双机高可用、负载均衡、MySQL(读写分离、主从自动切换)架构设计

前几天网友来信说帮忙实现这样一个架构：只有两台机器，需要实现其中一台死机之后另一台能接管这台机器的服务，并且在两台机器正常服务时，两台机器都能用上。于是设计了如下的架构。

架构简介

此架构主要是由keepalived实现双机高可用，维护了一个外网VIP,一个内网VIP。正常情况时，外网VIP和内网VIP都绑定在server1服务器，web请求发送到server1的Nginx，nginx对于静态资源请求就直接在本机检索并返回，对于PHP的动态请求，则负载均衡到server1和server2。对于SQL请求，会将此类请求发送到Atlas mysql中间件，Atlas接收到请求之后，把涉及写操作的请求发送到内网VIP，读请求操作发送到server2，这样就实现了读写分离。

当主服务器server1宕机时，keepalived检测到后，立即把外网VIP和内网VIP绑定到server2，并把server2的mysql切换成主库。此时由于外网VIP已经转移到了server2，web请求将发送给server2的nginx。nginx检测到server1宕机，不再把请求转发到server1的php-fpm。之后的sql请求照常发送给本地的atlas，atlas把写操作发送给内网VIP，读操作发送给server2 mysql，由于内网VIP已经绑定到server2了，server2的mysql同时接受写操作和读操作。

当主服务器server1恢复后，keepalived不抢占server2的VIP，继续正常服务。我们可以把server1的mysql切换成主，也可以切换成从。

架构要求

要实现此架构，需要三个条件：

服务器可以设置内网ip，并且设置的内网IP互通;
服务器可以随意绑定IDC分配给我们使用的外网IP，即外网IP没有绑定MAC地址;
MySQL服务器支持GTID，即MySQL-5.6.5以上版本。

环境说明

server1

eth0: 10.96.153.110(对外IP)

eth1: 192.168.3.100(对内IP)

server2

eth0: 10.96.153.114(对外IP)

eth1: 192.168.3.101(对内IP)

系统都是CentOS-6。

对外VIP: 10.96.153.239

对内VIP: 192.168.3.150

hosts设置

/etc/hosts:

192.168.3.100 server1

192.168.3.101 server2

Nginx PHP MySQL安装

这几个软件的安装推荐使用EZHTTP来完成。

Nginx配置

Server1配置

http {

[...]

    upstream php-server {

           server 192.168.3.101:9000;

           server 127.0.0.1:9000;

           keepalive 100;

    }

[...]

 server {

    [...]

        location ~ \.php$ {

                        fastcgi_pass   php-server;

                        fastcgi_index  index.php;

                        fastcgi_param  SCRIPT_FILENAME  $document_root$fastcgi_script_name;

                        include        fastcgi_params;

        }

    [...]

 }

[...]

}

Server2配置

http {

[...]

    upstream php-server {

           server 192.168.3.100:9000;

           server 127.0.0.1:9000;

           keepalive 100;

    }

[...]

 server {

    [...]

        location ~ \.php$ {

                        fastcgi_pass   php-server;

                        fastcgi_index  index.php;

                        fastcgi_param  SCRIPT_FILENAME  $document_root$fastcgi_script_name;

                        include        fastcgi_params;

        }

    [...]

 }

[...]

}

这两个配置主要的作用是设置php请求的负载均衡。

MySQL配置

mysql util安装

我们需要安装mysql util里的主从配置工具来实现主从切换。

cd /tmp

wget http://dev.mysql.com/get/Downloads/MySQLGUITools/mysql-utilities-1.5.3.tar.gz

tar xzf mysql-utilities-1.5.3.tar.gz

cd mysql-utilities-1.5.3

python setup.py build

python setup.py install

mysql my.cnf配置

server1:

[mysql]

[...]

protocol=tcp

[...]

[...]

[mysqld]

[...]

# BINARY LOGGING #

log-bin = /usr/local/mysql/data/mysql-bin

expire-logs-days = 14

sync-binlog = 1

binlog-format=ROW

log-slave-updates=true

gtid-mode=on

enforce-gtid-consistency =true

master-info-repository=TABLE

relay-log-info-repository=TABLE

sync-master-info=1

server-id=1

report-host=server1

report-port=3306

[...]

server2:

[mysql]

[...]

protocol=tcp

[...]

[mysqld]

[...]

# BINARY LOGGING #

log-bin = /usr/local/mysql/data/mysql-bin

expire-logs-days = 14

sync-binlog = 1

binlog-format=ROW

log-slave-updates=true

gtid-mode=on

enforce-gtid-consistency =true

master-info-repository=TABLE

relay-log-info-repository=TABLE

sync-master-info=1

server-id=2

report-host=server2

report-port=3306

[...]

