MySQL ProxySQL读写分离实践
目的
在上一篇文章MySQL ProxySQL读写分离使用初探里初步介绍了ProxySQL的使用,本文继续介绍它的一些特点和DBProxy的性能差异。深入一些去了解ProxySQL,通过例子来说明ProxySQL的一些特性和用sysbench对其进行测试来对比分析说明。
环境:
Distributor ID: Ubuntu
Description : Ubuntu 14.04.5 LTS
Release : 14.04
Codename : trusty
MySQL Master :192.168.200.202
MySQL Slave :192.168.200.132
APP IP :192.168.200.25/64
测试
本文测试环境是在上一篇文章的基础上进行的,所以已经做了一主一从的读写分离。在此基础上,若从库挂了,会怎么样?这里先把从库(132)shutdown掉,看看读去了哪里。
tips:如何修改管理接口的用户名密码?除了初始化时候修改配置文件,还有一个方法是在管理端口设置:
admin@127.0.0.1 : (none) 12:52:53>set admin-admin_credentials='zjy:zjy'; Query OK, 1 row affected (0.00 sec)
admin模块修改(select * from global_variables where variable_name like 'admin%';)需要用admin加载:
admin@127.0.0.1 : (none) 12:53:02>load admin variables to runtime;
Query OK, 0 rows affected (0.00 sec) admin@127.0.0.1 : (none) 12:53:33>save admin variables to disk;
Query OK, 9 rows affected (0.00 sec)
1,从库不可用
① 关闭从库
当前ProxySQL下后端MySQL的运行情况:
admin@127.0.0.1 : (none) 11:25:26>select hostgroup_id,hostname,port,status from runtime_mysql_servers;
+--------------+-----------------+------+--------+
| hostgroup_id | hostname | port | status |
+--------------+-----------------+------+--------+
| 100 | 192.168.200.202 | 3306 | ONLINE |
| 1000 | 192.168.200.132 | 3306 | ONLINE |
+--------------+-----------------+------+--------+
shutdown从库(132)后,后端MySQL的运行情况:
admin@127.0.0.1 : (none) 11:33:24>select hostgroup_id,hostname,port,status from runtime_mysql_servers;
+--------------+-----------------+------+---------+
| hostgroup_id | hostname | port | status |
+--------------+-----------------+------+---------+
| 100 | 192.168.200.202 | 3306 | ONLINE |
| 1000 | 192.168.200.132 | 3306 | SHUNNED |
+--------------+-----------------+------+---------+
此时读的操作会报超时:
sbuser@192.168.200.24 : sbtest 11:30:40>select * from x;
ERROR 9001 (HY000): Max connect timeout reached while reaching hostgroup 1000 after 10000ms
原因是proxysql的核心都在规则,shutdown从之后,proxysql还是想路由到 hostgroup=1000,它不会自动选择默认的100(mysql_users里配置的default_hostgroup) 。
这里解决的办法是:在mysql_servers的hostgroup 1000 里面要插一条主库的记录,然后把weight设小,当读不到从库,回去主库查询。
admin@127.0.0.1 : (none) 11:50:13>insert into mysql_servers(hostgroup_id,hostname,port,weight,max_connections,max_replication_lag,comment) values(1000,'192.168.200.202',3306,1,1000,10,'test proxysql');
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 11:50:42>update mysql_servers set weight=9 where hostgroup_id=1000 and hostname='192.168.200.132';
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 11:53:41>select hostgroup_id,hostname,port,weight from mysql_servers;
+--------------+-----------------+------+--------+
| hostgroup_id | hostname | port | weight |
+--------------+-----------------+------+--------+
| 100 | 192.168.200.202 | 3306 | 1 |
| 1000 | 192.168.200.132 | 3306 | 9 |
| 1000 | 192.168.200.202 | 3306 | 1 |
+--------------+-----------------+------+--------+ admin@127.0.0.1 : (none) 11:54:03>load mysql servers to runtime;
Query OK, 0 rows affected (0.01 sec) admin@127.0.0.1 : (none) 11:54:28>save mysql servers to disk;
Query OK, 0 rows affected (0.00 sec) admin@127.0.0.1 : (none) 11:54:38>select hostgroup_id,hostname,port,status from runtime_mysql_servers;
+--------------+-----------------+------+---------+
| hostgroup_id | hostname | port | status |
+--------------+-----------------+------+---------+
| 100 | 192.168.200.202 | 3306 | ONLINE |
| 1000 | 192.168.200.132 | 3306 | SHUNNED |
| 1000 | 192.168.200.202 | 3306 | ONLINE |
+--------------+-----------------+------+---------+
此时读的操作正常:
sbuser@192.168.200.24 : sbtest 11:52:37>select * from x;
+------+
| id |
+------+
| 123 |
| 123 |
| 123 |
+------+
3 rows in set (0.01 sec)
说明从关闭了之后读操作确实去主上执行了。当从库恢复之后,以后的读操作主库也可以处理 1/10 的读请求。
② 从库延迟/从库停止复制
在上一篇文章中已经建立了监控账号:proxysql,由于需要执行show slave status的命令来获得延迟时间,所以需要权限SUPER 和 REPLICATION CLIENT。并且需要设置mysql_servers.max_replication_lag的值,由于mysql_servers.max_replication_lag仅适用于从,但也可以将其配置为所有主机,无论是从还是主(不会有任何影响)。
-- 设置监控账号权限
dba@192.168.200.202 : sbtest 10:44:38>GRANT SUPER, REPLICATION CLIENT ON *.* TO 'proxysql'@'192.168.200.24' IDENTIFIED BY PASSWORD '*BF27B4C7AAD278126E228AA8427806E870F64F39';
Query OK, 0 rows affected (0.00 sec) -- 设置延迟的阈值
admin@127.0.0.1 : (none) 11:04:50>UPDATE mysql_servers SET max_replication_lag=5; Query OK, 3 rows affected (0.00 sec) -- 应用配置
admin@127.0.0.1 : (none) 11:04:54>load mysql servers to runtime;
Query OK, 0 rows affected (0.01 sec) admin@127.0.0.1 : (none) 11:05:04>save mysql servers to disk;
Query OK, 0 rows affected (0.01 sec)
主从复制正常的情况下,后端MySQL的情况:
admin@127.0.0.1 : (none) 11:05:13>select hostgroup_id,hostname,port,status,max_replication_lag from runtime_mysql_servers;
+--------------+-----------------+------+--------+---------------------+
| hostgroup_id | hostname | port | status | max_replication_lag |
+--------------+-----------------+------+--------+---------------------+
| 1000 | 192.168.200.132 | 3306 | ONLINE | 5 |
| 1000 | 192.168.200.202 | 3306 | ONLINE | 5 |
| 100 | 192.168.200.202 | 3306 | ONLINE | 5 |
+--------------+-----------------+------+--------+---------------------+
3 rows in set (0.00 sec)
从库执行stop slave之后,后端MySQL的情况:
admin@127.0.0.1 : (none) 11:06:52>select hostgroup_id,hostname,port,status,max_replication_lag from runtime_mysql_servers;
+--------------+-----------------+------+---------+---------------------+
| hostgroup_id | hostname | port | status | max_replication_lag |
+--------------+-----------------+------+---------+---------------------+
| 1000 | 192.168.200.132 | 3306 | SHUNNED | 5 |
| 1000 | 192.168.200.202 | 3306 | ONLINE | 5 |
| 100 | 192.168.200.202 | 3306 | ONLINE | 5 |
+--------------+-----------------+------+---------+---------------------+
