JDBC批量Insert深度优化(有事务)

JDBC批量Insert深度优化(有事务)
 
环境:
MySQL 5.1
RedHat Linux AS 5
JavaSE 1.5
DbConnectionBroker 微型数据库连接池
 
测试的方案:
执行10万次Insert语句,使用不同方式。
 
A组:静态SQL,自动提交,没事务控制(MyISAM引擎)
1、逐条执行10万次
2、分批执行将10万分成m批,每批n条,分多种分批方案来执行。
 
B组:预编译模式SQL,自动提交,没事务控制(MyISAM引擎)
1、逐条执行10万次
2、分批执行将10万分成m批,每批n条,分多种分批方案来执行。
-------------------------------------------------------------------------------------------
C组:静态SQL,不自动提交,有事务控制(InnoDB引擎)
1、逐条执行10万次
2、分批执行将10万分成m批,每批n条,分多种分批方案来执行。
 
D组:预编译模式SQL,不自动提交,有事务控制(InnoDB引擎)
1、逐条执行10万次
2、分批执行将10万分成m批,每批n条,分多种分批方案来执行。
 
本次主要测试C、D组,并得出测试结果。
 
SQL代码
DROP TABLE IF EXISTS tuser;

CREATE TABLE tuser (
    id bigint(20) NOT NULL AUTO_INCREMENT,
     name varchar(12) DEFAULT NULL,
    remark varchar(24) DEFAULT NULL,
    createtime datetime DEFAULT NULL,
    updatetime datetime DEFAULT NULL,
     PRIMARY KEY (id)
) ENGINE=InnoDB AUTO_INCREMENT=1 DEFAULT CHARSET=utf8;
 
C、D组测试代码:
package testbatch;

import java.io.IOException;
import java.sql.*;

/**
* JDBC批量Insert优化(下)
*
* @author leizhimin 2009-7-29 10:03:10
*/

public class TestBatch {
         public static DbConnectionBroker myBroker = null;

         static {
                 try {
                        myBroker = new DbConnectionBroker( "com.mysql.jdbc.Driver",
                                         "jdbc:mysql://192.168.104.163:3306/testdb",
                                        "vcom", "vcom", 2, 4,
                                        "c:\\testdb.log", 0.01);
                } catch (IOException e) {
                        e.printStackTrace();
                }
        }

        /**
         * 初始化测试环境
         *
         * @throws SQLException 异常时抛出
         */

        public static void init() throws SQLException {
                Connection conn = myBroker.getConnection();
                conn.setAutoCommit(false);
                Statement stmt = conn.createStatement();
                stmt.addBatch("DROP TABLE IF EXISTS tuser");
                stmt.addBatch("CREATE TABLE tuser (\n" +
                                "    id bigint(20) NOT NULL AUTO_INCREMENT,\n" +
                                "    name varchar(12) DEFAULT NULL,\n" +
                                "    remark varchar(24) DEFAULT NULL,\n" +
                                "    createtime datetime DEFAULT NULL,\n" +
                                "    updatetime datetime DEFAULT NULL,\n" +
                                "    PRIMARY KEY (id)\n" +
                                ") ENGINE=InnoDB AUTO_INCREMENT=1 DEFAULT CHARSET=utf8");
                stmt.executeBatch();
                conn.commit();
                myBroker.freeConnection(conn);
        }

        /**
         * 100000条静态SQL插入
         *
         * @throws Exception 异常时抛出
         */

        public static void testInsert() throws Exception {
                init();         //初始化环境
                Long start = System.currentTimeMillis();
                for (int i = 0; i < 100000; i++) {
                        String sql = "\n" +
                                        "insert into testdb.tuser \n" +
                                        "\t(name, \n" +
                                        "\tremark, \n" +
                                        "\tcreatetime, \n" +
                                        "\tupdatetime\n" +
                                        "\t)\n" +
                                        "\tvalues\n" +
                                        "\t('" + RandomToolkit.generateString(12) + "', \n" +
                                        "\t'" + RandomToolkit.generateString(24) + "', \n" +
                                        "\tnow(), \n" +
                                        "\tnow()\n" +
                                        ")";
                        Connection conn = myBroker.getConnection();
                        conn.setAutoCommit(false);
                        Statement stmt = conn.createStatement();
                        stmt.execute(sql);
                        conn.commit();
                        myBroker.freeConnection(conn);
                }
                Long end = System.currentTimeMillis();
                System.out.println("单条执行100000条Insert操作,共耗时:" + (end - start) / 1000f + "秒!");
        }

