探究MySQL MGR的读写分离

1：现有环境

机器	MySQL	环境
172.16.128.240	MGR NODE1	MGR
172.16.128.241	MGR NODE2	MGR
172.16.128.242	MGR NODE3	MGR
172.16.128.239	VIP	keepalived

环境介绍，VIP 172.16.128.239在172.16.128.240上，我们是基于172.16.128.240和172.16.128.241做的keepalived。MGR为多主模式。我们考虑使用中间件能够实现MGR的读写分离。中间件选型为Atlas,Atlas是360团队开源的一套基于MySQL-Proxy基础之上的代理，修改了MySQL-Proxy的一些BUG，并且进行了优化。

2：安装Atlas

[root@YC-ss1 tmp]# rpm -ivh Atlas-2.2.1.el6.x86_64.rpm
Preparing...                          ################################# [100%]
Updating / installing...
   1:Atlas-2.2.1-1                    ################################# [100%]
[root@YC-ss1 mysql-proxy]# pwd
/usr/local/mysql-proxy
[root@YC-ss1 mysql-proxy]# ls
bin  conf  lib  log

我们看一下，安装完Atlas 以后会生成四个文件夹。首先我们要配置密码的加密：

[root@YC-ss1 bin]# ./encrypt 123456
/iZxz+0GRoA=

我的root@%账号密码就是123456，比较简单。我们先记录这个加密字符串，等会要在配置文件中配置。配置配置文件

[root@YC-ss1 conf]# cp test.cnf atlas.cnf
[root@YC-ss1 conf]# cat atlas.cnf
[mysql-proxy]

#带#号的为非必需的配置项目

#管理接口的用户名
admin-username = user

#管理接口的密码
admin-password = pwd

#Atlas后端连接的MySQL主库的IP和端口，可设置多项，用逗号分隔
proxy-backend-addresses = 172.16.128.239:3306

#Atlas后端连接的MySQL从库的IP和端口，@后面的数字代表权重，用来作负载均衡，若省略则默认为1，可设置多项，用逗号分隔
proxy-read-only-backend-addresses = 172.16.128.240:3306,172.16.128.241:3306,172.16.128.242:3306

#用户名与其对应的加密过的MySQL密码，密码使用PREFIX/bin目录下的加密程序encrypt加密，下行的user1和user2为示例，将其替换为你的MySQL的用户名和加密密码！
pwds = root:/iZxz+0GRoA=

#设置Atlas的运行方式，设为true时为守护进程方式，设为false时为前台方式，一般开发调试时设为false，线上运行时设为true,true后面不能有空格。
daemon = true

#设置Atlas的运行方式，设为true时Atlas会启动两个进程，一个为monitor，一个为worker，monitor在worker意外退出后会自动将其重启，设为false时只有worker，没有monitor，一般开发调试时设为false，线上运行时设为true,true后面不能有空格。
keepalive = true

#工作线程数，对Atlas的性能有很大影响，可根据情况适当设置
event-threads = 16

#日志级别，分为message、warning、critical、error、debug五个级别
log-level = message

#日志存放的路径
log-path = /usr/local/mysql-proxy/log

#SQL日志的开关，可设置为OFF、ON、REALTIME，OFF代表不记录SQL日志，ON代表记录SQL日志，REALTIME代表记录SQL日志且实时写入磁盘，默认为OFF
#sql-log = OFF

#慢日志输出设置。当设置了该参数时，则日志只输出执行时间超过sql-log-slow（单位：ms)的日志记录。不设置该参数则输出全部日志。
#sql-log-slow = 10

#实例名称，用于同一台机器上多个Atlas实例间的区分
#instance = test

#Atlas监听的工作接口IP和端口
proxy-address = 0.0.0.0:6002

#Atlas监听的管理接口IP和端口
admin-address = 0.0.0.0:6003

#分表设置，此例中person为库名，mt为表名，id为分表字段，3为子表数量，可设置多项，以逗号分隔，若不分表则不需要设置该项
#tables = person.mt.id.3

#默认字符集，设置该项后客户端不再需要执行SET NAMES语句
#charset = utf8

#允许连接Atlas的客户端的IP，可以是精确IP，也可以是IP段，以逗号分隔，若不设置该项则允许所有IP连接，否则只允许列表中的IP连接
#client-ips = 127.0.0.1, 192.168.1