3 rows in set (0.00 sec)
此时,132从库不可用,读都到了HG 1000的202上去了,可以自行测试。 也可以在日志里看到:
2017-05-11 11:06:43 MySQL_HostGroups_Manager.cpp:934:replication_lag_action(): [WARNING] Shunning server 192.168.200.132:3306 with replication lag of 60 second
日志显示延迟60s,这个是怎么回事?这里需要说明下几个变量:
mysql-monitor_replication_lag_interval:主从延迟检测时间,默认10秒。
mysql-monitor_slave_lag_when_null:当为null时,设置的延迟值,默认为60。
admin@127.0.0.1 : (none) 11:08:35>select * from global_variables where variable_name like 'mysql-monitor%lag%';
+----------------------------------------+----------------+
| variable_name | variable_value |
+----------------------------------------+----------------+
| mysql-monitor_replication_lag_interval | 10000 |
| mysql-monitor_replication_lag_timeout | 1000 |
| mysql-monitor_slave_lag_when_null | 60 |
+----------------------------------------+----------------+
3 rows in set (0.00 sec)
根据mysql_servers.max_replication_lag设置的阈值,这2个参数可以根据自己的情况来设置,比如设置检测时间为1500。延迟的记录也可以通过表来查看:
admin@127.0.0.1 : (none) 11:19:47>select * from mysql_server_replication_lag_log limit 3;
+-----------------+------+------------------+-----------------+----------+-------+
| hostname | port | time_start_us | success_time_us | repl_lag | error |
+-----------------+------+------------------+-----------------+----------+-------+
| 192.168.200.132 | 3306 | 1494472189886932 | 411 | 0 | NULL |
| 192.168.200.202 | 3306 | 1494472189887224 | 372 | NULL | NULL |
| 192.168.200.202 | 3306 | 1494472189887640 | 325 | NULL | NULL |
+-----------------+------+------------------+-----------------+----------+-------+
3 rows in set (0.00 sec)
主从延迟的情况和stop slave的情况一样,只是stop slave是把延迟设置成了60s。
小结:通过上面的测试说明ProxySQL可以在从库不可用时进行下线,不需要人为再进行干预,等到恢复正常之后自动上线提供服务。
2,多路由规则
① 根据库路由
在现有基础上再增加一个主从:
M:192.168.200.97
S:192.168.200.245
授权账号:程序和监控账号
dba@192.168.200.97 : proxysql 12:39:39>GRANT SUPER, REPLICATION CLIENT ON *.* TO 'proxysql'@'192.168.200.24' IDENTIFIED BY PASSWORD '*BF27B4C7AAD278126E228AA8427806E870F64F39';
Query OK, 0 rows affected (0.01 sec)
dba@192.168.200.97 : proxysql 12:42:50>grant select,insert,update,delete on proxysql.* to proxysql@192.168.200.24 identified by 'proxysql';
Query OK, 0 rows affected (0.00 sec)
配置ProxySQL:
admin@127.0.0.1 : (none) 12:43:35>insert into mysql_servers(hostgroup_id,hostname,port,weight,max_connections,max_replication_lag,comment) values(101,'192.168.200.97',3306,1,1000,10,'test proxysql');
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 12:45:15>insert into mysql_servers(hostgroup_id,hostname,port,weight,max_connections,max_replication_lag,comment) values(1001,'192.168.200.245',3306,9,1000,10,'test proxysql');
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 12:45:24>insert into mysql_servers(hostgroup_id,hostname,port,weight,max_connections,max_replication_lag,comment) values(1001,'192.168.200.97',3306,1,1000,10,'test proxysql');
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 12:45:36>insert into mysql_users(username,password,active,default_hostgroup,transaction_persistent) values('proxysql','proxysql',1,101,1);
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 12:46:55>INSERT INTO mysql_query_rules(active,schemaname,match_pattern,destination_hostgroup,apply) VALUES(1,'proxysql','^SELECT.*FOR UPDATE$',101,1);
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 12:56:47>
admin@127.0.0.1 : (none) 12:56:47>INSERT INTO mysql_query_rules(active,schemaname,match_pattern,destination_hostgroup,apply) VALUES(1,'proxysql','^SELECT',1001,1);
Query OK, 1 row affected (0.00 sec) -- 应用保存配置
admin@127.0.0.1 : (none) 12:56:55>load mysql servers to runtime;
admin@127.0.0.1 : (none) 12:57:00>load mysql users to runtime;
admin@127.0.0.1 : (none) 12:57:04>load mysql query rules to runtime;
admin@127.0.0.1 : (none) 12:57:11>save mysql servers to disk;
admin@127.0.0.1 : (none) 12:57:17>save mysql users to disk;
admin@127.0.0.1 : (none) 12:57:21>save mysql query rules to disk;
rules、servers、users信息:
admin@127.0.0.1 : (none) 03:28:11>select rule_id,active,username,schemaname,client_addr,destination_hostgroup,match_pattern,flagIN,flagOUT,apply from mysql_query_rules;
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| rule_id | active | username | schemaname | client_addr | destination_hostgroup | match_pattern | flagIN | flagOUT | apply |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| 3 | 1 | NULL | NULL | NULL | 100 | ^SELECT.*FOR UPDATE$ | 0 | NULL | 1 |
| 4 | 1 | NULL | NULL | NULL | 1000 | ^SELECT | 0 | NULL | 1 |
| 5 | 1 | NULL | proxysql | NULL | 101 | ^SELECT.*FOR UPDATE$ | 0 | NULL | 1 |
| 6 | 1 | NULL | proxysql | NULL | 1001 | ^SELECT | 0 | NULL | 1 |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
4 rows in set (0.00 sec) admin@127.0.0.1 : (none) 03:29:10>select username,default_hostgroup from mysql_users;
+----------+-------------------+
| username | default_hostgroup |
+----------+-------------------+
| sbuser | 100 |
| proxysql | 101 |
+----------+-------------------+
2 rows in set (0.00 sec) admin@127.0.0.1 : (none) 03:29:28>select hostgroup_id,hostname,port,status from mysql_servers;
+--------------+-----------------+------+--------+
| hostgroup_id | hostname | port | status |
+--------------+-----------------+------+--------+
| 1000 | 192.168.200.132 | 3306 | ONLINE |
| 100 | 192.168.200.202 | 3306 | ONLINE |
| 1000 | 192.168.200.202 | 3306 | ONLINE |
| 101 | 192.168.200.97 | 3306 | ONLINE |
| 1001 | 192.168.200.245 | 3306 | ONLINE |
| 1001 | 192.168.200.97 | 3306 | ONLINE |
+--------------+-----------------+------+--------+
6 rows in set (0.00 sec)
模拟app连接:
/Users/jinyizhou [15:32:09] ~$ mysql -uproxysql -pproxysql -h192.168.200.24 -P6033 -A
...
