一、概述 1.简述： MariaDB Galera Cluster 是一套在mysql innodb存储引擎上面实现multi-master及数据实时同步的系统架构，业务层面无需做读写分离工作，数据库读写压力都能按照既定的规则分发到各个节点上去。在数据方面完全兼容 MariaDB、Percona Server和
一、概述
1.简述： MariaDB Galera Cluster 是一套在mysql innodb存储引擎上面实现multi-master及数据实时同步的系统架构，业务层面无需做读写分离工作，数据库读写压力都能按照既定的规则分发到各个节点上去。在数据方面完全兼容 MariaDB、Percona Server和MySQL。
2.特性：
(1).同步复制 Synchronous replication
(2).Active-active multi-master 拓扑逻辑
(3).可对集群中任一节点进行数据读写
(4).自动成员控制，故障节点自动从集群中移除
(5).自动节点加入
(6).真正并行的复制，基于行级
(7).直接客户端连接，原生的 MySQL 接口
(8).每个节点都包含完整的数据副本
(9).多台数据库中数据同步由 wsrep 接口实现
3.局限性:
(1).目前的复制仅仅支持InnoDB存储引擎,任何写入其他引擎的表，包括mysql.*表将不会复制,但是DDL语句会被复制的,因此创建用户将会被复制,但是insert into mysql.user…将不会被复制的 (2).DELETE操作不支持没有主键的表,没有主键的表在不同的节点顺序将不同,如果执行SELECT…LIMIT… 将出现不同的结果集
(3).在多主环境下LOCK/UNLOCK TABLES不支持,以及锁函数GET_LOCK(), RELEASE_LOCK()…
(4).查询日志不能保存在表中。如果开启查询日志，只能保存到文件中
(5).允许最大的事务大小由wsrep_max_ws_rows和wsrep_max_ws_size定义。任何大型操作将被拒绝。如大型的LOAD DATA操作
(6).由于集群是乐观的并发控制，事务commit可能在该阶段中止。如果有两个事务向在集群中不同的节点向同一行写入并提交，失败的节点将中止。对于集群级别的中止，集群返回死锁错误代码(Error: 1213 SQLSTATE: 40001 (ER_LOCK_DEADLOCK))
(7).XA事务不支持，由于在提交上可能回滚
(8).整个集群的写入吞吐量是由最弱的节点限制，如果有一个节点变得缓慢，那么整个集群将是缓慢的。为了稳定的高性能要求，所有的节点应使用统一的硬件
(9).集群节点建议最少3个
(10).如果DDL语句有问题将破坏集群。
二、架构介绍
1.Keepalived+LVS的经典组合作为前端负载均衡和高可用保障，可以使用单独两台主机分别作为主、备，如果数据库集群数量不多，比如两台，也可以直接在数据库主机上使用此组合
2.一共5台主机，2台作为keepalived+LVS的主备，另外三台分别为mdb1、mdb2和mdb3，mdb1作为参考节点，不执行任何客户端SQL，这样做的好处有如下几条:
（1）.数据一致性:因为"参考节点"本身不执行任何客户端SQL,所以在这个节点上发生transaction冲突的可能性最小。因此如果发现集群有数据不一致的时候，"参考节点"上的数据应该是集群中最准确的。
（2）.数据安全性:因为"参考节点"本身不执行任何客户端SQL,所以在这个节点上发生灾难事件的可能性最小。因此当整个集群宕掉的时候,"参考节点"应该是恢复集群的最佳节点。
（3）.高可用："参考节点"可以作为专门state snapshot donor。因为"参考节点"不服务于客户端，因此当使用此节点进行SST的时候，不会影响用户体验，并且前端的负载均衡设备也不需要重新配置。
三、 环境准备
1.系统和软件
| 系统环境 | |
| 系统 |
CentOS release 6.5 |
| 系统位数 | x86_64 |
| 内核版本 |
2.6.32-431 |
| 软件版本 | |
| Keepalived |
1.2.13 |
| LVS | 1.24 |
| MaridDB | 10.0.16 |
| socat | 1.7.3.0 |
2.主机环境
| mdb1（参考点） | 172.16.21.180 |
| mdb2 | 172.16.21.181 |
| mdb3 | 172.16.21.182 |
| ha1（keepalived+lvs主） | 172.16.21.201 |
| ha2（keepalived+lvs备） | 172.16.21.202 |
| VIP | 172.16.21.188 |
四、 集群安装配置
以主机mdb1为例：
1.配置hosts文件
编辑/etc/hosts加入下列内容
[root@mdb1 ~]# vi /etc/hosts 127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4 ::1 localhost localhost.localdomain localhost6 localhost6.localdomain6 172.16.21.201 ha1 172.16.21.202 ha2 172.16.21.180 mdb1 172.16.21.181 mdb2 172.16.21.182 mdb3
2. 准备YUM源
除了系统自带的官方源，再添加epel,Percona,MariaDB的源
