Keepalived

Keepalived は、Virtual Router Redundancy Protocol（VRRP）を使用して高可用性機能とロードバランシングを提供する、Linux 仮想サーバー（LVS）クラスター用のフェイルオーバーおよび監視デーモンです。

インストール

keepalived パッケージを インストール してください。

keepalived.service を 有効化 してください。

設定

このサービスは /etc/keepalived/keepalived.conf で設定されます。

マスターとバックアップのフェイルオーバー設定

以下は、仮想 IP アドレスを共有する 2 つのサーバー/ルーターで高可用性クラスターを設定する基本的な設定を提供します。両方とも keepalived を実行している必要があります。

ホスト A で：

/etc/keepalived/keepalived.conf

vrrp_instance VI_1 {
    state MASTER
    interface eth0
    virtual_router_id 101
    priority 20
    advert_int 1
    authentication {
        auth_type PASS
        auth_pass pass1234 
    }
    virtual_ipaddress {
        172.16.0.100
    }
}

ホスト B で：

/etc/keepalived/keepalived.conf

vrrp_instance VI_1 {
    state BACKUP
    interface eth0
    virtual_router_id 101
    priority 15
    advert_int 1
    authentication {
        auth_type PASS
        auth_pass pass1234
    }
    virtual_ipaddress {
        172.16.0.100
    }
}

この state は、サービスが開始され、マスターの役割を引き継ぐ最も高い優先度を持つホストを決定する前に、ホストの初期役割を設定します。この設定により、より高い優先度を持つホスト A が VIP（仮想 IP アドレス）を管理し、ホスト A が応答しなくなった場合にのみ、ホスト B が VIP を引き継ぎます。デフォルトの切り替えまでの時間は 1 秒です。

Failover setup with monitoring

The following provides a failover cluster for service (sshd) running on the same hosts as keepalived.

On host A:

/etc/keepalived/keepalived.conf

global_defs {
   notification_email {
     admin@example.net
   }
   notification_email_from lb1@example.net
   smtp_server mail.example.net
   smtp_connect_timeout 30
}

# enter failed state when the sshd process is down
vrrp_track_process track_sshd {
    process sshd
    delay 1
}

vrrp_instance VI_1 {
    state BACKUP
    nopreempt
    interface eth0
    virtual_router_id 101
    priority 20
    advert_int 1
    authentication {
        auth_type AH
        auth_pass key12345
    }
    virtual_ipaddress {
        172.16.0.100
    }
    track_process {
        track_sshd
    }
    smtp_alert
}

On host B:

/etc/keepalived/keepalived.conf

global_defs {
   notification_email {
     admin@example.net
   }
   notification_email_from lb2@example.net
   smtp_server mail.example.net
   smtp_connect_timeout 30
}

# enter failed state when the sshd process is down
vrrp_track_process track_sshd {
    process sshd
    delay 1
}

vrrp_instance VI_1 {
    state BACKUP
    nopreempt
    interface eth0
    virtual_router_id 101
    priority 15
    advert_int 1
    authentication {
        auth_type AH
        auth_pass key12345
    }
    virtual_ipaddress {
        172.16.0.100
    }
    track_process {
        track_sshd
    }
    smtp_alert
}

This setup provides a more dynamic cluster for the sshd service running on the failover host itself. Both servers start in the backup state and the nopreempt option allows the server with the lower priority to keep the master role even when the server with the higher priority returns after a failure. This prevents the interruption of connections to the VIP when the former master server comes back alive. The configuration also includes process tracking to reduce the priority of a server when the target service crashes.

Failover with load-balancing

Keepalived also provides load balancing with a number of possible algorithms (see keepalived.conf(5)).

This would be a simple example for HTTP load balancing that can be added to the above configuration:

/etc/keepalived/keepalived.conf

virtual_server 172.16.0.100 80 {
    delay_loop 6
    lb_algo rr
    lb_kind NAT
    protocol TCP

    real_server 172.16.1.20 80 {
        TCP_CHECK {
                connect_timeout 10
        }
    }
    real_server 172.16.1.21 80 {
        TCP_CHECK {
                connect_timeout 10
        }
    }
    real_server 172.16.1.22 80 {
        TCP_CHECK {
                connect_timeout 10
        }
    }
    real_server 172.16.1.23 80 {
        TCP_CHECK {
                connect_timeout 10
        }
    }
}

Alternatively, keepalived can also be used for failover with HAproxy acting as the loadbalancer.

Security considerations

The VRRP protocol used by keepalived does not provide significant security against attackers who are already on the same subnet. The main benefit of using authentication in this service is to protect against accidentally adding servers that disrupt an existing cluster. When using auth_type PASS, the password is sent in plaintext over the local subnet with every multicast packet. Using auth_type AH (IP Authentication Header) will provide slightly better resilience against attacks but basic disruption and attacks such as ARP spoofing are still possible. Use network restrictions to protect the VRRP traffic if security is a high priority for you, e.g. use an isolated subnet for the cluster and block VRRP traffic from other networks in your firewall.

See also

• Load Balancer Administration
• Configuring, Attacking and Securing VRRP on Linux