proxysql@192.168.200.24 : (none) 03:32:11>show databases;
+--------------------+
| Database |
+--------------------+
| information_schema |
| proxysql |
+--------------------+
2 rows in set (0.00 sec) proxysql@192.168.200.24 : (none) 03:32:13>use proxysql
Database changed
proxysql@192.168.200.24 : proxysql 03:32:17>show tables;
+--------------------+
| Tables_in_proxysql |
+--------------------+
| xx |
+--------------------+
1 row in set (0.00 sec) proxysql@192.168.200.24 : proxysql 03:32:24>insert into xx values(999);
Query OK, 1 row affected (0.00 sec) proxysql@192.168.200.24 : proxysql 03:35:49>select * from xx;
ERROR 1044 (#4200): Access denied for user 'proxysql'@'192.168.200.24' to database 'proxysql'
只有select的时候没有权限,其他insert,update等都是有权限的,为啥呢?原因是这里的路由关系,ProxySQL的读写分离核心就是路由,这里因为select的路由错了,到了HG为1000的主从上了:
admin@127.0.0.1 : (none) 03:32:28>select hostgroup,schemaname,username,digest_text from stats_mysql_query_digest;
+-----------+--------------------+----------+----------------------------------+
| hostgroup | schemaname | username | digest_text |
+-----------+--------------------+----------+----------------------------------+
| 1000 | proxysql | proxysql | select * from xx |
| 101 | proxysql | proxysql | show tables |
| 101 | information_schema | proxysql | show databases |
| 1000 | information_schema | proxysql | SELECT DATABASE() |
| 101 | information_schema | proxysql | select USER() |
| 101 | information_schema | proxysql | select @@version_comment limit ? |
+-----------+--------------------+----------+----------------------------------+
mysql_query_rules是整个ProxySQL的核心,上篇文章已经对该表进行了说明,在这里对这例子再次讲解下:
rule_id是表的自增主键,路由规则处理是以 rule_id 的顺序进行匹配,若没有找到规则就直接去mysql_users.default_hostgroup字段里找。上面信息中除了select之外的其他操作都找不到规则就直接去users表里取,所以这些操作不会报错。而我们执行的select被rule_id为4的规则匹配上,因为rule_id=4的是匹配所有库并且apply=1表示该正则匹配后,将不再接受其他匹配,直接转发。这样就转发到了HG为1000上面的主机上了,就报没有权限的错误。若apply=0则继续匹配下面,若没有找到路由规则,则返回再看flagOUT是否为NULL,是NULL则直接匹配,否则报错。大致的流程如下:
flagIN, flagOUT, apply: 用来定义路由链 chains of rules 首先会检查 flagIN=0 的规则,以rule_id的顺序;如果都没匹配上,则走这个用户的default_hostgroup
当匹配一条规则后,会检查 apply是否为1,是1则直接转发,不是1则继续匹配,匹配到就转发,否则看flagOUT,
如果不为NULL,并且 flagIN != flagOUT ,则进入以flagIN为上一个flagOUT值的新规则链
如果不为NULL,并且 flagIN = flagOUT,则应用这条规则
如果为NULL,则结束,应用这条规则
通过上面的说明,如何读取到正确的HG呢?这里可以设置apply=0
admin@127.0.0.1 : (none) 04:18:45>update mysql_query_rules set apply=0 where rule_id in (3,4);
Query OK, 2 rows affected (0.00 sec) admin@127.0.0.1 : (none) 04:18:56>load mysql query rules to runtime; Query OK, 0 rows affected (0.00 sec) admin@127.0.0.1 : (none) 04:18:59>save mysql query rules to disk; Query OK, 0 rows affected (0.00 sec) admin@127.0.0.1 : (none) 04:19:01>select rule_id,active,username,schemaname,client_addr,destination_hostgroup,match_pattern,flagIN,flagOUT,apply from mysql_query_rules;
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| rule_id | active | username | schemaname | client_addr | destination_hostgroup | match_pattern | flagIN | flagOUT | apply |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| 3 | 1 | NULL | NULL | NULL | 100 | ^SELECT.*FOR UPDATE$ | 0 | NULL | 0 |
| 4 | 1 | NULL | NULL | NULL | 1000 | ^SELECT | 0 | NULL | 0 |
| 5 | 1 | NULL | proxysql | NULL | 101 | ^SELECT.*FOR UPDATE$ | 0 | NULL | 1 |
| 6 | 1 | NULL | proxysql | NULL | 1001 | ^SELECT | 0 | NULL | 1 |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
4 rows in set (0.00 sec)
和上面一样模拟app连接,得到的信息:发现全部走了正确的路由。
admin@127.0.0.1 : (none) 05:58:55>select hostgroup,schemaname,username,digest_text from stats_mysql_query_digest;
+-----------+------------+----------+----------------------------------+
| hostgroup | schemaname | username | digest_text |
+-----------+------------+----------+----------------------------------+
| 101 | proxysql | proxysql | insert into xx values(?) |
| 1001 | proxysql | proxysql | select * from xx |
| 1001 | proxysql | proxysql | SELECT DATABASE() |
| 101 | proxysql | proxysql | select USER() |
| 101 | proxysql | proxysql | select @@version_comment limit ? |
+-----------+------------+----------+----------------------------------+
5 rows in set (0.00 sec)
查看路由规则的命中情况:
admin@127.0.0.1 : (none) 05:59:19>select * from stats_mysql_query_rules;
+---------+------+
| rule_id | hits |
+---------+------+
| 3 | 0 |
| 4 | 4 |
| 5 | 0 |
| 6 | 4 |
+---------+------+
4 rows in set (0.00 sec)
从上面看到,apply=0 & falgOUT=null,会继续往下找路由,找到了rule_id=6的,直接进行转发。apply=1 直接转发,flagOUT != null 直接转发。
小结:通过上面的测试说明ProxySQL只要设置好路由规则,可以有多个主库。
② 根据用户名路由
和多主路由一样,区别是写入到路由表的字段不一样:
admin@127.0.0.1 : (none) 06:09:20>INSERT INTO mysql_query_rules(active,username,match_pattern,destination_hostgroup,apply) VALUES(1,'proxysql','^SELECT.*FOR UPDATE$',101,1);
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 06:10:09>INSERT INTO mysql_query_rules(active,username,match_pattern,destination_hostgroup,apply) VALUES(1,'proxysql','^SELECT',1001,1);