[root@mdb1 ~]# vi /etc/yum.repos.d/MariDB.repo # MariaDB 5.5 RedHat repository list - created 2015-03-04 02:45 UTC # http://mariadb.org/mariadb/repositories/ [mariadb] name = MariaDB baseurl = http://yum.mariadb.org/10.0/rhel6-amd64 gpgkey=https://yum.mariadb.org/RPM-GPG-KEY-MariaDB gpgcheck=1 [root@mdb1 ~]#rpm -ivh http://dl.fedoraproject.org/pub/epel/6/x86_64/epel-release-6-8.noarch.rpm [root@mdb1 ~]#rpm --import https://yum.mariadb.org/RPM-GPG-KEY-MariaDB [root@mdb1 ~]# vi /etc/yum.repos.d/Percona.repo [percona] name = CentOS $releasever - Percona baseurl=http://repo.percona.com/centos/$releasever/os/$basearch/ enabled = 1 gpgkey = file:///etc/pki/rpm-gpg/RPM-GPG-KEY-percona gpgcheck = 1 [root@mdb1 ~]#wget -O /etc/pki/rpm-gpg/RPM-GPG-KEY-percona http://www.percona.com/downloads/RPM-GPG-KEY-percona [root@mdb1 ~]#yum clean all
3.安装socat
socat是一个多功能的网络工具，名字来由是”Socket CAT”，可以看作是netcat的N倍加强版。
事实证明，如果不安装socat，MariaDB-Galera-server最后的数据同步会失败报错，网上很多配置文档都没有讲到这点，请记住一定要安装
[root@mdb1 ~]# tar -xzvf socat-1.7.3.0.tar.gz [root@mdb1 ~]# cd socat-1.7.3.0 [root@mdb1 socat-1.7.3.0]# ./configure --prefix=/usr/local/socat [root@mdb1 socat-1.7.3.0]# make && make install [root@mdb1 socat-1.7.3.0]# ln -s /usr/local/socat/bin/socat /usr/sbin/
4.安装MariaDB、galera、xtrabackup
[root@mdb1 ~]# rpm -e --nodeps mysql-libs [root@mdb1 ~]# yum install MariaDB-Galera-server galera MariaDB-client xtrabackup [root@mdb1 ~]#chkconfig mysql on [root@mdb1 ~]#service mysql start
5.置MariaDB的root密码，并做安全加固
[root@mdb1 ~]#/usr/bin/mysql_secure_installation
6.创建用于同步数据库的SST帐号
[root@mdb1 ~]# mysql -uroot -p Enter password: Welcome to the MariaDB monitor. Commands end with ; or \g. Your MariaDB connection id is 12 Server version: 10.0.16-MariaDB-wsrep-log MariaDB Server, wsrep_25.10.r4144 Copyright (c) 2000, 2015, Oracle, MariaDB Corporation Ab and others. Type 'help;' or '\h' for help. Type '\c' to clear the current input statement. MariaDB [(none)]> grant all privileges on *.* to sst@'%' identified by '123456'; MariaDB [(none)]> flush privileges; MariaDB [(none)]> quit
7.创建wsrep.cnf文件
[root@mdb1 ~]#cp /usr/share/mysql/wsrep.cnf /etc/my.cnf.d/ [root@mdb1 ~]# vi /etc/my.cnf.d/wsrep.cnf 只需要修改如下4行: wsrep_provider=/usr/lib64/galera/libgalera_smm.so wsrep_cluster_address="gcomm://" wsrep_sst_auth=sst:123456 wsrep_sst_method=xtrabackup
注意:
"gcomm://" 是特殊的地址,仅仅是Galera cluster初始化启动时候使用。
如果集群启动以后，我们关闭了第一个节点，那么再次启动的时候必须先修改
"gcomm://"为其他节点的集群地址,例如下次启动时需要更改
wsrep_cluster_address="gcomm://172.16.21.182:4567"
650) this.width=650;" src="http://www.68idc.cn/help/uploads/allimg/151111/12133C510-1.jpg" title="704836f4t7c7d58fd365e&690.png" alt="wKiom1T4DgCS9gylAAD_erOXuYo868.jpg" />