#Atlas前面挂接的LVS的物理网卡的IP(注意不是虚IP)，若有LVS且设置了client-ips则此项必须设置，否则可以不设置
#lvs-ips = 192.168.1.1

启动Atlas

[root@YC-ss1 bin]# ./mysql-proxyd  atlas start
OK: MySQL-Proxy of atlas is started

OK，现在我们的Atlas已经启动了。我们登录查看一下信息：

[root@YC-ss1 conf]#  mysql -h127.0.0.1 -P6003 -uuser -ppwd
mysql: [Warning] Using a password on the command line interface can be insecure.
Welcome to the MySQL monitor.  Commands end with ; or \g.
Your MySQL connection id is 1
Server version: 5.0.99-agent-admin

Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.

Oracle is a registered trademark of Oracle Corporation and/or its
affiliates. Other names may be trademarks of their respective
owners.

Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.
(user@127.0.0.1) [(none)]> SELECT * FROM help
    -> ;
+----------------------------+---------------------------------------------------------+
| command                    | description                                             |
+----------------------------+---------------------------------------------------------+
| SELECT * FROM help         | shows this help                                         |
| SELECT * FROM backends     | lists the backends and their state                      |
| SET OFFLINE $backend_id    | offline backend server, $backend_id is backend_ndx's id |
| SET ONLINE $backend_id     | online backend server, ...                              |
| ADD MASTER $backend        | example: "add master 127.0.0.1:3306", ...               |
| ADD SLAVE $backend         | example: "add slave 127.0.0.1:3306", ...                |
| REMOVE BACKEND $backend_id | example: "remove backend 1", ...                        |
| SELECT * FROM clients      | lists the clients                                       |
| ADD CLIENT $client         | example: "add client 192.168.1.2", ...                  |
| REMOVE CLIENT $client      | example: "remove client 192.168.1.2", ...               |
| SELECT * FROM pwds         | lists the pwds                                          |
| ADD PWD $pwd               | example: "add pwd user:raw_password", ...               |
| ADD ENPWD $pwd             | example: "add enpwd user:encrypted_password", ...       |
| REMOVE PWD $pwd            | example: "remove pwd user", ...                         |
| SAVE CONFIG                | save the backends to config file                        |
| SELECT VERSION             | display the version of Atlas                            |
+----------------------------+---------------------------------------------------------+
16 rows in set (0.00 sec)

(user@127.0.0.1) [(none)]>  SELECT * FROM backends;
+-------------+---------------------+-------+------+
| backend_ndx | address             | state | type |
+-------------+---------------------+-------+------+
|           1 | 172.16.128.239:3306 | up    | rw   |
|           2 | 172.16.128.240:3306 | up    | ro   |
|           3 | 172.16.128.241:3306 | up    | ro   |
|           4 | 172.16.128.242:3306 | up    | ro   |
+-------------+---------------------+-------+------+
4 rows in set (0.00 sec)

然后使用服务端口访问一下：

[root@YC-ss1 ~]# mysql -h127.0.0.1 -P6002 -uroot -p123456
mysql: [Warning] Using a password on the command line interface can be insecure.
Welcome to the MySQL monitor.  Commands end with ; or \g.
Your MySQL connection id is 4
Server version: 5.0.81-log MySQL Community Server (GPL)

Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.

Oracle is a registered trademark of Oracle Corporation and/or its
affiliates. Other names may be trademarks of their respective
owners.

Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.

(root@127.0.0.1) [(none)]> show databases;
+--------------------+
| Database           |
+--------------------+
| information_schema |
| mxqteset           |
| mysql              |
| performance_schema |
| sys                |
+--------------------+
5 rows in set (0.00 sec)

(root@127.0.0.1) [(none)]> select @@server_id;
+-------------+
| @@server_id |
+-------------+
|  1282403306 |
+-------------+
1 row in set (0.00 sec)

(root@127.0.0.1) [(none)]> select @@server_id;
+-------------+
| @@server_id |
+-------------+
|  1282413306 |
+-------------+
1 row in set (0.00 sec)

(root@127.0.0.1) [(none)]> select @@server_id;
+-------------+
| @@server_id |
+-------------+
|  1282423306 |
+-------------+
1 row in set (0.00 sec)

可以看到我们查询了三次。已经路由到了MGR的三个节点上了。读已经是没有问题的。写的话都会通过VIP:172.16.128.239来写。而且我们的VIP是会自动故障转移，也就是说保证是可写的。这样基本上就实现的读写分离。如果我们不想让240这台机器发生读，就可以把优先级配置的低一些，优先路由倒其他的机器。