这两个配置主要是设置了binlog和启用gtid-mode，并且需要设置不同的server-id和report-host。

开放root帐号远程权限：

我们需要在两台mysql服务器设置root帐号远程访问权限。

mysql> grant all on *.* to 'root'@'192.168.3.%' identified by 'Xp29at5F37' with grant option;

mysql> grant all on *.* to 'root'@'server1' identified by 'Xp29at5F37' with grant option;

mysql> grant all on *.* to 'root'@'server2' identified by 'Xp29at5F37' with grant option;

mysql> flush privileges;

设置mysql主从

在任意一台执行如下命令：

mysqlreplicate --master=root:Xp29at5F37@server1:3306 --slave=root:Xp29at5F37@server2:3306 --rpl-user=rpl:o67DhtaW

# master on server1: ... connected.

# slave on server2: ... connected.

# Checking for binary logging on master...

# Setting up replication...

# ...done.

显示主从关系

mysqlrplshow --master=root:Xp29at5F37@server1 --discover-slaves-login=root:Xp29at5F37

# master on server1: ... connected.

# Finding slaves for master: server1:3306

# Replication Topology Graph

server1:3306 (MASTER)

|

+--- server2:3306 - (SLAVE)

检查主从状态

mysqlrplcheck --master=root:Xp29at5F37@server1 --slave=root:Xp29at5F37@server2

# master on server1: ... connected.

# slave on server2: ... connected.

test Description Status

---------------------------------------------------------------------------

Checking for binary logging on master [pass]

Are there binlog exceptions? [pass]

Replication user exists? [pass]

Checking server_id values [pass]

Checking server_uuid values [pass]

Is slave connected to master? [pass]

Check master information file [pass]

Checking InnoDB compatibility [pass]

Checking storage engines compatibility [pass]

Checking lower_case_table_names settings [pass]

Checking slave delay (seconds behind master) [pass]

# ...done.

在server2建立主从切换脚本

vi /data/sh/mysqlfailover.sh

#!/bin/bash

mysqlrpladmin --slave=root:Xp29at5F37@server2:3306 failover

chmod +x /data/sh/mysqlfailover.sh

Keepalived配置

keepalived安装(两台都装)

yum -y install keepalived

chkconfig keepalived on

keepalived配置(server1)

vi /etc/keepalived/keepalived.conf

vrrp_sync_group VG_1 {

group {

inside_network

outside_network

}

}

vrrp_instance inside_network {

state BACKUP

interface eth1

virtual_router_id 51

priority 101

advert_int 1

authentication {

auth_type PASS

auth_pass 3489

}

virtual_ipaddress {

192.168.3.150/24

}

nopreempt

}

vrrp_instance outside_network {

state BACKUP

interface eth0

virtual_router_id 50

priority 101

advert_int 1

authentication {

auth_type PASS

auth_pass 3489

}

virtual_ipaddress {

10.96.153.239/24

}

nopreempt

}

keepalived配置(server2)

vrrp_sync_group VG_1 {

group {

inside_network

outside_network

}

}

vrrp_instance inside_network {

state BACKUP

interface eth1

virtual_router_id 51

priority 100

advert_int 1

authentication {

auth_type PASS

auth_pass 3489

}

virtual_ipaddress {

192.168.3.150

}

notify_master /data/sh/mysqlfailover.sh

}

vrrp_instance outside_network {

state BACKUP

interface eth0

virtual_router_id 50

priority 100

advert_int 1

authentication {

auth_type PASS

auth_pass 3489

}

virtual_ipaddress {

10.96.153.239/24

}

}

此keepalived配置需要注意的是：

两台server的state都设置为backup，server1增加nopreempt配置，并且server1 priority比server2高，这样用来实现当server1从宕机恢复时，不抢占VIP;
server2设置notify_master /data/sh/mysqlfailover.sh,意味着server2接管server1后，执行这个脚本，以把server2的mysql提升为主。

Atlas设置

atlas安装

到这里下载最新版本，https://github.com/Qihoo360/Atlas/releases

cd /tmp

wget https://github.com/Qihoo360/Atlas/releases/download/2.2.1/Atlas-2.2.1.el6.x86_64.rpm

rpm -i Atlas-2.2.1.el6.x86_64.rpm

atlas配置

cd /usr/local/mysql-proxy/conf

cp test.cnf my.cnf

vi my.cnf

调整如下参数，

proxy-backend-addresses = 192.168.3.150:3306

proxy-read-only-backend-addresses = 192.168.3.101:3306

pwds = root:qtyU1btXOo074Itvx0UR9Q==

event-threads = 8

注意：

proxy-backend-addresse设置为内网VIP
proxy-read-only-backend-addresses设置为server2的IP
root:qtyU1btXOo074Itvx0UR9Q==设置数据库的用户和密码，密码是通过/usr/local/mysql-proxy/bin/encrypt Xp29at5F37生成。更详细参数解释请查看，Atlas配置详解。