图中的Node A就是我们的mdb1,Node N就是后面需要添加的主机mdb3
8.修改/etc/my.cnf
添加如下一行
!includedir /etc/my.cnf.d/
另外最好在/etc/my.cnf中指定datadir路径
datadir = /var/lib/mysql
否则可能会遇到报错说找不到路径，所以最好加上这条
9.关闭防火墙iptables和selinux
很多人在启动数据库集群时总是失败，很可能就是因为防火墙没有关闭或者没有打开相应端口，最好的办法就是清空iptables并关闭selinux
[root@mdb1 ~]# iptables -F [root@mdb1 ~]# iptables-save > /etc/sysconfig/iptables [root@mdb1 ~]# setenforce 0 [root@mdb1 ~]# vi /etc/selinux/config # This file controls the state of SELinux on the system. # SELINUX= can take one of these three values: # enforcing - SELinux security policy is enforced. # permissive - SELinux prints warnings instead of enforcing. # disabled - No SELinux policy is loaded. SELINUX=disabled # SELINUXTYPE= can take one of these two values: # targeted - Targeted processes are protected, # mls - Multi Level Security protection. SELINUXTYPE=targeted
10.重启MariaDB
[root@mdb1 ~]# service mysql restart [root@mdb1 ~]# netstat -tulpn | grep -e 4567 -e 3306 tcp 0 0 0.0.0.0:4567 0.0.0.0:* LISTEN 11325/mysqld tcp 0 0 0.0.0.0:3306 0.0.0.0:* LISTEN 11325/mysqld
到此，单节点的配置完成
11.添加mdb2、mdb3到集群
整个集群就是首位相连，简单说来就是在"gcomm://"处的IP不一样，mdb3—>mdb2—>mdb1—>mdb3，在生产环境，可以考虑将mdb1作为参考节点，不执行客户端的SQL，用来保障数据一致性和数据恢复时用。具体构造方法如下：
（1）按照上述1-10的步骤安装和配置另外两条主机
（2）除了第7步wsrep_cluster_address要改为对应的主机地址
mdb2：wsrep_cluster_address="gcomm://172.16.21.180:4567"
mdb3：wsrep_cluster_address="gcomm://172.16.21.181:4567"
如果有更多主机要加入集群，以此类推，将wsrep_cluster_address指向前一个主机地址，而集群第一台主机指向最后一台的地址就行了
12.最后将mdb2和mdb3启动
[root@mdb2 ~]# service mysql start [root@mdb3 ~]# service mysql start
13.给集群加入Galera arbitrator
对于只有2个节点的Galera Cluster和其他集群软件一样，需要面对极端情况下的"脑裂"状态。
为了避免这种问题，Galera引入了"arbitrator(仲裁人)"。
"仲裁人"节点上没有数据,它在集群中的作用就是在集群发生分裂时进行仲裁，集群中可以有多个"仲裁人"节点。
"仲裁人"节点加入集群的方法很简单,运行如下命令即可:
[root@mdb1 ~]# garbd -a gcomm://172.16.21.180:4567 -g my_wsrep_cluster -d
参数说明:
-d 以daemon模式运行
-a 集群地址
-g 集群名称
14.确认galera集群正确安装和运行
MariaDB [(none)]> show status like 'ws%'; +------------------------------+----------------------------------------------------------+ | Variable_name | Value | +------------------------------+----------------------------------------------------------+ | wsrep_local_state_uuid | 64784714-c23a-11e4-b7d7-5edbdea0e62c uuid 集群唯一标记 | | wsrep_protocol_version | 5 | | wsrep_last_committed | 94049 sql 提交记录 | | wsrep_replicated | 0 | | wsrep_replicated_bytes | 0 | | wsrep_repl_keys | 0 | | wsrep_repl_keys_bytes | 0 | | wsrep_repl_data_bytes | 0 | | wsrep_repl_other_bytes | 0 | | wsrep_received | 3 | | wsrep_received_bytes | 287 | | wsrep_local_commits | 0 本地执行的 sql | | wsrep_local_cert_failures | 0 本地失败事务 | | wsrep_local_replays | 0 | | wsrep_local_send_queue | 0 | | wsrep_local_send_queue_avg | 0.333333 队列平均时间间隔 | | wsrep_local_recv_queue | 0 | | wsrep_local_recv_queue_avg | 0.000000 | | wsrep_local_cached_downto | 18446744073709551615 | | wsrep_flow_control_paused_ns | 0 | | wsrep_flow_control_paused | 0.000000 | | wsrep_flow_control_sent | 0 | | wsrep_flow_control_recv | 0 | | wsrep_cert_deps_distance | 0.000000 并发数量 | | wsrep_apply_oooe | 0.000000 | | wsrep_apply_oool | 0.000000 | | wsrep_apply_window | 0.000000 | | wsrep_commit_oooe | 0.000000 | | wsrep_commit_oool | 0.000000 | | wsrep_commit_window | 0.000000 | | wsrep_local_state | 4 | | wsrep_local_state_comment | Synced | | wsrep_cert_index_size | 0 | | wsrep_causal_reads | 0 | | wsrep_cert_interval | 0.000000 | | wsrep_incoming_addresses | 172.16.21.180:3306,172.16.21.182:3306,172.16.21.188:3306 | | wsrep_cluster_conf_id | 19 | | wsrep_cluster_size | 3 集群成员个数 | | wsrep_cluster_state_uuid | 64784714-c23a-11e4-b7d7-5edbdea0e62c | | wsrep_cluster_status | Primary 主服务器 | | wsrep_connected | ON 当前是否连接中 | | wsrep_local_bf_aborts | 0 | | wsrep_local_index | 0 | | wsrep_provider_name | Galera | | wsrep_provider_vendor | Codership Oy <info@codership.com> | | wsrep_provider_version | 25.3.5(rXXXX) | | wsrep_ready | ON | | wsrep_thread_count | 3 | +------------------------------+----------------------------------------------------------+
wsrep_ready为ON,则说明MariaDB Galera集群已经正确运行了
监控状态说明：
（1）集群完整性检查:
wsrep_cluster_state_uuid:在集群所有节点的值应该是相同的,有不同值的节点,说明其没有连接入集群.
wsrep_cluster_conf_id:正常情况下所有节点上该值是一样的.如果值不同,说明该节点被临时”分区”了.当节点之间网络连接恢复的时候应该会恢复一样的值.
wsrep_cluster_size:如果这个值跟预期的节点数一致,则所有的集群节点已经连接.
wsrep_cluster_status:集群组成的状态.如果不为”Primary”,说明出现”分区”或是”split-brain”状况.
（2）节点状态检查:
wsrep_ready: 该值为ON,则说明可以接受SQL负载.如果为Off,则需要检查wsrep_connected.
wsrep_connected: 如果该值为Off,且wsrep_ready的值也为Off,则说明该节点没有连接到集群.(可能是wsrep_cluster_address或wsrep_cluster_name等配置错造成的.具体错误需要查看错误日志)
wsrep_local_state_comment:如果wsrep_connected为On,但wsrep_ready为OFF,则可以从该项查看原因.
（3）复制健康检查:
wsrep_flow_control_paused:表示复制停止了多长时间.即表明集群因为Slave延迟而慢的程度.值为0~1,越靠近0越好,值为1表示复制完全停止.可优化wsrep_slave_threads的值来改善.
wsrep_cert_deps_distance:有多少事务可以并行应用处理.wsrep_slave_threads设置的值不应该高出该值太多.
wsrep_flow_control_sent:表示该节点已经停止复制了多少次.
wsrep_local_recv_queue_avg:表示slave事务队列的平均长度.slave瓶颈的预兆.
最慢的节点的wsrep_flow_control_sent和wsrep_local_recv_queue_avg这两个值最高.这两个值较低的话,相对更好.
