mysql负载均衡完美解决方案

http://blog.csdn.net/v1v1wang/article/details/8771974

1.环境： 
mysql 5
ubuntu10.04 x86_64

mdb1        eth0    192.168.5.11 
mdb2        eth0    192.168.5.12
sdb1        eth0    192.168.5.21 
sdb2        eth0    192.168.5.22 
sdb3        eth0    192.168.5.23 
sdb4        eth0    192.168.5.24
haproxy 
        eth0    192.168.5.10    (mdb  vip write) 
        eth1    192.168.5.20    (sdb  vip read)
说明：mdb vip用于DB的写，sdb vip用于DB读，实现读写分离和负载均衡，带故障检测自动切换

2.架构图 
        web1    web2    web3 
         |              |          | 
        —————————- 
                    | 
            haproxy(lb db write/read) 
                    | 
        ———————————- 
        |                                | 
        mdb1                     mdb2 
          |                              | 
        ————–             —————- 
        |             |              |            | 
        sdb1    sdb2        sdb3      sdb4
说明： 
1）mdb1和mdb1配置成主-主模式，相互同步,通过haproxy提供一个lb的写ip 
2）sdb1和sdb2配置为mdb1的从，sdb3和sdb4配置为mdb2的从 
3）sdb1,sdb2,sdb3,sdb4这4台从库，通过haproxy提供一个lb的读ip 
4) 当mdb2停止复制，mdb1为主库，haproxy停止发送请求到mdb2和sdb3,sdb4 
5) 当mdb1停止复制，mdb2为主库，haproxy停止发送请求到mdb1和sdb1,sdb2 
6) 当mdb1和mdb2同时停止复制，这时2台主库变成readonly模式，数据库不能写入 
7）当mdb2 offline时，mdb1进入backup mode,停止发送请求到mdb2,sdb3,sdb4 
8）当mdb1 offline时，mdb2进入backup mode,停止发送请求到mdb1,sdb1,sdb2 
9) 当mdb1 mdb2同时offline,整个DB停止工作


3.安装mysql-server 
登录mdb1,mdb2,sdb1,sdb2,sdb3,sdb4，输入以下命令进行安装： 
apt-get install mysql-server -y
安装时会提示输入mysql root用户密码，输入gaojinbo.com
修改mysql配置，监听所有接口 
vi /etc/mysql/my.cnf 
修改为： 
bind-address            = 0.0.0.0
重启mysql 
/etc/init.d/mysql restart


4.配置mdb1,mdb2主-主同步 
1)mdb1: 
vi /etc/mysql/my.cnf 
server-id               = 1 
log_bin                 = mysql-bin
log-slave-updates                #很重要，从前一台机器上同步过来的数据才能同步到下一台机器 
expire_logs_days        = 10 
max_binlog_size         = 100M 
auto_increment_offset    = 1 
auto_increment_increment = 2

2)mdb2: 
vi /etc/mysql/my.cnf 
server-id               = 2 
log_bin                 = mysql-bin
log-slave-updates                #很重要，从前一台机器上同步过来的数据才能同步到下一台机器 
expire_logs_days        = 10 
max_binlog_size         = 100M 
auto_increment_offset    = 2 
auto_increment_increment = 2

3)mdb1和mdb2: 
重启mysql 
/etc/init.d/mysql restart
添加复制用户 
mysql -uroot -pgaojinbo.com 
GRANT REPLICATION SLAVE ON *.* TO ‘repl’@’192.168.5.%’ IDENTIFIED BY ‘gaojinbo’;
记录日志文件和pos 
mysql -uroot -pgaojinbo.com 
show master statusG


4)mdb1: 
change master to master_host=’192.168.5.12′,master_port=3306,master_user=’repl’,master_password=’gaojinbo’,master_log_file=’mysql-bin.000003′,master_log_pos=106; 
start slave; 
show slave statusG 
说明：mysql-bin.000003和106是主库配置第3)步记录的信息
出现以下内容，说明同步ok 
           Slave_IO_Running: Yes 
          Slave_SQL_Running: Yes

5)mdb2: 
change master to master_host=’192.168.5.11′,master_port=3306,master_user=’repl’,master_password=’gaojinbo’,master_log_file=’mysql-bin.000001′,master_log_pos=249; 
start slave; 
show slave statusG 
说明：mysql-bin.000001和249是主库配置第3)步记录的信息
出现以下内容，说明同步ok 
           Slave_IO_Running: Yes 
          Slave_SQL_Running: Yes