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 06:10:32>select rule_id,active,username,schemaname,client_addr,destination_hostgroup,match_pattern,flagIN,flagOUT,apply from mysql_query_rules;
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| rule_id | active | username | schemaname | client_addr | destination_hostgroup | match_pattern | flagIN | flagOUT | apply |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| 3 | 1 | NULL | NULL | NULL | 100 | ^SELECT.*FOR UPDATE$ | 0 | NULL | 1 |
| 4 | 1 | NULL | NULL | NULL | 1000 | ^SELECT | 0 | NULL | 1 |
| 1405 | 1 | proxysql | NULL | NULL | 101 | ^SELECT.*FOR UPDATE$ | 0 | NULL | 1 |
| 1406 | 1 | proxysql | NULL | NULL | 1001 | ^SELECT | 0 | NULL | 1 |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
4 rows in set (0.00 sec)
3,flagIN/flahOUT规则链实现多实例(推荐)
和2中的条件一样,先配置ProxySQL的servers,users:
admin@127.0.0.1 : (none) 01:09:52>insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment)
-> values
-> (100, '192.168.200.202', 3306, 1, 10, 'ReadWrite'),
-> (1000, '192.168.200.202', 3306, 1, 10, 'ReadWrite'),
-> (1000, '192.168.200.132', 3306, 9, 10, 'ReadOnly');
Query OK, 3 rows affected (0.00 sec) admin@127.0.0.1 : (none) 01:09:54>insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment)
-> values
-> (101, '192.168.200.97', 3306, 1, 10, 'ReadWrite'),
-> (1001, '192.168.200.97', 3306, 1, 10, 'ReadWrite'),
-> (1001, '192.168.200.245', 3306, 9, 10, 'ReadOnly');
Query OK, 3 rows affected (0.01 sec) admin@127.0.0.1 : (none) 01:11:01>insert into mysql_users(username, password,active,default_hostgroup,transaction_persistent)
-> values
-> ('sbuser', 'sbuser', 1, 100, 1),
-> ('proxysql', 'proxysql', 1, 101, 1);
Query OK, 2 rows affected (0.00 sec) admin@127.0.0.1 : (none) 01:19:44>set mysql-monitor_username='proxysql';
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 01:19:44>set mysql-monitor_password='proxysql';
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 01:56:09>select hostgroup_id,hostname,port,status from mysql_servers;
+--------------+-----------------+------+--------+
| hostgroup_id | hostname | port | status |
+--------------+-----------------+------+--------+
| 100 | 192.168.200.202 | 3306 | ONLINE |
| 1000 | 192.168.200.202 | 3306 | ONLINE |
| 1000 | 192.168.200.132 | 3306 | ONLINE |
| 101 | 192.168.200.97 | 3306 | ONLINE |
| 1001 | 192.168.200.97 | 3306 | ONLINE |
| 1001 | 192.168.200.245 | 3306 | ONLINE |
+--------------+-----------------+------+--------+
6 rows in set (0.00 sec) admin@127.0.0.1 : (none) 01:58:18>select username,default_hostgroup from mysql_users;
+----------+-------------------+
| username | default_hostgroup |
+----------+-------------------+
| sbuser | 100 |
| proxysql | 101 |
+----------+-------------------+
再配置flagOUT/flagIN,flag20是读,flag21是写:
admin@127.0.0.1 : (none) 01:21:34>INSERT INTO mysql_query_rules(rule_id,active,match_pattern,apply,flagOUT) VALUES(49,1,'^SELECT.*FOR UPDATE$',0,21);
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 01:27:18>INSERT INTO mysql_query_rules(rule_id,active,match_pattern,apply,flagOUT) VALUES(50,1,'^SELECT',0,20);
Query OK, 1 row affected (0.00 sec) admin@127.0.0.1 : (none) 01:32:11>insert into mysql_query_rules(active,schemaname,destination_hostgroup,apply,flagIN,flagOUT) values
-> (1,'sbtest',100,1,21,21),
-> (1,'proxysql',101,1,21,21);
Query OK, 2 rows affected (0.00 sec) admin@127.0.0.1 : (none) 01:32:53>insert into mysql_query_rules(active,schemaname,destination_hostgroup,apply,flagIN,flagOUT) values
-> (1,'sbtest',1000,1,20,20),
-> (1,'proxysql',1001,1,20,20);
Query OK, 2 rows affected (0.00 sec) admin@127.0.0.1 : (none) 01:58:28>select rule_id,active,username,schemaname,client_addr,destination_hostgroup,match_pattern,flagIN,flagOUT,apply from mysql_query_rules;
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| rule_id | active | username | schemaname | client_addr | destination_hostgroup | match_pattern | flagIN | flagOUT | apply |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
| 49 | 1 | NULL | NULL | NULL | NULL | ^SELECT.*FOR UPDATE$ | 0 | 21 | 0 |
| 50 | 1 | NULL | NULL | NULL | NULL | ^SELECT | 0 | 20 | 0 |
| 51 | 1 | NULL | sbtest | NULL | 100 | NULL | 21 | 21 | 1 |
| 52 | 1 | NULL | proxysql | NULL | 101 | NULL | 21 | 21 | 1 |
| 53 | 1 | NULL | sbtest | NULL | 1000 | NULL | 20 | 20 | 1 |
| 54 | 1 | NULL | proxysql | NULL | 1001 | NULL | 20 | 20 | 1 |
+---------+--------+----------+------------+-------------+-----------------------+----------------------+--------+---------+-------+
最后保存上线:
-- 应用
load mysql users to runtime;
load mysql servers to runtime;
load mysql variables to runtime;
LOAD MYSQL QUERY RULES TO RUN; -- 保存到磁盘
save mysql users to disk;
save mysql servers to disk;
save mysql variables to disk;
SAVE MYSQL QUERY RULES TO DISK; save mysql users to mem; -- 可以屏蔽看到的明文密码
app连接测试:
1)连接实例202
[zhoujy@localhost ~]$ mysql -usbuser -psbuser -h192.168.200.24 -P6033
...