（4）检测慢网络问题:
wsrep_local_send_queue_avg:网络瓶颈的预兆.如果这个值比较高的话,可能存在网络瓶
冲突或死锁的数目:
wsrep_last_committed:最后提交的事务数目
wsrep_local_cert_failures和wsrep_local_bf_aborts:回滚,检测到的冲突数目
15.测试数据是否能同步
分别在每个节点创建库和表，再删除，查看其它节点是否同步，如若配置正确，应该是同步的，具体操作省略
五、 Keepalived+LVS配置
1.使用YUM方式安装
[root@ha1 ~]# yum install keepalived ipvsadm [root@ha2 ~]# yum install keepalived ipvsadm
2.Keepalived配置
主机ha1的配置
[root@ha1 ~]# vi /etc/keepalived/keepalived.conf
global_defs {
notification_email {
xx@xxxx.com
}
notification_email_from root@localhost
smtp_server 127.0.0.1
smtp_connect_timeout 30
router_id LVS_201
}
vrrp_instance VI_1 {
state MASTER
interface eth0
virtual_router_id 51
priority 100
advert_int 1
authentication {
auth_type PASS
auth_pass 1111
}
virtual_ipaddress {
172.16.21.188/24 dev eth0 label eth0:0
}
}
virtual_server 172.16.21.188 3306 {
delay_loop 6
lb_algo rr
lb_kind DR
nat_mask 255.255.255.0
persistence_timeout 50
protocol TCP
real_server 172.16.21.181 3306 {
weight 1
TCP_CHECK {
connect_timeout 3
nb_get_retry 3
delay_before_retry 3
connect_port 3306
}
}
real_server 172.16.21.182 3306 {
weight 1
TCP_CHECK {
connect_timeout 3
nb_get_retry 3
delay_before_retry 3
connect_port 3306
}
}
} 备机ha2的配置
global_defs {
notification_email {
xx@xxxx.com
}
notification_email_from root@localhost
smtp_server 127.0.0.1
smtp_connect_timeout 30
router_id LVS_202
}
vrrp_instance VI_1 {
state BACKUP
interface eth0
virtual_router_id 51
priority 99
advert_int 1
authentication {
auth_type PASS
auth_pass 1111
}
virtual_ipaddress {
172.16.21.188/24 dev eth0 label eth0:0
}
}
virtual_server 172.16.21.188 3306 {
delay_loop 6
lb_algo rr
lb_kind DR
nat_mask 255.255.255.0
persistence_timeout 50
protocol TCP
real_server 172.16.21.181 3306 {
weight 1
TCP_CHECK {
connect_timeout 3
nb_get_retry 3
delay_before_retry 3
connect_port 3306
}
}
real_server 172.16.21.182 3306 {
weight 1
TCP_CHECK {
connect_timeout 3
nb_get_retry 3
delay_before_retry 3
connect_port 3306
}
}
} 3.LVS脚本配置
两台realserver服务器上都要配置如下脚本
[root@mdb2 ~]#vi /etc/init.d/lvsdr.sh
#!/bin/bash
# description: Config realserver lo and apply noarp
VIP=172.16.21.188
. /etc/rc.d/init.d/functions
case "$1" in
start)
/sbin/ifconfig lo down
/sbin/ifconfig lo up
echo "1" >/proc/sys/net/ipv4/conf/lo/arp_ignore
echo "2" >/proc/sys/net/ipv4/conf/lo/arp_announce
echo "1" >/proc/sys/net/ipv4/conf/all/arp_ignore
echo "2" >/proc/sys/net/ipv4/conf/all/arp_announce
/sbin/sysctl -p >/dev/null 2>&1
/sbin/ifconfig lo:0 $VIP netmask 255.255.255.255 up
/sbin/route add -host $VIP dev lo:0
echo "LVS-DR real server starts successfully."
;;
stop)
/sbin/ifconfig lo:0 down
/sbin/route del $VIP >/dev/null 2>&1
echo "0" >/proc/sys/net/ipv4/conf/lo/arp_ignore
echo "0" >/proc/sys/net/ipv4/conf/lo/arp_announce
echo "0" >/proc/sys/net/ipv4/conf/all/arp_ignore
echo "0" >/proc/sys/net/ipv4/conf/all/arp_announce
echo "LVS-DR real server stopped."