6）测试主-主同步 
mdb1: 
mysql -uroot -pgaojinbo.com 
show databases; 
create database gaojinbo;
mdb2: 
mysql -uroot -pgaojinbo.com 
show databases; 
即可看到在mdb1上建立的数据库gaojinbo

至此mdb1,mdb2主-主配置完成！

5.4台从库配置 
sdb1-4配置(注：server-id不能相同)： 
vi /etc/MySQL/my.cnf 
server-id               = 3 
log_bin                 = mysql-bin
重启mysql 
/etc/init.d/mysql restart

sdb1和sdb2配置成mdb1的从库: 
mysql -uroot -pgaojinbo.com 
change master to master_host=’192.168.5.11′,master_port=3306,master_user=’repl’,master_password=’gaojinbo’,master_log_file=’mysql-bin.000001′,master_log_pos=345; 
start slave; 
show slave statusG 
说明：mysql-bin.000001和345是主库配置第3)步记录的信息
出现以下内容，说明同步ok 
          Slave_IO_Running: Yes 
          Slave_SQL_Running: Yes

sdb3和sdb4配置成mdb2的从库: 
change master to master_host=’192.168.5.12′,master_port=3306,master_user=’repl’,master_password=’gaojinbo’,master_log_file=’mysql-bin.000003′,master_log_pos=106; 
start slave; 
show slave statusG 
说明：mysql-bin.000003和106是主库配置第3)步记录的信息
出现以下内容，说明同步ok 
          Slave_IO_Running: Yes 
          Slave_SQL_Running: Yes

测试： 
mdb1: 
mysql -uroot -pgaojinbo.com 
show databases; 
create database eossc;
在其他DB上，这时会看到刚建立的数据库eossc

至此4台从数据库配置完成！

6.编写mysql检测脚本 
1）mdb1和mdb2: 
vi /etc/xinetd.d/mysqlchk 
# 
# /etc/xinetd.d/mysqlchk 
# 
service mysqlchk_write 
{ 
        flags           = REUSE 
        socket_type     = stream 
        port            = 9200 
        wait            = no 
        user            = nobody 
        server          = /opt/mysqlchk_status.sh 
        log_on_failure  += USERID 
        disable         = no 
        only_from       = 192.168.5.0/24 # recommended to put the IPs that need 
                                    # to connect exclusively (security purposes) 
} 
  
service mysqlchk_replication 
{ 
        flags           = REUSE 
        socket_type     = stream 
        port            = 9201 
        wait            = no 
        user            = nobody 
        server          = /opt/mysqlchk_replication.sh 
        log_on_failure  += USERID 
        disable         = no 
        only_from       = 192.168.5.0/24 # recommended to put the IPs that need 
                                    # to connect exclusively (security purposes) 
}
添加服务端口 
vi /etc/services 
mysqlchk_write      9200/tcp                #mysqlchk_write 
mysqlchk_replication    9201/tcp                #mysqlchk_replication
mdb1上操作： 
vi /opt/mysqlchk_status.sh 
#!/bin/bash
MYSQL_HOST="192.168.5.11" 
MYSQL_PORT="3306" 
MYSQL_USERNAME="root" 
MYSQL_PASSWORD="gaojinbo.com" 
  
  
ERROR_MSG=`/usr/bin/mysql –host=$MYSQL_HOST –port=$MYSQL_PORT –user=$MYSQL_USERNAME –password=$MYSQL_PASSWORD -e "show databases;" 2>/dev/null` 
  
if [ "$ERROR_MSG" != "" ] 
then 
        # mysql is fine, return http 200 
        /bin/echo -e "HTTP/1.1 200 OK " 
        /bin/echo -e "Content-Type: Content-Type: text/plain " 
        /bin/echo -e " " 
        /bin/echo -e "MySQL is running. " 
        /bin/echo -e " " 
else 
        # mysql is down, return http 503 
        /bin/echo -e "HTTP/1.1 503 Service Unavailable " 
        /bin/echo -e "Content-Type: Content-Type: text/plain " 
        /bin/echo -e " " 
        /bin/echo -e "MySQL is *down*. " 
        /bin/echo -e " " 
fi
vi /opt/mysqlchk_replication.sh 
#!/bin/bash
MYSQL_HOST="192.168.5.11" 
MYSQL_PORT="3306" 
MYSQL_USERNAME="root" 
MYSQL_PASSWORD="gaojinbo.com" 
  