        /**
         * 批处理执行静态SQL测试
         *
         * @param m 批次
         * @param n 每批数量
         * @throws Exception 异常时抛出
         */

        public static void testInsertBatch(int m, int n) throws Exception {
                init();             //初始化环境
                Long start = System.currentTimeMillis();
                for (int i = 0; i < m; i++) {
                        //从池中获取连接
                        Connection conn = myBroker.getConnection();
                        conn.setAutoCommit(false);
                        Statement stmt = conn.createStatement();
                        for (int k = 0; k < n; k++) {
                                String sql = "\n" +
                                                "insert into testdb.tuser \n" +
                                                "\t(name, \n" +
                                                "\tremark, \n" +
                                                "\tcreatetime, \n" +
                                                "\tupdatetime\n" +
                                                "\t)\n" +
                                                "\tvalues\n" +
                                                "\t('" + RandomToolkit.generateString(12) + "', \n" +
                                                "\t'" + RandomToolkit.generateString(24) + "', \n" +
                                                "\tnow(), \n" +
                                                "\tnow()\n" +
                                                ")";
                                //加入批处理
                                stmt.addBatch(sql);
                        }
                        stmt.executeBatch();    //执行批处理
                        conn.commit();
//                        stmt.clearBatch();        //清理批处理
                        stmt.close();
                        myBroker.freeConnection(conn); //连接归池
                }
                Long end = System.currentTimeMillis();
                System.out.println("批量执行" + m + "*" + n + "=" + m * n + "条Insert操作,共耗时:" + (end - start) / 1000f + "秒!");
        }

        /**
         * 100000条预定义SQL插入
         *
         * @throws Exception 异常时抛出
         */

        public static void testInsert2() throws Exception {     //单条执行100000条Insert操作,共耗时:40.422秒!
                init();         //初始化环境
                Long start = System.currentTimeMillis();
                String sql = "" +
                                "insert into testdb.tuser\n" +
                                "    (name, remark, createtime, updatetime)\n" +
                                "values\n" +
                                "    (?, ?, ?, ?)";
                for (int i = 0; i < 100000; i++) {
                        Connection conn = myBroker.getConnection();
                        conn.setAutoCommit(false);
                        PreparedStatement pstmt = conn.prepareStatement(sql);
                        pstmt.setString(1, RandomToolkit.generateString(12));
                        pstmt.setString(2, RandomToolkit.generateString(24));
                        pstmt.setDate(3, new Date(System.currentTimeMillis()));
                        pstmt.setDate(4, new Date(System.currentTimeMillis()));
                        pstmt.executeUpdate();
                        conn.commit();
                        pstmt.close();
                        myBroker.freeConnection(conn);
                }
                Long end = System.currentTimeMillis();
                System.out.println("单条执行100000条Insert操作,共耗时:" + (end - start) / 1000f + "秒!");
        }

        /**
         * 批处理执行预处理SQL测试
         *
         * @param m 批次
         * @param n 每批数量
         * @throws Exception 异常时抛出
         */

        public static void testInsertBatch2(int m, int n) throws Exception {
                init();             //初始化环境
                Long start = System.currentTimeMillis();
                String sql = "" +
                                "insert into testdb.tuser\n" +
                                "    (name, remark, createtime, updatetime)\n" +
                                "values\n" +
                                "    (?, ?, ?, ?)";
                for (int i = 0; i < m; i++) {
                        //从池中获取连接
                        Connection conn = myBroker.getConnection();
                        conn.setAutoCommit(false);
                        PreparedStatement pstmt = conn.prepareStatement(sql);
                        for (int k = 0; k < n; k++) {
                                pstmt.setString(1, RandomToolkit.generateString(12));
                                pstmt.setString(2, RandomToolkit.generateString(24));
                                pstmt.setDate(3, new Date(System.currentTimeMillis()));
                                pstmt.setDate(4, new Date(System.currentTimeMillis()));
                                //加入批处理
                                pstmt.addBatch();
                        }
                        pstmt.executeBatch();    //执行批处理
                        conn.commit();
//                        pstmt.clearBatch();        //清理批处理
                        pstmt.close();
                        myBroker.freeConnection(conn); //连接归池
                }
                Long end = System.currentTimeMillis();
                System.out.println("批量执行" + m + "*" + n + "=" + m * n + "条Insert操作,共耗时:" + (end - start) / 1000f + "秒!");
        }