启动atlas

/usr/local/mysql-proxy/bin/mysql-proxy --defaults-file=/usr/local/mysql-proxy/conf/my.cnf

之后程序里配置mysql就配置127.0.0.1:1234就好。

server1主宕机测试

测试keepalived是否工作正常，我们来模拟server1宕机。在server1上执行shutdown关机命令。此时我们登录server2，执行ip addr命令，输出如下：

1: lo: mtu 16436 qdisc noqueue state UNKNOWN

link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00

inet 127.0.0.1/8 scope host lo

inet6 ::1/128 scope host

valid_lft forever preferred_lft forever

2: eth0: mtu 1500 qdisc pfifo_fast state UP qlen 1000

link/ether 00:0c:29:81:9d:42 brd ff:ff:ff:ff:ff:ff

inet 10.96.153.114/24 brd 10.96.153.255 scope global eth0

inet 10.96.153.239/24 scope global secondary eth0

inet6 fe80::20c:29ff:fe81:9d42/64 scope link

valid_lft forever preferred_lft forever

3: eth1: mtu 1500 qdisc pfifo_fast state UP qlen 1000

link/ether 00:0c:29:81:9d:4c brd ff:ff:ff:ff:ff:ff

inet 192.168.3.101/24 brd 192.168.3.255 scope global eth1

inet 192.168.3.150/32 scope global eth1

inet6 fe80::20c:29ff:fe81:9d4c/64 scope link

valid_lft forever preferred_lft forever

我们看到对外VIP 10.96.153.239和对内IP 192.168.3.150已经转移到server2了，证明keepalived运行正常。

测试是否自动切换了主从，登录server2的mysql服务器，执行show status;命令，如下：

mysql> show slave statusG

Empty set (0.00 sec)

我们发现从状态已经为空，证明已经切换为主了。

测试server1是否抢占VIP，为什么要测试这个呢？如果server1恢复之后抢占了VIP，而我们的Atlas里后端设置的是VIP，这样server1启动之后，sql的写操作就会向server1的mysql发送，而server1的mysql数据是旧于server2的，所以这样会造成数据不一致，这个是非常重要的测试。

我们先来启动server1，之后执行ip addr，输出如下：

1: lo: mtu 16436 qdisc noqueue state UNKNOWN

link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00

inet 127.0.0.1/8 scope host lo

inet6 ::1/128 scope host

valid_lft forever preferred_lft forever

2: eth0: mtu 1500 qdisc pfifo_fast state UP qlen 1000

link/ether 00:0c:29:f1:4f:4e brd ff:ff:ff:ff:ff:ff

inet 10.96.153.110/24 brd 10.96.153.255 scope global eth0

inet6 fe80::20c:29ff:fef1:4f4e/64 scope link

valid_lft forever preferred_lft forever

3: eth1: mtu 1500 qdisc pfifo_fast state UP qlen 1000

link/ether 00:0c:29:f1:4f:58 brd ff:ff:ff:ff:ff:ff

inet 192.168.3.100/24 brd 192.168.3.255 scope global eth1

inet6 fe80::20c:29ff:fef1:4f58/64 scope link

valid_lft forever preferred_lft forever

我们看到，server1并没有抢占VIP，测试正常。不过另人郁闷的是，在虚拟机的环境并没有测试成功，不知道为什么。

如何恢复server1

设置server1 mysql为从，server1从宕机中恢复之后，mysql的数据已经旧于server2的数据了，这时我们先设置server1 mysql为从。

mysqlreplicate --master=root:Xp29at5F37@server2:3306 --slave=root:Xp29at5F37@server1:3306 --rpl-user=rpl:o67DhtaW

# master on server2: ... connected.

# slave on server1: ... connected.

# Checking for binary logging on master...

# Setting up replication...

# ...done.

看到提示是设置成功了。

获取server1 mysql数据数据同步情况，server1 mysql刚从宕机恢复，有可能数据远远落后于server2 mysql，所以我们先查看它们之间的数据同步情况。登录server1 mysql，执行如下sql:

mysql> show slave statusG

*************************** 1. row ***************************

Slave_IO_State: Waiting for master to send event

Master_Host: server2

Master_User: rpl

Master_Port: 3306

Connect_Retry: 60

Master_Log_File: mysql-bin.000004

Read_Master_Log_Pos: 2894

Relay_Log_File: mysql-relay-bin.000002

Relay_Log_Pos: 408

Relay_Master_Log_File: mysql-bin.000004

Slave_IO_Running: yes

Slave_SQL_Running: Yes

我们记下Read_Master_Log_Pos的值为2894，登录server2 mysql，执行如下sql:

mysql> show master statusG

*************************** 1. row ***************************

File: mysql-bin.000004

Position: 2894

Binlog_Do_DB:

Binlog_Ignore_DB:

Executed_Gtid_Set: 9347e042-9044-11e4-b4f0-000c29f14f4e:1-7,

f5bbfc15-904a-11e4-b519-000c29819d42:1-6

1 row in set (0.00 sec)

记下Position的值，并与Read_Master_Log_Pos比较，如果这两个值非常相近或相等，说明数据已经同步得差不多了，可以进行切换操作;如果差得很远，需要等待它们同步完成。

屏蔽mysql写操作

我们需要在切换时先禁止sql的写操作，如果不这样做，就会在切换时造成数据不一致的问题。屏蔽写操作我们在Atlas上操作。在server2执行登录Atlas命令：

mysql -h127.0.0.1 -P2345 -uuser -ppwd

mysql> SELECT * FROM backends;

+-------------+--------------------+-------+------+

| backend_ndx | address | state | type |

+-------------+--------------------+-------+------+

| 1 | 192.168.3.150:3306 | up | rw |

| 2 | 192.168.3.101:3306 | up | ro |

+-------------+--------------------+-------+------+

2 rows in set (0.00 sec)

执行SELECT * FROM backends;后我们看到backend id为1,所以我们执行SET OFFLINE 1;设置此后端下线。

mysql> SET OFFLINE 1;

+-------------+--------------------+---------+------+

| backend_ndx | address | state | type |

+-------------+--------------------+---------+------+

| 1 | 192.168.3.150:3306 | offline | rw |

+-------------+--------------------+---------+------+

1 row in set (0.00 sec)

mysql> SELECT * FROM backends;

+-------------+--------------------+---------+------+

| backend_ndx | address | state | type |

+-------------+--------------------+---------+------+

| 1 | 192.168.3.150:3306 | offline | rw |

| 2 | 192.168.3.101:3306 | up | ro |

+-------------+--------------------+---------+------+

2 rows in set (0.00 sec)

这时客户端就无法写入数据了。

恢复server1 mysql为主

mysqlrpladmin --master=root:Xp29at5F37@server2:3306 --new-master=root:Xp29at5F37@server1:3306 --demote-master --discover-slaves-login=root:Xp29at5F37 switchover

# Discovering slaves for master at server2:3306

# Discovering slave at server1:3306

# Found slave: server1:3306

# Checking privileges.

# Performing switchover from master at server2:3306 to slave at server1:3306.

# Checking candidate slave prerequisites.

# Checking slaves configuration to master.

# Waiting for slaves to catch up to old master.

# Stopping slaves.

# Performing STOP on all slaves.

# Demoting old master to be a slave to the new master.

# Switching slaves to new master.

# Starting all slaves.

# Performing START on all slaves.

# Checking slaves for errors.

# Switchover complete.

再次检查是否恢复成功.

mysqlrplcheck --master=root:Xp29at5F37@server1 --slave=root:Xp29at5F37@server2

# master on server1: ... connected.

# slave on server2: ... connected.

Test Description Status

---------------------------------------------------------------------------

Checking for binary logging on master [pass]

Are there binlog exceptions? [pass]

Replication user exists? [pass]

Checking server_id values [pass]

Checking server_uuid values [pass]

Is slave connected to master? [pass]

Check master information file [pass]

Checking InnoDB compatibility [pass]

Checking storage engines compatibility [pass]

Checking lower_case_table_names settings [pass]

Checking slave delay (seconds behind master) [pass]

# ...done.

设置VIP回到server1，在server2机器上执行：

/etc/init.d/keepalived restart

然后在两台机器分别执行ip addr查看ip绑定状态。

设置server2 atlas后端上线

server2上执行mysql -h127.0.0.1 -P2345 -uuser -ppwd登录，然后执行SET ONLINE 1;设置上线（这里1是后端的id，可以使用SELECT * FROM backends;查看）

mysql> SELECT * FROM backends;

+-------------+--------------------+-------+------+

| backend_ndx | address | state | type |

+-------------+--------------------+-------+------+

| 1 | 192.168.3.150:3306 | up | rw |

| 2 | 192.168.3.101:3306 | up | ro |

+-------------+--------------------+-------+------+

2 rows in set (0.00 sec)

到这里server1就恢复为主了。