  
/usr/bin/mysql –host=$MYSQL_HOST –port=$MYSQL_PORT –user=$MYSQL_USERNAME –password=$MYSQL_PASSWORD -e "show slave statusG;" >/tmp/check_repl.txt
iostat=`grep "Slave_IO_Running" /tmp/check_repl.txt |awk ‘{print $2}’ ` 
sqlstat=`grep "Slave_SQL_Running" /tmp/check_repl.txt |awk ‘{print $2}’ `
#echo iostat:$iostat and sqlstat:$sqlstat
if [ "$iostat" = "No" ] || [ "$sqlstat" = "No" ]; 
then 
        # mysql is down, return http 503 
        /bin/echo -e "HTTP/1.1 503 Service Unavailable " 
        /bin/echo -e "Content-Type: Content-Type: text/plain " 
        /bin/echo -e " " 
        /bin/echo -e "MySQL replication  is *down*. " 
        /bin/echo -e " " 
else 
        # mysql is fine, return http 200 
        /bin/echo -e "HTTP/1.1 200 OK " 
        /bin/echo -e "Content-Type: Content-Type: text/plain " 
        /bin/echo -e " " 
        /bin/echo -e "MySQL replication is running. " 
        /bin/echo -e " " 
fi
测试同步检测脚本： 
mysql -uroot -pgaojinbo.com 
stop slave sql_thread; #或者  stop slave io_thread;
/opt/mysqlchk_replication.sh
mdb2上操作： 
添加和mdb1一样的脚本，把 
/opt/mysqlchk_status.sh        里面的192.168.5.11修改为192.168.5.12 
/opt/mysqlchk_replication.sh    里面的192.168.5.11修改为192.168.5.12

2）sdb1,sdb2,sdb3,sdb4上操作： 
vi /etc/xinetd.d/mysqlchk 
# 
# /etc/xinetd.d/mysqlchk 
# 
  
service mysqlchk_replication 
{ 
        flags           = REUSE 
        socket_type     = stream 
        port            = 9201 
        wait            = no 
        user            = nobody 
        server          = /opt/mysqlchk_replication.sh 
        log_on_failure  += USERID 
        disable         = no 
        only_from       = 192.168.5.0/24 # recommended to put the IPs that need 
                                    # to connect exclusively (security purposes) 
}
vi /opt/mysqlchk_replication.sh 
#!/bin/bash
MYSQL_HOST="192.168.5.21" 
MYSQL_PORT="3306" 
MYSQL_USERNAME="root" 
MYSQL_PASSWORD="gaojinbo.com" 
  
  
/usr/bin/mysql –host=$MYSQL_HOST –port=$MYSQL_PORT –user=$MYSQL_USERNAME –password=$MYSQL_PASSWORD -e "show slave statusG;" >/tmp/check_repl.txt
iostat=`grep "Slave_IO_Running" /tmp/check_repl.txt |awk ‘{print $2}’ ` 
sqlstat=`grep "Slave_SQL_Running" /tmp/check_repl.txt |awk ‘{print $2}’ `
#echo iostat:$iostat and sqlstat:$sqlstat
if [ "$iostat" = "No" ] || [ "$sqlstat" = "No" ]; 
then 
        # mysql is down, return http 503 
        /bin/echo -e "HTTP/1.1 503 Service Unavailable " 
        /bin/echo -e "Content-Type: Content-Type: text/plain " 
        /bin/echo -e " " 
        /bin/echo -e "MySQL replication  is *down*. " 
        /bin/echo -e " " 
else 
        # mysql is fine, return http 200 
        /bin/echo -e "HTTP/1.1 200 OK " 
        /bin/echo -e "Content-Type: Content-Type: text/plain " 
        /bin/echo -e " " 
        /bin/echo -e "MySQL replication is running. " 
        /bin/echo -e " " 
fi
注：脚本/opt/mysqlchk_replication.sh里面的ip 
sdb1    MYSQL_HOST="192.168.5.21" 
sdb2    MYSQL_HOST="192.168.5.22" 
sdb3    MYSQL_HOST="192.168.5.23" 
sdb4    MYSQL_HOST="192.168.5.24"
添加服务端口 
vi /etc/services 
mysqlchk_replication    9201/tcp                #mysqlchk_replication