sbuser@192.168.200.24 : (none) 02:19:41>show databases; +--------------------+
| Database |
+--------------------+
| information_schema |
| sbtest |
+--------------------+
2 rows in set (0.00 sec) sbuser@192.168.200.24 : (none) 02:19:44>use sbtest
Database changed
sbuser@192.168.200.24 : sbtest 02:19:48>show tables;
...
sbuser@192.168.200.24 : sbtest 02:19:57>insert into x values(10000);
...
sbuser@192.168.200.24 : sbtest 02:20:10>select * from x;
...
相关信息:路由的信息都是正确的
admin@127.0.0.1 : (none) 02:24:15>select hostgroup,schemaname,username,digest_text,count_star from stats_mysql_query_digest;
+-----------+--------------------+----------+----------------------------------+------------+
| hostgroup | schemaname | username | digest_text | count_star |
+-----------+--------------------+----------+----------------------------------+------------+
| 1000 | sbtest | sbuser | select * from x | 5 |
| 100 | sbtest | sbuser | insert into x values(?) | 5 |
| 100 | sbtest | sbuser | show tables | 2 |
| 100 | sbtest | sbuser | show databases | 1 |
| 100 | information_schema | sbuser | SELECT DATABASE() | 1 |
| 100 | information_schema | sbuser | show databases | 1 |
| 100 | information_schema | sbuser | select USER() | 1 |
| 100 | information_schema | sbuser | select @@version_comment limit ? | 1 |
+-----------+--------------------+----------+----------------------------------+------------+
8 rows in set (0.00 sec)
--路由命中
admin@127.0.0.1 : (none) 02:25:13>admin@1* from stats_mysql_query_rules;
+---------+------+
| rule_id | hits |
+---------+------+
| 49 | 0 |
| 50 | 6 |
| 51 | 0 |
| 52 | 0 |
| 53 | 5 |
| 54 | 0 |
+---------+------+
6 rows in set (0.00 sec)
结论:通过实例202的账号访问ProxySQL,首先会检查flagIN=0,在其上面进行匹配(Proxysql入口都是flagIN =0,顺序往下), 匹配到之后检查flagOUT,发现 flagOUT不为NULL且flagIN !(0)= flagOUT (20),则进入以flagIN为上一个flagOUT值的新规则链,即20。再去flagIN=20里匹配,最终匹配到了rule_id=53的规则,最后转发。
2)连接实例97
相关情况和上面一样,最终通过rule_id=54进行转发。
建议:若要用ProxySQL来控制多主从实例的读写分离,推荐使用flagIN/flahOUT规则链实现多实例。
4,flagIN/flahOUT规则链实现分库
目的:客户端应用连接上 proxysql 的ip:port,连接时指定分库db名,自动路由到对应的实例、对应的库。
① :环境
APP:192.168.200.25、192.168.200.64 M1:
IP:192.168.200.202
Port:3306
DB:M1、M2、M3 S1:
IP:192.168.200.132
Port:3306
DB:M1、M2、M3 M2:
IP:192.168.200.97
Port:3306
DB:M4、M5、M6 S2:
IP:192.168.200.245
Port:3306
DB:M4、M5、M6 ProxySQL:192.168.200.24
② 搭建
和之前一样先在后端数据库创建程序和检测账号:
--程序账号
GRANT SELECT,INSERT,UPDATE,DELETE ON `mtest%`.* TO 'mtest'@'192.168.200.24' identified by 'mtest';
--健康检测账号
GRANT SUPER, REPLICATION CLIENT ON *.* TO 'proxysql'@'192.168.200.24' IDENTIFIED BY 'proxysql';
配置ProxySQL:
--插入后端用户信息
insert into mysql_users(username, password,active,transaction_persistent) values('mtest','mtest',1,1); --插入后端数据库信息
insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment) values(100,'192.168.200.202',3306,1,10,'test proxysql');
insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment) values(1000,'192.168.200.132',3306,9,10,'test proxysql');
insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment) values(1000,'192.168.200.202',3306,1,10,'test proxysql');
insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment) values(101,'192.168.200.97',3306,1,10,'test proxysql');
insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment) values(1001,'192.168.200.245',3306,9,10,'test proxysql');
insert into mysql_servers(hostgroup_id,hostname,port,weight,max_replication_lag,comment) values(1001,'192.168.200.97',3306,1,10,'test proxysql'); --配置健康检测信息
set mysql-monitor_username='proxysql';
set mysql-monitor_password='proxysql';
应用保存配置:
-- 应用
load mysql users to runtime;
load mysql servers to runtime;
load mysql variables to runtime;
-- 保存到磁盘
save mysql users to disk;
save mysql servers to disk;
save mysql variables to disk;
save mysql users to mem; -- 可以屏蔽看到的明文密码
配置路由规则:
----添加读写分离的路由
--写:写的入口 flagIN=0
INSERT INTO mysql_query_rules(rule_id,active,match_pattern,apply,flagOUT) VALUES(49,1,'^SELECT.*FOR UPDATE$',0,21);
--读:读的入口 flagIN=0
INSERT INTO mysql_query_rules(rule_id,active,match_pattern,apply,flagOUT) VALUES(50,1,'^SELECT',0,20);
--反向匹配,相当于对 match_digest/match_pattern 的匹配取反,非select,即写
INSERT INTO mysql_query_rules(rule_id, active,match_pattern,negate_match_pattern,apply,flagOUT) values(60, 1,'^SELECT',1,0,21); ----为后端服务器配置路由
--读
insert into mysql_query_rules(active,schemaname,destination_hostgroup,apply,flagIN,flagOUT) values(1,'M1',1000,1,20,20),(1,'M2',1000,1,20,20),(1,'M3',1000,1,20,20);
insert into mysql_query_rules(active,schemaname,destination_hostgroup,apply,flagIN,flagOUT) values(1,'M4',1001,1,20,20),(1,'M5',1001,1,20,20),(1,'M6',1001,1,20,20); --写
insert into mysql_query_rules(active,schemaname,destination_hostgroup,apply,flagIN,flagOUT) values(1,'M1',100,1,21,21),(1,'M2',100,1,21,21),(1,'M3',100,1,21,21);
insert into mysql_query_rules(active,schemaname,destination_hostgroup,apply,flagIN,flagOUT) values(1,'M4',101,1,21,21),(1,'M5',101,1,21,21),(1,'M6',101,1,21,21); --连接时,若没有指定数据库,则进行show databases/tables、use 等会超时出错,连接时,默认的数据库是在information_schema,所以写一条根据information_schema库的路由,直接返回错误信息。
insert into mysql_query_rules(rule_id,active,schemaname,apply,flagOUT) values(20,1,'information_schema',0,302);
