MySQL 读写分离与负载均衡

时间:2021-12-08 07:35:06

MySQL 数据库的读写分离和负载均衡一般是通过第三方软件来实现的。 也可以通过mysql驱动程序来实现,如com.mysql.jdbc.ReplicationDriver。

详细文档参见:http://dev.mysql.com/doc/refman/5.5/en/connector-j-info.html

import java.sql.Connection;
import java.sql.ResultSet;
import java.util.Properties;
  
import com.mysql.jdbc.ReplicationDriver;
  
public class ReplicationDriverDemo {
  
  public static void main(String[] args) throws Exception {
    ReplicationDriver driver = new ReplicationDriver();
  
    Properties props = new Properties();
  
    // We want this for failover on the slaves
    props.put("autoReconnect""true");
  
    // We want to load balance between the slaves
    props.put("roundRobinLoadBalance""true");
  
    props.put("user""foo");
    props.put("password""bar");
  
    //
    // Looks like a normal MySQL JDBC url, with a
    // comma-separated list of hosts, the first
    // being the 'master', the rest being any number
    // of slaves that the driver will load balance against
    //
  
    Connection conn =
        driver.connect("jdbc:mysql:replication://master,slave1,slave2,slave3/test",
            props);
  
    //
    // Perform read/write work on the master
    // by setting the read-only flag to "false"
    //
  
    conn.setReadOnly(false);
    conn.setAutoCommit(false);
    conn.createStatement().executeUpdate("UPDATE some_table ....");
    conn.commit();
  
    //
    // Now, do a query from a slave, the driver automatically picks one
    // from the list
    //
  
    conn.setReadOnly(true);
  
    ResultSet rs =
      conn.createStatement().executeQuery("SELECT a,b FROM alt_table");
  
     .......
  }
}

读写分离:

jdbc:mysql:replication://master:3306,slave1:3306,slave2:3306/dbname

When using the following connection string: jdbc:mysql:replication://dbmaster:3306,dbslave1:3306,dbslave2:3306/dbname

dbmaster is used for all write connections as expected and dbslave1 is used for all read connections, but dbslave2 is never used. I would have expected distributed reads between dbslave1 and dbslave2.

原理是:ReplicationDriver生成代理的connection对象,当设置这个connection.readOnly=true时,连接slave,当connection.readOnly=false时,连接master

负载均衡:

jdbc:mysql:loadbalance://master:3306,slave1:3306,slave2:3306/dbname

When using the following connection string: jdbc:mysql:loadbalance://dbmaster:3306,dbslave1:3306,dbslave2:3306/dbname
connections are load-balanced between all three servers for both read and write connections.

问题:

读写分离时可能会碰到刚写完master,再马上到slave进行查询的情况,而主从复制的时候有延迟,这时怎么解决呢?有两个办法:

1. 比如增加页面保存数据后马上跳转到列表页面,这时可能出不来数据,因为复制还没完成,这时可以在前台添加一些成功的提示,成功页面等进行一些页面跳转延迟处理,让服务器有时间去复制(复制延迟一般在毫秒级,而这种提示处理在秒级,所以时间上一般是足够的)

2. 第1种办法可能部分场景是可行的,但是有些场景要求比较高,需要实时的,这时可以在读取的时候进行处理,强制从master中读取,可以通过注解,加参数/标识等来指定从master读取数据

版权说明

文章标题:  MySQL 读写分离与负载均衡
本文地址:  http://www.ttlsa.com/mysql/read-and-write-separation-and-load-balanced/
除非注明,博客文章均为"运维生存时间"原创,转载请标明本文地址