3)所有DB上操作： 
增加检测脚本执行权限 
chmod +x /opt/mysql*.sh
重启系统 
reboot
查看监听端口 
netstat -antup|grep xinetd 
tcp        0      0 0.0.0.0:9200            0.0.0.0:*               LISTEN      903/xinetd      
tcp        0      0 0.0.0.0:9201            0.0.0.0:*               LISTEN      903/xinetd
注：sdb只有9201监听

7.haproxy安装配置 
下载编译安装： 
wget http://haproxy.1wt.eu/download/1.4/src/haproxy-1.4.11.tar.gz 
tar xvzf haproxy-1.4.11.tar.gz 
cd haproxy-1.4.11 
make TARGET=linux26 ARCH=x86_64 
make install

配置 
vi /etc/haproxy.cfg
global 
        maxconn 40000 
        debug 
        #quiet 
        user haproxy 
        group haproxy 
        nbproc 1 
        log 127.0.0.1 local3 
        spread-checks 2 
defaults 
         timeout server  3s 
         timeout connect 3s 
         timeout client  60s 
         timeout http-request 3s 
         timeout queue   3s
frontend db_write 
        bind 192.168.5.10:3306 
        default_backend cluster_db_write
frontend db_read 
        bind 192.168.5.20:3306 
        default_backend cluster_db_read
frontend web_haproxy_status 
        bind :80 
        default_backend web_status
frontend monitor_mdb1 
        bind 127.0.0.1:9301 
        mode http 
        acl no_repl_mdb1 nbsrv(mdb1_replication) eq 0 
        acl no_repl_mdb2 nbsrv(mdb2_replication) eq 0 
        acl no_mdb1 nbsrv(mdb1_status) eq 0 
        acl no_mdb2 nbsrv(mdb2_status) eq 0 
        monitor-uri /dbs 
        monitor fail unless no_repl_mdb1 no_repl_mdb2 no_mdb2 
        monitor fail if no_mdb1 no_mdb2
frontend monitor_mdb2 
        bind 127.0.0.1:9302 
        mode http 
        acl no_repl_mdb1 nbsrv(mdb1_replication) eq 0 
        acl no_repl_mdb2 nbsrv(mdb2_replication) eq 0 
        acl no_mdb1 nbsrv(mdb1_status) eq 0 
        acl no_mdb2 nbsrv(mdb2_status) eq 0 
        monitor-uri /dbs 
        monitor fail unless no_repl_mdb1 no_repl_mdb2 no_mdb1 
        monitor fail if no_mdb1 no_mdb2
frontend monitor_sdb1 
        bind 127.0.0.1:9303 
        mode http 
        acl no_repl_sdb1 nbsrv(sdb1_replication) eq 0 
        acl no_repl_mdb1 nbsrv(mdb1_replication) eq 0 
        acl no_mdb2 nbsrv(mdb2_status) eq 1 
        monitor-uri /dbs 
        monitor fail if no_repl_sdb1 
        monitor fail if no_repl_mdb1 no_mdb2
frontend monitor_sdb2 
        bind 127.0.0.1:9304 
        mode http 
        acl no_repl_sdb2 nbsrv(sdb2_replication) eq 0 
        acl no_repl_mdb1 nbsrv(mdb1_replication) eq 0 
        acl no_mdb2 nbsrv(mdb2_status) eq 1 
        monitor-uri /dbs 
        monitor fail if no_repl_sdb2 
        monitor fail if no_repl_mdb1 no_mdb2
frontend monitor_sdb3 
        bind 127.0.0.1:9305 
        mode http 
        acl no_repl_sdb3 nbsrv(sdb3_replication) eq 0 
        acl no_repl_mdb2 nbsrv(mdb2_replication) eq 0 
        acl no_mdb1 nbsrv(mdb1_status) eq 1 
        monitor-uri /dbs 
        monitor fail if no_repl_sdb3 
        monitor fail if no_repl_mdb2 no_mdb1
frontend monitor_sdb4 
        bind 127.0.0.1:9306 
        mode http 
        acl no_repl_sdb4 nbsrv(sdb4_replication) eq 0 
        acl no_repl_mdb2 nbsrv(mdb2_replication) eq 0 
        acl no_mdb1 nbsrv(mdb1_status) eq 1 
        monitor-uri /dbs 
        monitor fail if no_repl_sdb4 
        monitor fail if no_repl_mdb2 no_mdb1
frontend monitor_splitbrain 
        bind 127.0.0.1:9300 
        mode http 
        acl no_repl01 nbsrv(mdb1_replication) eq 0 
        acl no_repl02 nbsrv(mdb2_replication) eq 0 
        acl mdb1 nbsrv(mdb1_status) eq 1 
        acl mdb2 nbsrv(mdb2_status) eq 1 
        monitor-uri /dbs 
        monitor fail unless no_repl01 no_repl02 mdb1 mdb2
backend mdb1_replication 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server mdb1 192.168.5.11:3306 check port 9201 inter 1s rise 1 fall 1 
  