insert into mysql_query_rules(rule_id,active,apply, flagIN,flagOUT,error_msg) values(9999,1,1, 302,302,'No query rules matched (by ProxySQL)'); --连接时,若没有指定数据库,则可以使用 schemaname.tablename 的形式匹配数据。
insert into mysql_query_rules(rule_id,active,match_pattern,destination_hostgroup,apply,flagIN,flagOUT) values(1000,1,'([\s\`])M(1|2|3)([\.\`])',100,1,302,302);
insert into mysql_query_rules(rule_id,active,match_pattern,destination_hostgroup,apply,flagIN,flagOUT) values(1001,1,'([\s\`])M(4|5|6)([\.\`])',101,1,302,302);
应用规则:
LOAD MYSQL QUERY RULES TO RUN;
SAVE MYSQL QUERY RULES TO DISK;
最终的路由规则如下:
select rule_id,schemaname,destination_hostgroup,match_pattern,negate_match_pattern,flagIN,flagOUT,apply,error_msg from mysql_query_rules;
app连接测试:
~$ mysql -umtest -pmtest -h192.168.200.24 -P6033 -A
...
mtest@192.168.200.24 : (none) 11:27:29>show databases; --触发了定义的路由 ERROR 1148 (42000): No query rules matched (by ProxySQL)
mtest@192.168.200.24 : (none) 11:27:34>select * from M5.mtest5; --可以直接用schema.tables 访问
+------+
| id |
+------+
| 5 |
| 55 |
+------+
2 rows in set (0.00 sec) mtest@192.168.200.24 : (none) 11:27:47>use M1 --切换数据库
Database changed
mtest@192.168.200.24 : M1 11:27:52>show tables; --可以show了
+--------------+
| Tables_in_M1 |
+--------------+
| mtest1 |
+--------------+
1 row in set (0.00 sec) mtest@192.168.200.24 : M1 11:27:56>select * from mtest1;
+------+
| id |
+------+
| 1 |
| 11 |
| 111 |
| 1111 |
+------+
4 rows in set (0.00 sec) mtest@192.168.200.24 : M1 11:28:02>insert into mtest1 values(11111);
Query OK, 1 row affected (0.00 sec) mtest@192.168.200.24 : M1 11:28:11>select * from mtest1;
+-------+
| id |
+-------+
| 1 |
| 11 |
| 111 |
| 1111 |
| 11111 |
+-------+
5 rows in set (0.00 sec) mtest@192.168.200.24 : M1 11:28:12>show databases; --可以show了
+--------------------+
| Database |
+--------------------+
| information_schema |
| M1 |
| M2 |
| M3 |
+--------------------+
4 rows in set (0.00 sec) mtest@192.168.200.24 : M1 11:28:20>use M5 --切换到另一个实例的db
Database changed
mtest@192.168.200.24 : M5 11:28:52>show databases;
+--------------------+
| Database |
+--------------------+
| information_schema |
| M4 |
| M5 |
| M6 |
+--------------------+
4 rows in set (0.00 sec) mtest@192.168.200.24 : M5 11:28:55>select * from mtest5;
+------+
| id |
+------+
| 5 |
| 55 |
+------+
2 rows in set (0.00 sec) mtest@192.168.200.24 : M5 11:29:03>insert into mtest5 values(555);
Query OK, 1 row affected (0.01 sec) mtest@192.168.200.24 : M5 11:29:12>select * from mtest5;
+------+
| id |
+------+
| 5 |
| 55 |
| 555 |
+------+
3 rows in set (0.00 sec)
查看路由命中率:
select active,hits, mysql_query_rules.rule_id, schemaname,match_pattern,destination_hostgroup hostgroup,flagIn,flagOUT FROM mysql_query_rules NATURAL JOIN stats.stats_mysql_query_rules ORDER BY mysql_query_rules.rule_id;
查看SQL统计信息:
admin@127.0.0.1 : (none) 11:36:46>select hostgroup,schemaname,username,substr(digest_text,120,-120),count_star from stats_mysql_query_digest;
+-----------+--------------------+----------+----------------------------------+------------+
| hostgroup | schemaname | username | substr(digest_text,120,-120) | count_star |
+-----------+--------------------+----------+----------------------------------+------------+
| 101 | M5 | mtest | show databases | 1 |
| 1000 | M1 | mtest | SELECT DATABASE() | 1 |
| 101 | M5 | mtest | insert into mtest5 values(?) | 1 |
| 100 | M1 | mtest | show databases | 1 |
| 100 | M1 | mtest | insert into mtest1 values(?) | 1 |
| 1000 | M1 | mtest | select * from mtest1 | 2 |
| 1001 | M5 | mtest | select * from mtest5 | 2 |
| 100 | M1 | mtest | show tables | 1 |
| 101 | information_schema | mtest | select * from M5.mtest5 | 1 |
| 0 | information_schema | mtest | show databases | 1 |
| 0 | information_schema | mtest | SELECT DATABASE() | 1 |
| 0 | information_schema | mtest | select USER() | 1 |
| 0 | information_schema | mtest | select @@version_comment limit ? | 1 |
+-----------+--------------------+----------+----------------------------------+------------+
具体的说明可以看ProxySQL之读写分离与分库路由演示,到此读写分离的测试介绍完毕,
5,查询重写
查询重写这种对于线上环境SQL问题引起的紧急故障处理还是很有用处的。如果定位到了问题所在,必须修改SQL,时间紧急,这时查询重写这个东西就非常有用了。类似于MySQL5.7的查询重写插件。这里做下相关的说明:
ProxySQL的核心就是路由,查询重写也只是添加一条路由而已,在4的基础上进行测试:
select * from mtest1 order by id
重写成
select * from mtest1
添加路由:
--查询的路由,flagIN=0,当匹配上规则后进行重写,并且不应用,而通过flagOUT下去继续查询
INSERT INTO mysql_query_rules (rule_id,active,match_pattern,replace_pattern,apply,flagOUT) VALUES (48,1,'(.*)order by id','\1',0,20);
其实查询重写的实现在proxysql中也实现为正则匹配替换,表示当proxysql匹配到<若干字符>order by id这个模式后,就将这个模式的order by id去掉。那么\1是什么意思呢,就是sed的向前引用。
加载load和save完rules之后,查看是否重写成功:
--初始
#查询路由命中信息
admin@127.0.0.1 : (none) 02:44:52>select * from stats_mysql_query_rules; +---------+------+
| rule_id | hits |
+---------+------+
| 20 | 0 |
| 48 | 0 |
| 49 | 0 |
| 50 | 0 |
| 60 | 0 |
| 61 | 0 |
| 62 | 0 |
| 63 | 0 |
| 64 | 0 |
| 65 | 0 |
| 66 | 0 |
| 67 | 0 |
| 68 | 0 |
| 69 | 0 |
| 70 | 0 |
| 71 | 0 |
| 72 | 0 |
| 1000 | 0 |
| 1001 | 0 |
| 9999 | 0 |
+---------+------+
20 rows in set (0.00 sec) #查询统计信息
admin@127.0.0.1 : (none) 02:45:09>select * from stats_mysql_query_digest;
Empty set (0.00 sec) --操作
~$ mysql -umtest -pmtest -h192.168.200.24 -P6033 -A
...