;;
status)
isLoOn=`/sbin/ifconfig lo:0 | grep "$VIP"`
isRoOn=`/bin/netstat -rn | grep "$VIP"`
if [ "$isLoOn" == "" -a "$isRoOn" == "" ]; then
echo "LVS-DR real server has to run yet."
else
echo "LVS-DR real server is running."
fi
exit 3
;;
*)
echo "Usage: $0 {start|stop|status}"
exit 1
esac
exit 0
[root@mdb2 ~]# chmod +x /etc/init.d/lvsdr.sh
[root@mdb3 ~]# chmod +x /etc/init.d/lvsdr.sh 4.启动Keepalived和LVS
[root@mdb2 ~]# /etc/init.d/lvsdr.sh start [root@mdb3 ~]# /etc/init.d/lvsdr.sh start [root@ha1 ~]# service keepalived start [root@ha2 ~]# service keepalived start
5.加入开机自动启动
[root@mdb2 ~]#echo "/etc/init.d/lvsdr.sh start" >> /etc/rc.d/rc.local [root@mdb3 ~]#echo "/etc/init.d/lvsdr.sh start" >> /etc/rc.d/rc.local [root@ha1 ~]# chkconfig keepalived on [root@ha2 ~]# chkconfig keepalived on
6.测试
将主服务器ha1的keepalived关闭，在备机ha2上观察日志和IP变化
[root@ha1 ~]#service keepalived stop
[root@ha2 ~]#tail -f /var/log/messages
Mar 5 10:36:03 ha2 Keepalived_healthcheckers[11249]: Opening file '/etc/keepalived/keepalived.conf'.
Mar 5 10:36:03 ha2 Keepalived_healthcheckers[11249]: Configuration is using : 14697 Bytes
Mar 5 10:36:03 ha2 Keepalived_vrrp[11250]: Opening file '/etc/keepalived/keepalived.conf'.
Mar 5 10:36:03 ha2 Keepalived_vrrp[11250]: Configuration is using : 63250 Bytes
Mar 5 10:36:03 ha2 Keepalived_vrrp[11250]: Using LinkWatch kernel netlink reflector...
Mar 5 10:36:03 ha2 Keepalived_healthcheckers[11249]: Using LinkWatch kernel netlink reflector...
Mar 5 10:36:03 ha2 Keepalived_healthcheckers[11249]: Activating healthchecker for service [172.16.21.181]:3306
Mar 5 10:36:03 ha2 Keepalived_healthcheckers[11249]: Activating healthchecker for service [172.16.21.182]:3306
Mar 5 10:36:03 ha2 Keepalived_vrrp[11250]: VRRP_Instance(VI_1) Entering BACKUP STATE
Mar 5 10:36:03 ha2 Keepalived_vrrp[11250]: VRRP sockpool: [ifindex(2), proto(112), unicast(0), fd(10,11)]
Mar 6 08:41:53 ha2 Keepalived_vrrp[11250]: VRRP_Instance(VI_1) Transition to MASTER STATE
Mar 6 08:41:54 ha2 Keepalived_vrrp[11250]: VRRP_Instance(VI_1) Entering MASTER STATE
Mar 6 08:41:54 ha2 Keepalived_vrrp[11250]: VRRP_Instance(VI_1) setting protocol VIPs.
Mar 6 08:41:54 ha2 Keepalived_vrrp[11250]: VRRP_Instance(VI_1) Sending gratuitous ARPs on eth0 for 172.16.21.188
Mar 6 08:41:54 ha2 Keepalived_healthcheckers[11249]: Netlink reflector reports IP 172.16.21.188 added
Mar 6 08:41:59 ha2 Keepalived_vrrp[11250]: VRRP_Instance(VI_1) Sending gratuitous ARPs on eth0 for 172.16.21.188
[root@ha2 ~]#ifconfig
eth0 Link encap:Ethernet HWaddr 00:0C:29:1D:77:9C
inet addr:172.16.21.202 Bcast:172.16.21.255 Mask:255.255.255.0
inet6 addr: fe80::20c:29ff:fe1d:779c/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:2969375670 errors:0 dropped:0 overruns:0 frame:0
TX packets:2966841735 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:225643845081 (210.1 GiB) TX bytes:222421642143 (207.1 GiB)
eth0:0 Link encap:Ethernet HWaddr 00:0C:29:1D:77:9C
inet addr:172.16.21.188 Bcast:0.0.0.0 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
inet6 addr: ::1/128 Scope:Host
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:55694 errors:0 dropped:0 overruns:0 frame:0
TX packets:55694 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:3176387 (3.0 MiB) TX bytes:3176387 (3.0 MiB) 将ha1的keepalived启动再观察ha1的日志和IP
[root@ha1 ~]#service keepalived start
[root@ha1 ~]#tail -f /var/log/messages
Mar 6 08:54:42 ha1 Keepalived[13310]: Starting Keepalived v1.2.13 (10/15,2014)
Mar 6 08:54:42 ha1 Keepalived[13311]: Starting Healthcheck child process, pid=13312
Mar 6 08:54:42 ha1 Keepalived[13311]: Starting VRRP child process, pid=13313
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Netlink reflector reports IP 172.16.21.181 added
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Netlink reflector reports IP fe80::20c:29ff:fe4d:8e83 added
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Netlink reflector reports IP 172.16.21.181 added
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Registering Kernel netlink reflector
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Netlink reflector reports IP fe80::20c:29ff:fe4d:8e83 added
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Registering Kernel netlink command channel
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Registering gratuitous ARP shared channel
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Registering Kernel netlink reflector
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Registering Kernel netlink command channel
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Opening file '/etc/keepalived/keepalived.conf'.