3：压力测试

我们使用tpcc-mysql进行测试。先进行安装，下载地址：http://imysql.com/wp-content/uploads/2014/09/tpcc-mysql-src.tgz

[root@YC-ss1 tpccmysql]# gunzip tpcc-mysql-src.tgz
[root@YC-ss1 tpccmysql]#  tar xf tpcc-mysql-src.tar
[root@YC-ss1 tpccmysql]# cd src/
-bash: cd: src/: No such file or directory
[root@YC-ss1 tpccmysql]# ls
tpcc-mysql  tpcc-mysql-src.tar
[root@YC-ss1 tpccmysql]# cd tpcc-mysql
[root@YC-ss1 tpcc-mysql]# ls
add_fkey_idx.sql  count.sql  create_table.sql  drop_cons.sql  load.sh  README  schema2  scripts  src
[root@YC-ss1 tpcc-mysql]# cd src/
[root@YC-ss1 src]# make
cc -w -O2 -g -I. `mysql_config --include`  -c load.c
cc -w -O2 -g -I. `mysql_config --include`  -c support.c
cc load.o support.o `mysql_config --libs_r` -lrt -o ../tpcc_load
cc -w -O2 -g -I. `mysql_config --include`  -c main.c
cc -w -O2 -g -I. `mysql_config --include`  -c spt_proc.c
cc -w -O2 -g -I. `mysql_config --include`  -c driver.c
cc -w -O2 -g -I. `mysql_config --include`  -c sequence.c
cc -w -O2 -g -I. `mysql_config --include`  -c rthist.c
cc -w -O2 -g -I. `mysql_config --include`  -c neword.c
cc -w -O2 -g -I. `mysql_config --include`  -c payment.c
cc -w -O2 -g -I. `mysql_config --include`  -c ordstat.c
cc -w -O2 -g -I. `mysql_config --include`  -c delivery.c
cc -w -O2 -g -I. `mysql_config --include`  -c slev.c
cc main.o spt_proc.o driver.o support.o sequence.o rthist.o neword.o payment.o ordstat.o delivery.o slev.o `mysql_config --libs_r` -lrt -o ../tpcc_start
[root@YC-ss1 src]# cd ..
[root@YC-ss1 tpcc-mysql]# ls
add_fkey_idx.sql  count.sql  create_table.sql  drop_cons.sql  load.sh  README  schema2  scripts  src  tpcc_load  tpcc_start

初始化压测数据：

(root@127.0.0.1) [tpccmysql]> create database tpccmysql;
Query OK, 1 row affected (0.02 sec)

(root@127.0.0.1) [tpccmysql]> use tpccmysql
Database changed
(root@127.0.0.1) [tpccmysql]> source /tmp/create_table.sql;
Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.01 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.01 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.01 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.01 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.01 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.01 sec)

Query OK, 0 rows affected, 1 warning (0.01 sec)

Query OK, 0 rows affected (0.00 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.18 sec)

Query OK, 0 rows affected, 1 warning (0.00 sec)

Query OK, 0 rows affected (0.00 sec)

但是没有外键的话是不能压测的，及时压测很多也是失败的。唉，坑爹，还好有一个TPCC去除外键的，这个应该是叶大大搞的，github地址：https://github.com/yejr/tpcc-mysql-autoinc-pk