mtest@192.168.200.24 : (none) 02:45:27>use M1 Database changed
mtest@192.168.200.24 : M1 02:45:31>show tables;
+--------------+
| Tables_in_M1 |
+--------------+
| mtest1 |
+--------------+
1 row in set (0.00 sec) mtest@192.168.200.24 : M1 02:45:33>select * from mtest1;
+-------+
| id |
+-------+
| 1 |
| 11 |
| 111 |
| 1111 |
| 11111 |
+-------+
5 rows in set (0.00 sec) mtest@192.168.200.24 : M1 02:45:37>select * from mtest1 order by id;
+-------+
| id |
+-------+
| 1 |
| 11 |
| 111 |
| 1111 |
| 11111 |
+-------+
5 rows in set (0.00 sec) mtest@192.168.200.24 : M1 02:45:46>select * from mtest1 order by id;
+-------+
| id |
+-------+
| 1 |
| 11 |
| 111 |
| 1111 |
| 11111 |
+-------+
5 rows in set (0.01 sec) ----以上执行了2次order by id和1此没有order by id的查询,去查询统计应该得到的值是3次没有order by id的查询。 --验证
#查询统计信息,查看没有order by id的SQL出现了3次,没有出现有order by id的SQL
admin@127.0.0.1 : (none) 02:49:49>select hostgroup,schemaname,digest_text,count_star from stats_mysql_query_digest;
+-----------+--------------------+----------------------------------+------------+
| hostgroup | schemaname | digest_text | count_star |
+-----------+--------------------+----------------------------------+------------+
| 1000 | M1 | select * from mtest1 | 3 |
| 100 | M1 | show tables | 1 |
| 0 | information_schema | SELECT DATABASE() | 1 |
| 0 | information_schema | select USER() | 1 |
| 0 | information_schema | select @@version_comment limit ? | 1 |
+-----------+--------------------+----------------------------------+------------+
5 rows in set (0.00 sec) #重写查询的路由命中了2次
admin@127.0.0.1 : (none) 02:50:12>select * from stats_mysql_query_rules;
+---------+------+
| rule_id | hits |
+---------+------+
| 20 | 1 |
| 48 | 2 |
| 49 | 0 |
| 50 | 1 |
| 60 | 1 |
| 61 | 3 |
| 62 | 0 |
| 63 | 0 |
| 64 | 0 |
| 65 | 0 |
| 66 | 0 |
| 67 | 1 |
| 68 | 0 |
| 69 | 0 |
| 70 | 0 |
| 71 | 0 |
| 72 | 0 |
| 1000 | 0 |
| 1001 | 0 |
| 9999 | 1 |
+---------+------+
20 rows in set (0.00 sec)
从上面的结果看,查询重写已经测试通过。到此,关于ProxySQL的相关测试已经结束,下面分析下和DBProxy的特性差别和性能差异。
性能测试
环境:
ProxySQL:192.168.200.24
DBProxy :192.168.200.24 M:
IP:192.168.200.202
Port:3306
DB:sbtest
S:
IP:192.168.200.132
Port:3306
DB:sbtest
读写混合(oltp_read_write.lua)测试对比:
直连数据库:
./bin/sysbench --test=./share/sysbench/oltp_read_write.lua --mysql-host=192.168.200.202 --mysql-port=3306 --mysql-user=sbuser --mysql-password=sbuser --mysql-db=sbtest --report-interval=10 --max-requests=0 --time=120 --threads=1 --tables=3 --table-size=1000000 prepare/run/cleanup
ProxySQL连接数据库:
./bin/sysbench --test=./share/sysbench/oltp_read_write.lua --mysql-host=192.168.200.24 --mysql-port=6033 --mysql-user=sbuser --mysql-password=sbuser --mysql-db=sbtest --report-interval=10 --max-requests=0 --time=120 --threads=1 --tables=3 --table-size=1000000 --skip-trx=on --db-ps-mode=disable --mysql-ignore-errors=1062 prepare/run/cleanup
DBProxy连接数据库
./bin/sysbench --test=./share/sysbench/oltp_read_write.lua --mysql-host=192.168.200.24 --mysql-port=3308 --mysql-user=sbuser --mysql-password=sbuser --mysql-db=sbtest --report-interval=10 --max-requests=0 --time=120 --threads=1 --tables=3 --table-size=1000000 --skip-trx=on --db-ps-mode=disable --mysql-ignore-errors=1062 prepare/run/cleanup
测试的线程:1、2、4、8、16、32、64、128
把上面数据以曲线图的形式表现:
TPS:
QPS:
测试小结:
