管理系列的学习文章，请参阅：玩转企业集群运维管理专栏，本系列持续更新中。

Keepalived 脑裂问题的起因与解决

Keepalived脑裂

脑裂（split）：指在一个高可用（HA）系统中，当联系着的两个节点断开联系时，本来为一个整体的系统，分裂为两个独立节点，这时两个节点开始争抢共享资源，结果会导致系统混乱，数据损坏。对于无状态服务的HA，无所谓脑裂不脑裂；但对有状态服务(比如MySQL)的HA，必须要严格防止脑裂。

在高可用HA系统中，当联系2个节点的“心跳线”断开时，本来为一整体，一个VRRP协议组，动作协调的HA系统，就分裂为两个独立的个体。由于相互失去了联系，都以为对方出了故障；两个节点的HA软件像“连体人”一样，有共同的身体却拥有两个脑袋，争抢“共享资源”身体，争抢服务器里的应用服务，就会发送严重后果，或者共享资源被瓜分，两边服务都起不来或者都起来都为Master，假如两端服务器发生了脑裂现象就会成为各自的Master，会同时读写“共享存储”，导致数据损坏（常见的如数据库轮询着的联机日志出错）对付HA系统“裂脑”的对策，目前达成共识的的大概有以下几条：

1）添加冗余的心跳线，例如：双线条线（心跳线也HA），尽量减少“裂脑”发生几率；
2）启用磁盘锁。正在服务一方锁住共享磁盘，“裂脑”发生时，让对方完全“抢不走”共享磁盘资源。但使用锁磁盘也会有一个不小的问题，如果占用共享盘的一方不主动“解锁”，另一方就永远得不到共享磁盘。现实中假如服务节点突然死机或崩溃，就不可能执行解锁命令。后备节点也就接管不了共享资源和应用服务。于是有人在HA中设计了“智能”锁。即：正在服务的一方只在发现心跳线全部断开（察觉不到对端）时才启用磁盘锁。平时就不上锁了。
3）设置仲裁机制。例如设置参考IP（如网关IP），当心跳线完全断开时，2个节点都各自ping一下参考IP，不通则表明断点就出在本端。不仅“心跳”、还兼对外“服务”的本端网络链路断了，即使启动（或继续）应用服务也没有用了，那就主动放弃竞争，让能够ping通参考IP的一端去起服务。更保险一些，ping不通参考IP的一方干脆就自我重启，以彻底释放有可能还占用着的那些共享资源。

脑裂产生的原因

一般来说，裂脑的发生，有以下几种原因：

1）高可用服务器对之间心跳线链路发生故障，导致无法正常通信。
2）因心跳线坏了（包括断了，老化）。
3）因网卡及相关驱动坏了，ip配置及冲突问题（网卡直连）。
4）因心跳线间连接的设备故障（网卡及交换机）。
5）因仲裁的机器出问题（采用仲裁的方案）。
6）高可用服务器上开启了 iptables防火墙阻挡了心跳消息传输。
7）高可用服务器上心跳网卡地址等信息配置不正确，导致发送心跳失败。
8）其他服务配置不当等原因，如心跳方式不同，心跳广插冲突、软件Bug等提示：Keepalived配置里同一 VRRP实例如果 virtual_router_id两端参数配置不一致也会导致裂脑问题发生。

iptables导致keepalived脑裂的问题

曾经碰到的一个keepalived脑裂的问题（如果启用了iptables，不设置"系统接收VRRP协议"的规则，就会出现脑裂）。

曾经在做keepalived+Nginx主备架构的环境时，当重启了备用机器后，发现两台机器都拿到了VIP。这也就是意味着出现了keepalived的脑裂现象，检查了两台主机的网络连通状态，发现网络是好的。然后在备机上抓包：

# tcpdump -i eth0|grep VRRP 
tcpdump: verbose output suppressed, use -v or -vv for full protocol decode 
listening on eth0, link-type EN10MB (Ethernet), capture size 65535 bytes 
 
22:10:17.146322 IP 192.168.1.54 > vrrp.mcast.net: VRRPv2, Advertisement, vrid 51, prio 160, authtype simple, intvl 1s, length 20 
22:10:17.146577 IP 192.168.1.96 > vrrp.mcast.net: VRRPv2, Advertisement, vrid 51, prio 50, authtype simple, intvl 1s, length 20 
22:10:17.146972 IP 192.168.1.54 > vrrp.mcast.net: VRRPv2, Advertisement, vrid 51, prio 160, authtype simple, intvl 1s, length 20 
22:10:18.147136 IP 192.168.1.96 > vrrp.mcast.net: VRRPv2, Advertisement, vrid 51, prio 50, authtype simple, intvl 1s, length 20

抓包发现备机能接收到master发过来的VRRP广播，那为什么还会有脑裂现象？

接着发现iptables开启着，检查了防火墙配置。发现系统不接收VRRP协议。于是修改iptables，添加允许系统接收VRRP协议的配置：自己添加了下面的iptables规则：

-A INPUT -s 192.168.1.0/24 -d 224.0.0.18 -j ACCEPT       #允许组播地址通信
-A INPUT -s 192.168.1.0/24 -p vrrp -j ACCEPT             #允许VRRP（虚拟路由器冗余协）通信

最后重启iptables，发现备机上的VIP没了。虽然问题解决了，但备机明明能抓到master发来的VRRP广播包，却无法改变自身状态。只能说明网卡接收到数据包是在iptables处理数据包之前。

预防keepalived脑裂问题

在实际生产环境中，我们可以从以下几个方面来防止裂脑问题的发生：

同时使用串行电缆和以太网电缆连接，同时用两条心跳线路，这样一条线路坏了，另一个还是好的，依然能传送心跳消息。
可以采用第三方仲裁的方法。由于keepalived体系中主备两台机器所处的状态与对方有关。如果主备机器之间的通信出了网题，就会发生脑裂，此时keepalived体系中会出现双主的情况，产生资源竞争。
一般可以引入仲裁来解决这个问题，即每个节点必须判断自身的状态。最简单的一种操作方法是，在主备的keepalived的配置文件中增加check配置，服务器周期性地ping一下网关，如果ping不通则认为自身有问题。
最容易的是借助keepalived提供的vrrp_script及track_script实现。如下所示：

#vim /etc/keepalived/keepalived.conf
   ...... 
   vrrp_script check_local {
    script "/root/check_gateway.sh" 
    interval 5
    }
   ...... 
   track_script {     
   check_local                   
   }

脚本内容：

# cat /root/check_gateway.sh
#!/bin/sh 
VIP=$1
GATEWAY=192.168.1.1
/sbin/arping -I em1 -c 5 -s $VIP $GATEWAY &>/dev/null

check_gateway.sh 就是我们的仲裁逻辑，发现ping不通网关，则关闭keepalived。

推荐自己写脚本，写一个while循环，每轮ping网关，累计连续失败的次数，当连续失败达到一定次数则运行service keepalived stop关闭keepalived服务。

如果发现又能够ping通网关，再重启keepalived服务。最后在脚本开头再加上脚本是否已经运行的判断逻辑，将该脚本加到crontab里面。更多关于企业集群运维管理系列的学习文章，请参阅：玩转企业集群运维管理专栏，本系列持续更新中。

参考链接：https://blog.csdn.net/qq_34556414/article/ details/109905166 https://blog.csdn.net/qq_34556414 /article/details/107696010 https://blog.csdn.net/ qq_34556414/article/details/107641766

本文参与腾讯云自媒体分享计划，分享自微信公众号。

原始发表：2023-12-14，如有侵权请联系 cloudcommunity@tencent.com 删除

企业