我们使用去除外键的TPCC进行压测，先进行安装

[root@YC-ss1 tpccmysql]# cd tpcc-mysql-autoinc-pk-master
[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ls
add_fkey_idx.sql  create_table-autoinc-pk.sql  load.sh  README  README.md  run_tpcc.sh  schema2  scripts  src  tpcc_load_parallel.sh
[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# cd src/
[root@YC-ss1 src]# make
cc -w -O2 -g -I. `mysql_config --include`  -c load.c
cc -w -O2 -g -I. `mysql_config --include`  -c support.c
cc load.o support.o `mysql_config --libs_r` -lrt -o ../tpcc_load
cc -w -O2 -g -I. `mysql_config --include`  -c main.c
cc -w -O2 -g -I. `mysql_config --include`  -c spt_proc.c
cc -w -O2 -g -I. `mysql_config --include`  -c driver.c
cc -w -O2 -g -I. `mysql_config --include`  -c sequence.c
cc -w -O2 -g -I. `mysql_config --include`  -c rthist.c
cc -w -O2 -g -I. `mysql_config --include`  -c neword.c
cc -w -O2 -g -I. `mysql_config --include`  -c payment.c
cc -w -O2 -g -I. `mysql_config --include`  -c ordstat.c
cc -w -O2 -g -I. `mysql_config --include`  -c delivery.c
cc -w -O2 -g -I. `mysql_config --include`  -c slev.c
cc main.o spt_proc.o driver.o support.o sequence.o rthist.o neword.o payment.o ordstat.o delivery.o slev.o `mysql_config --libs_r` -lrt -o ../tpcc_start
[root@YC-ss1 src]# ls
delivery.c  driver.c  load.c       load.o  main.o    neword.c  ordstat.c  parse_port.h  payment.o  rthist.h  sequence.c  sequence.o  slev.o      spt_proc.h  support.c  tpc.h
delivery.o  driver.o  load.c.orig  main.c  Makefile  neword.o  ordstat.o  payment.c     rthist.c   rthist.o  sequence.h  slev.c      spt_proc.c  spt_proc.o  support.o  trans_if.h
[root@YC-ss1 src]# cd ..
[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ls
add_fkey_idx.sql  create_table-autoinc-pk.sql  load.sh  README  README.md  run_tpcc.sh  schema2  scripts  src  tpcc_load  tpcc_load_parallel.sh  tpcc_start

初始化数据

(root@127.0.0.1) [(none)]> create database tpcc;
Query OK, 1 row affected (0.07 sec)

(root@127.0.0.1) [(none)]> use tpcc
Database changed
(root@127.0.0.1) [tpcc]> source /home/tpccmysql/tpcc-mysql-autoinc-pk-master/create_table-autoinc-pk.sql;

使用tpcc_load加载数据：

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_load 127.0.0.1:6002  tpcc root "123456" 30
*************************************
*** ###easy### TPC-C Data Loader  ***
*************************************
<Parameters>
     [server]: 127.0.0.1
     [port]: 6002
     [DBname]: tpcc
       [user]: root
       [pass]: 123456
  [warehouse]: 30
TPCC Data Load Started...
Loading Item
.................................................. 50000
.................................................. 100000

这个过程有点漫长，可以去楼下抽支烟。然后进行压力测试，测试一个小时，这个时间也有点久，可以下楼买个冰棍了。

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_start -h 127.0.0.1 -P 6002 -d tpcc -u root -p 123456 -w 30 -c 8 -r 120 -l 3600  -f tpcc_0705
***************************************
*** ###easy### TPC-C Load Generator ***
***************************************
option h with value '127.0.0.1'
option P with value ''
option d with value 'tpcc'
option u with value 'root'
option p with value ''
option w with value ''
option c with value ''
option r with value ''
option l with value ''
option f with value 'tpcc_0705'
<Parameters>
     [server]: 127.0.0.1
     [port]: 6002
     [DBname]: tpcc
       [user]: root
       [pass]: 123456
  [warehouse]: 30
 [connection]: 8
     [rampup]: 120 (sec.)
    [measure]: 3600 (sec.)

等待擦测试结果

STOPPING THREADS........

<Raw Results>
  [] sc:24209  lt:1  rt:0  fl:0
  [] sc:24167  lt:0  rt:0  fl:0
  [] sc:2421  lt:1  rt:0  fl:0
  [] sc:2420  lt:0  rt:0  fl:0
  [] sc:2423  lt:0  rt:0  fl:0
 in 3600 sec.

<Raw Results2(sum ver.)>
  [] sc:24209  lt:1  rt:0  fl:0
  [] sc:24210  lt:0  rt:0  fl:0
  [] sc:2421  lt:1  rt:0  fl:0
  [] sc:2421  lt:0  rt:0  fl:0
  [] sc:2423  lt:0  rt:0  fl:0 

<Constraint Check> (all must be [OK])
 [transaction percentage]
        Payment: 43.43% (>=43.0%) [OK]
   Order-Status: 4.35% (>= 4.0%) [OK]
       Delivery: 4.35% (>= 4.0%) [OK]
    Stock-Level: 4.35% (>= 4.0%) [OK]
 [response time (at least 90% passed)]
      New-Order: 100.00%  [OK]
        Payment: 100.00%  [OK]
   Order-Status: 99.96%  [OK]
       Delivery: 100.00%  [OK]
    Stock-Level: 100.00%  [OK]