在读写混合的模式下:线程越少差距越大,测试结果和美团点评DBProxy读写分离使用说明里的测试报告基本吻合,这里主要对比ProxySQL和DBProxy的性能情况,从上图看到二者性能差不多,不过DBProxy的CPU消耗是ProxySQL的1到1.5倍。
总结:
通过上面的一些基本介绍,大致了解了ProxySQL读写分离功能的使用,关于ProxySQL的其他功能内容在手册里有了详尽的介绍,具体的情况请参考手册说明。现在大致整理下ProxySQL和DBproxy的差别:
①:连接池,是 multiplexing。
②:强大的正则路由,可以自己干预读写路由算法。
③:从库不可用自动下线,不需要人为干预,支持多主库。
④:支持重写SQL。
⑤:足够轻量,配置简单。
但是在安全配置上面,DBProxy比ProxySQL要强,ProxySQL前后端账号未分离,可以通过mysql_users查看,前后端公用一个账号,但是在runting_mysql_users里面前后端账号是分离的(backend、frontend)。其他的相关安全可以参考美团点评DBProxy读写分离使用说明。最后根据情况选择到底使用哪个proxy,要是使用的是MySQL Server 5.7,因DBProxy没有对5.7进行测试,所以推荐使用ProxySQL。至于如何防止ProxySQL的单点问题,也可以用lvs来解决,具体的说可以看LVS+Keepalived实现DBProxy的高可用。
参考文档
https://github.com/sysown/proxysql/wiki
http://proxysql.blogspot.jp/2015/09/proxysql-tutorial-setup-in-mysql.html
https://severalnines.com/blog/how-set-read-write-split-galera-cluster-using-proxysql
MySQL ProxySQL读写分离实践的更多相关文章
- MySQL ProxySQL读写分离使用初探
目的 在美团点评DBProxy读写分离使用说明文章中已经说明了使用目的,本文介绍ProxySQL的使用方法以及和DBProxy的性能差异.具体的介绍可以看官网的相关说明,并且这个中间件也是percon ...
- ProxySQL 读写分离实践
前言 ProxySQL是一个高性能的MySQL中间件,拥有强大的规则引擎.具有以下特性: 连接池,而且是 multiplexing 主机和用户的最大连接数限制 自动下线后端DB 延迟超过阀值 ping ...
- Centos7源码安装mysql及读写分离,互为主从
Linux服务器 -源码安装mysql 及读写分离,互为主从 一.环境介绍: Linux版本: CentOS 7 64位 mysq版本: mysql-5.6.26 这是我安装时所使用的版本, ...
- 应用集成mycat,实现mycat的高可用与mysql的读写分离
前言 开心一刻 一个女人自朋友圈写道:我家老公昨天和别人家的老婆出去旅游,迄今未归,我则被别人家的老公折腾了一天,好累哦! 圈子下面,评论无数,老公在下面评论到:能不能好好说话,我只不过陪女儿去毕业旅 ...
- MySQL Router实现MySQL的读写分离
1.简介 MySQL Router是MySQL官方提供的一个轻量级MySQL中间件,用于取代以前老版本的SQL proxy. 既然MySQL Router是一个数据库的中间件,那么MySQL Rout ...
- ProxySQL读写分离
我们首先看一下自己的环境: MHA已经搭建: master: slave: slave: MHA manager在172.16.16.34,配置文件如下: [root@localhost bin]# ...
- mysqlfailover高可用与proxysql读写分离配置
proxysql官方推荐两种高可用方案: 1.MHA+proxysql 2.mysqlrpladmin+proxysql MySQLfailover工具包含在mysqlrpladmin工具中,所以两者 ...
- MySQL的读写分离的几种选择
MySQL的读写分离的几种选择 MySQL主从复制(Master-Slave)与读写分离(MySQL-Proxy)实践 原址如下: http://heylinux.com/archives/1004. ...
- MHA+ProxySQL 读写分离高可用
文档结构如下: 1.ProxySQL说明 ProxySQL是mysql的一款中间件的产品,是灵活的mysql代理层,可以实现读写分离,支持query路由器的功能,支持动态指定sql进行缓存,支持动态加 ...
随机推荐
- javaWEB与EL表达式
EL表达式1. EL是JSP内置的表达式语言! * jsp2.0开始,不让再使用java脚本,而是使用el表达式和动态标签来替代java脚本! * EL替代的是<%= ... %>,也 ...
- phpcms2008常用函数小结
{$head[title]} 页面标题,用法: <title>{$head[title]}-{$PHPCMS[sitename]}</title> {$PHPCMS[siten ...
- (删)Java线程同步实现一:synchronzied和wait()/notify()
上面文章(2.Java多线程总结系列:Java的线程控制实现)讲到了如何对线程进行控制,其中有一个是线程同步问题.下面我们先来看一个例子: 1.一个典型的Java线程安全例子 package com. ...
- SQL语句中的日期查询
我们先创建一个表: CREATE TABLE `student` ( `id` INT NOT NULL AUTO_INCREMENT , `name` VARCHAR(10) NULL COMMEN ...
- C++11 新特性总结
前言 转载请注明出处,感谢! C++11 的新特性 1 变量和基本类型 1.1 long long 类型 扩展精度浮点数,10位有效数字 1.2 列表初始化 初始化的几种不同形式,其中用花括号来初始化 ...
- Kubernetes DNS 简介
环境 $ sudo lsb_release -a No LSB modules are available. Distributor ID: Ubuntu Description: Ubuntu 16 ...
- MySQL元数据库——information_schema
平时使用MySQL客户端操作数据库的同学,只要稍微留神都会发现,除了我们建的库之外,还经常看到三个数据库的影子: 1. information_schema 2. performance_schema ...
- Linux-进程描述(3)之进程状态僵尸进程与孤儿进程
进程状态 进程状态反映进程执行过程的变化.这些状态随着进程的执行和外界条件的变化而转换.为了弄明正正在运行的进程是什么意思,我们需要知道进程的不同状态.一个进程可以有多个状态(在Linux内核中,进程 ...
- 第二章 Oracle数据库应用
第二章 Oracle数据库应用2.1 表空间和用户权限下管理 2.1.1 表空间 2.1.1.1 分类: 永久性表空间 临时性表空间 ...
- Unity属性的封装、继承、方法隐藏
(一)Unity属性封装.继承.方法隐藏的学习和总结 一.属性的封装 1.属性封装的定义:通过对属性的读和写来保护类中的域. 2.格式例子: private string departname; // ...