backend mdb2_replication 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server mdb2 192.168.5.12:3306 check port 9201 inter 1s rise 1 fall 1
backend sdb1_replication 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server sdb1 192.168.5.21:3306 check port 9201 inter 1s rise 1 fall 1
backend sdb2_replication 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server sdb2 192.168.5.22:3306 check port 9201 inter 1s rise 1 fall 1
backend sdb3_replication 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server sdb3 192.168.5.23:3306 check port 9201 inter 1s rise 1 fall 1
backend sdb4_replication 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server sdb4 192.168.5.24:3306 check port 9201 inter 1s rise 1 fall 1
backend mdb1_status 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server mdb1 192.168.5.11:3306 check port 9200 inter 1s rise 2 fall 2 
  
backend mdb2_status 
        mode tcp 
        balance roundrobin 
        option tcpka 
        option httpchk 
        server mdb2 192.168.5.12:3306 check port 9200 inter 1s rise 2 fall 2
backend cluster_db_write 
        mode    tcp 
        option  tcpka 
        balance roundrobin 
        option  httpchk GET /dbs 
        server  mdb1 192.168.5.11:3306 weight 1 check port 9201 inter 1s rise 5 fall 1 
        server  mdb2 192.168.5.12:3306 weight 1 check port 9201 inter 1s rise 5 fall 1 backup 
        server  mdb1_backup 192.168.5.11:3306 weight 1 check port 9301 inter 1s rise 2 fall 2 addr 127.0.0.1 backup 
        server  mdb2_backup 192.168.5.12:3306 weight 1 check port 9302 inter 1s rise 2 fall 2 addr 127.0.0.1 backup
backend cluster_db_read 
        mode    tcp 
        option  tcpka 
        balance roundrobin 
        option  httpchk GET /dbs 
        server  mdb1 192.168.5.11:3306 weight 1 track cluster_db_write/mdb1 
        server  mdb2 192.168.5.12:3306 weight 1 track cluster_db_write/mdb2 
        server  mdb1_backup 192.168.5.11:3306 weight 1 track cluster_db_write/mdb1_backup 
        server  mdb2_backup 192.168.5.12:3306 weight 1 track cluster_db_write/mdb2_backup 
        server  mdb1_splitbrain 192.168.5.11:3306 weight 1 check port 9300 inter 1s rise 1 fall 2 addr 127.0.0.1 
        server  mdb2_splitbrain 192.168.5.12:3306 weight 1 check port 9300 inter 1s rise 1 fall 2 addr 127.0.0.1
        server  sdb1_slave 192.168.5.21:3306 weight 1 check port 9303 inter 1s rise 5 fall 1 addr 127.0.0.1 
        server  sdb2_slave 192.168.5.22:3306 weight 1 check port 9304 inter 1s rise 5 fall 1 addr 127.0.0.1 
        server  sdb3_slave 192.168.5.23:3306 weight 1 check port 9305 inter 1s rise 5 fall 1 addr 127.0.0.1 
        server  sdb4_slave 192.168.5.24:3306 weight 1 check port 9306 inter 1s rise 5 fall 1 addr 127.0.0.1

backend  web_status 
         mode http 
         stats enable 
       # stats scope 
       # stats hide-version 
         stats refresh 5s 
         stats uri /status 
         stats realm Haproxy statistics 
         stats auth ylmf:gaojinbo

8.测试
1）正常情况，backup和splitbrain状态down


2）停止mdb2复制，mdb2和sdb3,sdb4状态down，数据库仍可读写


3)同时停止mdb1,mdb2复制，mdb1和sdb1,sdb2,sdb3,sdb4状态down，数据库只能读


4）关闭mdb1数据库，mdb1,mdb2和sdb1,sdb2状态down，数据库仍可读写


5）关闭mdb2数据库，mdb1,mdb2和sdb3,sdb4状态down，数据库仍可读写