<TpmC>
                 403.500 TpmC

我们可以看到在我这个级别压测下，Atlas+MGR还是很稳定的。下面我们看下故障转移

4：故障转移的测试

往上面回顾一下我们的机器环境

机器	MySQL	环境
172.16.128.240	MGR NODE1	MGR
172.16.128.241	MGR NODE2	MGR
172.16.128.242	MGR NODE3	MGR
172.16.128.239	VIP	keepalived

现在我们Atlas提供的服务端口是172.16.128.242:6002，我们模拟NODE1节点故障 ，然后看一下压测的结果最终结果是如何的。

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_start -h 127.0.0.1 -P 6002 -d tpcc -u root -p 123456 -w 30 -c 8 -r 120 -l 300  -f tpcc_0706
***************************************
*** ###easy### TPC-C Load Generator ***
***************************************
option h with value '127.0.0.1'
option P with value ''
option d with value 'tpcc'
option u with value 'root'
option p with value ''
option w with value ''
option c with value ''
option r with value ''
option l with value ''
option f with value 'tpcc_0706'
<Parameters>
     [server]: 127.0.0.1
     [port]: 6002
     [DBname]: tpcc
       [user]: root
       [pass]: 123456
  [warehouse]: 30
 [connection]: 8
     [rampup]: 120 (sec.)
    [measure]: 300 (sec.)

RAMP-UP TIME.(120 sec.)

第二部关闭NODE1，

2006, HY000, MySQL server has gone away
payment 4:1
2006, HY000, MySQL server has gone away
payment 4:1
2006, HY000, MySQL server has gone away
payment 4:1
2006, HY000, MySQL server has gone away

全部都TM的gone away了，也真是无奈的，我们在重新开始压测，发现是没问题的，

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_start -h 127.0.0.1 -P 6002 -d tpcc -u root -p 123456 -w 30 -c 8 -r 120 -l 300  -f tpcc_0706
***************************************
*** ###easy### TPC-C Load Generator ***
***************************************
option h with value '127.0.0.1'
option P with value ''
option d with value 'tpcc'
option u with value 'root'
option p with value ''
option w with value ''
option c with value ''
option r with value ''
option l with value ''
option f with value 'tpcc_0706'
<Parameters>
     [server]: 127.0.0.1
     [port]: 6002
     [DBname]: tpcc
       [user]: root
       [pass]: 123456
  [warehouse]: 30
 [connection]: 8
     [rampup]: 120 (sec.)
    [measure]: 300 (sec.)

RAMP-UP TIME.(120 sec.)

MEASURING START.

  10, 905(0):1.694|4.648, 905(0):0.449|4.042, 91(0):0.217|0.562, 90(0):2.178|3.151, 90(0):5.098|10.983
  20, 724(0):1.750|2.948, 725(0):0.548|1.545, 73(0):0.173|0.230, 72(0):1.947|2.309, 73(0):4.700|5.217
  30, 1050(0):1.746|4.647, 1046(0):0.372|0.746, 104(0):0.163|0.258, 105(0):1.843|1.966, 105(0):5.323|5.616
  40, 917(0):1.754|4.656, 918(0):0.392|0.985, 92(0):0.190|0.191, 91(0):2.042|2.340, 92(0):5.574|6.931
  50, 951(0):1.688|2.792, 950(1):0.440|5.618, 96(0):0.206|0.373, 96(0):2.240|3.440, 95(0):4.809|4.985
  60, 965(0):1.764|2.737, 966(0):0.393|0.836, 96(0):0.222|0.517, 97(0):2.220|2.448, 95(0):4.989|4.990
  70, 940(0):1.589|2.525, 941(0):0.463|1.766, 94(0):0.164|0.263, 94(0):2.446|3.339, 96(0):5.512|5.704
  80, 746(2):1.671|7.135, 742(1):0.445|5.139, 75(0):0.229|0.324, 74(0):2.353|2.653, 74(0):5.342|5.454
  90, 1022(0):1.784|4.030, 1022(0):0.421|1.311, 102(0):0.311|0.479, 102(0):2.093|3.696, 102(0):5.810|6.025
 100, 1012(0):1.809|2.774, 1012(0):0.436|0.961, 101(0):0.207|0.381, 102(0):2.135|2.262, 101(0):5.867|5.931
 110, 928(0):1.651|2.960, 930(0):0.397|1.347, 93(0):0.193|0.371, 92(0):2.157|3.873, 93(0):5.445|6.339
 120, 1000(0):1.670|2.486, 1001(0):0.408|0.888, 99(0):0.257|0.405, 100(0):1.943|2.189, 100(0):5.292|5.770
 130, 992(0):1.749|4.123, 990(0):0.463|1.120, 100(0):0.206|0.277, 99(0):1.742|1.759, 100(0):4.674|5.843
 140, 868(1):1.893|7.145, 864(0):0.447|1.527, 86(0):0.188|0.221, 87(0):2.262|2.292, 86(0):4.586|4.684
 150, 1024(0):1.641|2.351, 1027(0):0.415|0.680, 103(0):0.153|0.195, 102(0):1.936|5.337, 102(0):4.877|5.004

