keepalived BACKUP 无明显原因地过渡到 MASTER

Question

出于测试目的，我提出了 2 个 Debian jessie VM（使用 vagrant 和 virtualbox），并让它们都运行一个迷你 Web 服务器并配置了 keepalived。

问题是 BACKUP 服务器在几秒钟后service keepalived restart无缘无故地转换到 MASTER。

我还看到 VRRP 多播数据包到达两台服务器，它们看起来符合预期。（这似乎意味着Keepalived 不需要的过渡到 master 中建议的解决方案无关）

配置文件如下：

主服务器：

global_defs {
        lvs_id tom_lvs
}

vrrp_instance tom_lvs {
        state MASTER
        interface eth1
    virtual_router_id 1
    lvs_sync_daemon_interface eth1
    priority 100
    preempt
        authentication {
                auth_type PASS
                auth_pass 1234
        }

    advert_int 1
    virtual_ipaddress {
        172.28.128.10/24
    }

        virtual_server 172.28.128.10 3000 {
                delay_loop 10
                lb_algo wlc
                lb_kind DR
                protocol TCP
                persistence_timeout 1800
                sorry_server 172.28.128.3 3000
                real_server 172.28.128.4 3000 {
                        weight 5
                        HTTP_GET {
                                url {
                                        path /index.html
                                }
                        }
                }
        }
}

备份服务器：

global_defs {
        lvs_id tom_lvs
}

vrrp_instance tom_lvs {
        state BACKUP
        interface eth1
    virtual_router_id 2
        lvs_sync_daemon_interface eth1
    priority 96
    preempt
        authentication {
                auth_type PASS
                auth_pass 1234
        }

    advert_int 1
    virtual_ipaddress {
        172.28.128.10/24
    }

        virtual_server 172.28.128.10 3000 {
                delay_loop 10
                lb_algo wlc
                lb_kind DR
                protocol TCP
                persistence_timeout 1800
                sorry_server 172.28.128.3 3000
                real_server 172.28.128.3 3000 {
                        weight 5
                        HTTP_GET {
                                url {
                                        path /index.html
                                }
                        }
                }
        }
}

显示从 BACKUP 到 MASTER 的转换的备份服务器上的日志：

Feb 14 13:22:44 jessie Keepalived_vrrp[1145]: VRRP_Instance(tom_lvs) sending 0 priority
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: Registering Kernel netlink reflector
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: Registering Kernel netlink command channel
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: Registering gratuitous ARP shared channel
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: Opening file '/etc/keepalived/keepalived.conf'.
Feb 14 13:22:44 jessie Keepalived_healthcheckers[1175]: Registering Kernel netlink reflector
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: Configuration is using : 62104 Bytes
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: Using LinkWatch kernel netlink reflector...
Feb 14 13:22:44 jessie Keepalived_healthcheckers[1175]: Registering Kernel netlink command channel
Feb 14 13:22:44 jessie Keepalived_healthcheckers[1175]: Opening file '/etc/keepalived/keepalived.conf'.
Feb 14 13:22:44 jessie Keepalived_healthcheckers[1175]: Configuration is using : 12103 Bytes
Feb 14 13:22:44 jessie Keepalived_vrrp[1176]: VRRP_Instance(tom_lvs) Entering BACKUP STATE
Feb 14 13:22:44 jessie Keepalived_healthcheckers[1175]: Using LinkWatch kernel netlink reflector...
Feb 14 13:22:44 jessie Keepalived_healthcheckers[1175]: Activating healthchecker for service [172.28.128.3]:3000
Feb 14 13:22:48 jessie Keepalived_vrrp[1176]: VRRP_Instance(tom_lvs) Transition to MASTER STATE
Feb 14 13:22:49 jessie Keepalived_vrrp[1176]: VRRP_Instance(tom_lvs) Entering MASTER STATE

这是我在两台机器上看到的 tshark

Capturing on 'eth1'
  1   0.000000 172.28.128.4 -> 224.0.0.18   VRRP 60 Announcement (v2)
  2   0.001404 172.28.128.3 -> 224.0.0.18   VRRP 54 Announcement (v2)
  3   1.002283 172.28.128.4 -> 224.0.0.18   VRRP 60 Announcement (v2)
  4   1.004130 172.28.128.3 -> 224.0.0.18   VRRP 54 Announcement (v2)
  5   2.004466 172.28.128.4 -> 224.0.0.18   VRRP 60 Announcement (v2)
  6   2.006260 172.28.128.3 -> 224.0.0.18   VRRP 54 Announcement (v2)

Answer 1

我在研究类似行为时遇到了这个问题，所以我想我会为了其他同样困惑的人的利益发布答案。

两个配置文件都需要对 virtual_router_id 使用相同的值，因为这就是 keepalived 知道状态消息是针对同一个虚拟路由器的方式。

在上面的配置中，它们具有不同的值，因此独立运行。