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Opening file '/etc/keepalived/keepalived.conf'.
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Configuration is using : 63252 Bytes
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: Using LinkWatch kernel netlink reflector...
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Configuration is using : 14699 Bytes
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Using LinkWatch kernel netlink reflector...
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Activating healthchecker for service [172.16.21.181]:3306
Mar 6 08:54:42 ha1 Keepalived_healthcheckers[13312]: Activating healthchecker for service [172.16.21.182]:3306
Mar 6 08:54:42 ha1 Keepalived_vrrp[13313]: VRRP sockpool: [ifindex(2), proto(112), unicast(0), fd(10,11)]
Mar 6 08:54:43 ha1 Keepalived_vrrp[13313]: VRRP_Instance(VI_1) Transition to MASTER STATE
Mar 6 08:54:43 ha1 Keepalived_vrrp[13313]: VRRP_Instance(VI_1) Received lower prio advert, forcing new election
Mar 6 08:54:44 ha1 Keepalived_vrrp[13313]: VRRP_Instance(VI_1) Entering MASTER STATE
Mar 6 08:54:44 ha1 Keepalived_vrrp[13313]: VRRP_Instance(VI_1) setting protocol VIPs.
Mar 6 08:54:44 ha1 Keepalived_vrrp[13313]: VRRP_Instance(VI_1) Sending gratuitous ARPs on eth0 for 172.16.21.188
Mar 6 08:54:44 ha1 Keepalived_healthcheckers[13312]: Netlink reflector reports IP 172.16.21.188 added
Mar 6 08:54:49 ha1 Keepalived_vrrp[13313]: VRRP_Instance(VI_1) Sending gratuitous ARPs on eth0 for 172.16.21.188
[root@ha1 ~]#ifconfig
eth0 Link encap:Ethernet HWaddr 00:0C:29:4D:8E:83
inet addr:172.16.21.201 Bcast:172.16.21.255 Mask:255.255.255.0
inet6 addr: fe80::20c:29ff:fe4d:8e83/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:2968402607 errors:0 dropped:0 overruns:0 frame:0
TX packets:2966256067 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:224206102960 (208.8 GiB) TX bytes:221258814612 (206.0 GiB)
eth0:0 Link encap:Ethernet HWaddr 00:0C:29:4D:8E:83
inet addr:172.16.21.188 Bcast:0.0.0.0 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
inet6 addr: ::1/128 Scope:Host
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:54918 errors:0 dropped:0 overruns:0 frame:0
TX packets:54918 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:3096422 (2.9 MiB) TX bytes:3096422 (2.9 MiB)到此，所有配置就完成了
六、 总结
提到MySQL多主复制，大家很可能都是想到MySQL+MMM的架构，MariaDB Galera Cluster很好地替代了前者并且可靠性更高，具体比较可以参考http://www.oschina.net/translate/from-mysql-mmm-to-mariadb-galera-cluster-a-high-availability-makeover这篇文章。
当然，MariaDB Galera Cluster并不是适合所有需要复制的情形，你必须根据自己的需求来决定，比如，如果你是数据一致性考虑的多，而且写操作和更新的东西多，但写入量不是很大，MariaDB Galera Cluster就适合你；如果你是查询的多，且读写分离也容易实现，那就用replication好，简单易用，用一个master保证数据的一致性，可以有多个slave用来读去数据，分担负载，只要能解决好数据一致性和唯一性，replication就更适合你，毕竟MariaDB Galera Cluster集群遵循“木桶”原理，如果写的量很大，数据同步速度是由集群节点中IO最低的节点决定的，整体上，写入的速度会比replication慢许多。
如果文中有任何遗漏和错误