<Raw Results>
  [] sc:28932  lt:6  rt:0  fl:0
  [] sc:28921  lt:2  rt:0  fl:0
  [] sc:2894  lt:0  rt:0  fl:0
  [] sc:2893  lt:0  rt:0  fl:0
  [] sc:2894  lt:0  rt:0  fl:0
 in 300 sec.

<Raw Results2(sum ver.)>
  [] sc:28932  lt:6  rt:0  fl:0
  [] sc:28934  lt:2  rt:0  fl:0
  [] sc:2894  lt:0  rt:0  fl:0
  [] sc:2893  lt:0  rt:0  fl:0
  [] sc:2894  lt:0  rt:0  fl:0 

<Constraint Check> (all must be [OK])
 [transaction percentage]
        Payment: 43.47% (>=43.0%) [OK]
   Order-Status: 4.35% (>= 4.0%) [OK]
       Delivery: 4.35% (>= 4.0%) [OK]
    Stock-Level: 4.35% (>= 4.0%) [OK]
 [response time (at least 90% passed)]
      New-Order: 99.98%  [OK]
        Payment: 99.99%  [OK]
   Order-Status: 100.00%  [OK]
       Delivery: 100.00%  [OK]
    Stock-Level: 100.00%  [OK]

<TpmC>
                 5787.600 TpmC

我们再看一下Atlas的管理信息

(user@127.0.0.1) [(none)]> SELECT * FROM backends;
+-------------+---------------------+-------+------+
| backend_ndx | address             | state | type |
+-------------+---------------------+-------+------+
|           1 | 172.16.128.239:3306 | up    | rw   |
|           2 | 172.16.128.240:3306 | up    | ro   |
|           3 | 172.16.128.241:3306 | up    | ro   |
|           4 | 172.16.128.242:3306 | up    | ro   |
+-------------+---------------------+-------+------+
4 rows in set (0.00 sec)

发现有问题的NODE1节点的状态居然还是UP，也就是说我们还要自己想办法通知Atlas我这个节点挂了，要修改配置信息。不过最起码说明读写分离是可以做到了。所以说这个还是有点不靠谱的啊，我们没办法实现。我们选材mycat试试

MyCat读写分离高可用测试

安装使用mycat，贴一下mycat的配置文件

[root@YC-ss1 conf]# cat schema.xml
<?xml version="1.0"?>
<!DOCTYPE mycat:schema SYSTEM "schema.dtd">
<mycat:schema xmlns:mycat="http://io.mycat/">

    <schema name="TESTDB" checkSQLschema="false" sqlMaxLimit="100">
        <table name="customer" primaryKey="aid"  dataNode="tp1" />
        <table name="district" primaryKey="aid"  dataNode="tp1" />
        <table name="history" primaryKey="aid"  dataNode="tp1" />
        <table name="item" primaryKey="aid"  dataNode="tp1" />
        <table name="new_orders" primaryKey="aid"  dataNode="tp1" />
        <table name="order_line" primaryKey="aid"  dataNode="tp1" />
        <table name="orders" primaryKey="aid"  dataNode="tp1" />
        <table name="stock" primaryKey="aid"  dataNode="tp1" />
        <table name="warehouse" primaryKey="aid"  dataNode="tp1" />
    </schema>

    <dataNode name="tp1" dataHost="dbhost239" database="tpcc" />

    <dataHost name="dbhost239" maxCon="1000" minCon="10" balance="1"
                writeType="0" dbType="mysql" dbDriver="native" switchType="-1"  slaveThreshold="100">
                <heartbeat>select user()</heartbeat>
                <!-- can have multi write hosts -->
                <writeHost host="hostM1" url="172.16.128.239:3306" user="root"
                        password="123456">
                </writeHost>
                <writeHost host="hostS1" url="172.16.128.240:3306" user="root"
                        password="123456" />
                <writeHost host="hostS2" url="172.16.128.241:3306" user="root"
                        password="123456" />
                <writeHost host="hostS3" url="172.16.128.242:3306" user="root"
                        password="123456" />
    </dataHost>

</mycat:schema>

不在赘述，直接通过节点一load数据

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_load 172.16.128.239:3306  tpcc root "123456" 10
*************************************
*** ###easy### TPC-C Data Loader  ***
*************************************
<Parameters>
     [server]: 172.16.128.239
     [port]: 3306
     [DBname]: tpcc
       [user]: root
       [pass]: 123456
  [warehouse]: 10
TPCC Data Load Started...
Loading Item
.................................................. 50000
.................................................. 100000
Item Done.
Loading Warehouse
Loading Stock Wid=1
.................................................. 50000
.................................................. 100000
 Stock Done.
Loading District
Loading Stock Wid=2
.................................................. 50000
.................................................. 100000

然后先压测：

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_start -h 127.0.0.1 -P 8066 -d TESTDB -u root -p 123456 -w 10 -c 8 -r 120 -l 300  -f tpcc_0710

看一下压测结果

 250, 1320(0):1.793|4.080, 1320(0):0.370|1.357, 133(0):0.160|0.185, 131(0):1.993|2.155, 132(0):2.476|3.372
 260, 1154(0):1.695|3.316, 1153(0):0.371|1.664, 114(0):0.138|0.144, 116(0):1.936|3.117, 114(0):3.079|3.506
 270, 1076(0):1.780|3.807, 1076(0):0.359|0.706, 108(0):0.155|0.195, 107(0):1.830|3.335, 109(0):2.296|2.368
 280, 1094(0):1.682|2.435, 1092(0):0.362|1.352, 109(0):0.142|0.163, 110(0):1.963|2.259, 109(0):2.760|2.868
 290, 1264(0):1.668|2.191, 1269(0):0.339|0.744, 127(0):0.159|0.188, 127(0):1.847|1.943, 126(0):2.909|3.890
 300, 1169(0):1.605|4.055, 1166(0):0.336|1.421, 117(0):0.160|0.192, 117(0):1.874|3.247, 118(0):2.456|2.595

STOPPING THREADS........

<Raw Results>
  [] sc:35560  lt:3  rt:0  fl:0
  [] sc:35563  lt:1  rt:0  fl:0
  [] sc:3557  lt:0  rt:0  fl:0
  [] sc:3557  lt:0  rt:0  fl:0
  [] sc:3557  lt:0  rt:0  fl:0
 in 300 sec.

<Raw Results2(sum ver.)>
  [] sc:35560  lt:3  rt:0  fl:0
  [] sc:35565  lt:1  rt:0  fl:0
  [] sc:3557  lt:0  rt:0  fl:0
  [] sc:3557  lt:0  rt:0  fl:0
  [] sc:3557  lt:0  rt:0  fl:0 

<Constraint Check> (all must be [OK])
 [transaction percentage]
        Payment: 43.48% (>=43.0%) [OK]
   Order-Status: 4.35% (>= 4.0%) [OK]
       Delivery: 4.35% (>= 4.0%) [OK]
    Stock-Level: 4.35% (>= 4.0%) [OK]
 [response time (at least 90% passed)]
      New-Order: 99.99%  [OK]
        Payment: 100.00%  [OK]
   Order-Status: 100.00%  [OK]
       Delivery: 100.00%  [OK]
    Stock-Level: 100.00%  [OK]

<TpmC>
                 7112.600 TpmC

由于大部分压测都是更新数据，所以都路由到了239VIP上。我们可以直接测一下读取

(root@127.0.0.1) [TESTDB]> select *,@@server_id from item limit 1;
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
| aid | i_id | i_im_id | i_name                | i_price | i_data                                  | @@server_id |
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
|  35 |    1 |    1320 | lUPdQrxNjkFvKSw698smv |   11.14 | jtIe2rjbaByDZp8wXr9QBbNFjV7VOIL8BdmBSTk |  1282403306 |
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
1 row in set (0.37 sec)

(root@127.0.0.1) [TESTDB]> select *,@@server_id from item limit 1;
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
| aid | i_id | i_im_id | i_name                | i_price | i_data                                  | @@server_id |
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
|  35 |    1 |    1320 | lUPdQrxNjkFvKSw698smv |   11.14 | jtIe2rjbaByDZp8wXr9QBbNFjV7VOIL8BdmBSTk |  1282413306 |
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
1 row in set (0.05 sec)
(root@127.0.0.1) [TESTDB]> select *,@@server_id from item limit 1;
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
| aid | i_id | i_im_id | i_name                | i_price | i_data                                  | @@server_id |
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
|  35 |    1 |    1320 | lUPdQrxNjkFvKSw698smv |   11.14 | jtIe2rjbaByDZp8wXr9QBbNFjV7VOIL8BdmBSTk |  1282423306 |
+-----+------+---------+-----------------------+---------+-----------------------------------------+-------------+
1 row in set (0.07 sec)

读的请求会路由倒三台机器上。现在假设240的主机挂掉，VIP239从240飘到241，我们看一下mycat会不会把连接有没有问题。 首先给一个持续的压测，

[root@YC-ss1 tpcc-mysql-autoinc-pk-master]# ./tpcc_start -h 127.0.0.1 -P 8066 -d TESTDB -u root -p 123456 -w 10 -c 8 -r 120 -l 1800

在压测的过程中我们停掉240主机，VIP会切换的241上。我们再看压测结果。

1184, HY000, Connection refused
payment 7:1
1184, HY000, Connection refused
neword 4:1
1184, HY000, Connection refused
payment 6:1
1184, HY000, Connection refused
payment 5:1
1184, HY000, Connection refused
payment 1:1
1184, HY000, Connection refused
payment 0:1
1184, HY000, Connection refused
neword 3:1
1184, HY000, Connection refused
payment 6:1
1184, HY000, Connection refused
neword 4:1
1184, HY000, Connection refused
payment 7:1
1184, HY000, Connection refused
  70, 723(2):1.654|47.728, 726(5):0.655|76.966, 72(0):0.164|0.298, 72(0):1.957|2.127, 72(0):2.550|2.571
  80, 1178(0):1.769|2.702, 1182(0):0.416|0.870, 118(0):0.159|0.288, 117(0):1.938|1.971, 118(0):2.769|3.452
  90, 1332(0):1.766|4.103, 1330(0):0.425|1.254, 134(0):0.175|0.178, 133(0):2.207|2.644, 134(0):3.027|3.508
 100, 901(0):1.822|2.738, 902(0):0.431|1.109, 90(0):0.152|0.158, 90(0):2.040|2.049, 90(0):2.991|3.176
 -------
 <Raw Results>
  [] sc:22875  lt:3  rt:81654  fl:40
  [] sc:22879  lt:5  rt:87541  fl:40
  [] sc:2288  lt:0  rt:9995  fl:5
  [] sc:2287  lt:0  rt:13993  fl:7
  [] sc:2291  lt:0  rt:5997  fl:3
 in 240 sec.

<Raw Results2(sum ver.)>
  [] sc:22881  lt:3  rt:81655  fl:40
  [] sc:22881  lt:5  rt:87552  fl:40
  [] sc:2288  lt:0  rt:9995  fl:5
  [] sc:2287  lt:0  rt:13993  fl:7
  [] sc:2291  lt:0  rt:5997  fl:3 

<Constraint Check> (all must be [OK])
 [transaction percentage]
        Payment: 43.48% (>=43.0%) [OK]
   Order-Status: 4.35% (>= 4.0%) [OK]
       Delivery: 4.35% (>= 4.0%) [OK]
    Stock-Level: 4.35% (>= 4.0%) [OK]
 [response time (at least 90% passed)]
      New-Order: 99.99%  [OK]
        Payment: 99.98%  [OK]
   Order-Status: 100.00%  [OK]
       Delivery: 100.00%  [OK]
    Stock-Level: 100.00%  [OK]

<TpmC>
                 5719.500 TpmC

可以看到，在keepalived切换过程中，连接是失败的，但是最后VIP切换成功以后所有的连接就正常了。所以说使用MYCAT来实现读写分离是可行的。但是也有一个问题，session close的时候会有问题，但是新链接能够保证是没有问题的，从最后压测结果也是可以看出来的。但是从基本也满足了我们的基本需求。