        public static void main(String[] args) throws Exception {
                init();
                Long start = System.currentTimeMillis();
                System.out.println("--------C组测试----------");
                testInsert();
                testInsertBatch(100, 1000);
                testInsertBatch(250, 400);
                testInsertBatch(400, 250);
                testInsertBatch(500, 200);
                testInsertBatch(1000, 100);
                testInsertBatch(2000, 50);
                testInsertBatch(2500, 40);
                testInsertBatch(5000, 20);
                Long end1 = System.currentTimeMillis();
                System.out.println("C组测试过程结束,全部测试耗时:" + (end1 - start) / 1000f + "秒!");

                System.out.println("--------D组测试----------");
                testInsert2();
                testInsertBatch2(100, 1000);
                testInsertBatch2(250, 400);
                testInsertBatch2(400, 250);
                testInsertBatch2(500, 200);
                testInsertBatch2(1000, 100);
                testInsertBatch2(2000, 50);
                testInsertBatch2(2500, 40);
                testInsertBatch2(5000, 20);

                Long end2 = System.currentTimeMillis();
                System.out.println("D组测试过程结束,全部测试耗时:" + (end2 - end1) / 1000f + "秒!");
        }
}
 
执行结果:
--------C组测试----------
单条执行100000条Insert操作,共耗时:103.656秒!
批量执行100*1000=100000条Insert操作,共耗时:31.328秒!
批量执行250*400=100000条Insert操作,共耗时:31.406秒!
批量执行400*250=100000条Insert操作,共耗时:31.75秒!
批量执行500*200=100000条Insert操作,共耗时:31.438秒!
批量执行1000*100=100000条Insert操作,共耗时:31.968秒!
批量执行2000*50=100000条Insert操作,共耗时:32.938秒!
批量执行2500*40=100000条Insert操作,共耗时:33.141秒!
批量执行5000*20=100000条Insert操作,共耗时:35.265秒!
C组测试过程结束,全部测试耗时:363.656秒!
--------D组测试----------
单条执行100000条Insert操作,共耗时:107.61秒!
批量执行100*1000=100000条Insert操作,共耗时:32.64秒!
批量执行250*400=100000条Insert操作,共耗时:32.641秒!
批量执行400*250=100000条Insert操作,共耗时:33.109秒!
批量执行500*200=100000条Insert操作,共耗时:32.859秒!
批量执行1000*100=100000条Insert操作,共耗时:33.547秒!
批量执行2000*50=100000条Insert操作,共耗时:34.312秒!
批量执行2500*40=100000条Insert操作,共耗时:34.672秒!
批量执行5000*20=100000条Insert操作,共耗时:36.672秒!
D组测试过程结束,全部测试耗时:378.922秒!
 
 
测试结果意想不到吧,最短时间竟然超过上篇。观察整个测试结果,发现总时间很长,原因是逐条执行的效率太低了。
 
结论:
 
在本测试条件下,得出结论:
 
数据库连接池控制下,不自动提交,事务控制(InnoDB引擎)
 
1、逐条执行的效率很低很低,尽可能避免逐条执行。
2、事务控制下,静态SQL的效率超过预处理SQL。
3、分批的大小对效率影响挺大的,一般来说,事务控制下,分批大小在100-1000之间比较合适。
4、谈到优化方式,上面的批处理就是很好的优化策略。
 
 
大总结:
 
对比上篇没事务的测试结果,得出一个全面的结论:
 
1、连接池最基本的也是最重要的优化策略,总能大幅提高性能。
 
2、批处理在效率上总是比逐条处理有优势,要处理的数据的记录条数越大,批处理的优势越明显,批处理还有一个好处就是减少了对数据库的链接次数,从而减轻数据库的压力。
 
3、批处理执行SQL的时候,批处理的分批的大小与数据库的吞吐量以及硬件配置有很大关系,需要通过测试找到最佳的分批大小,一般在50-1000之间。
 
4、预处理SQL在没事务的表上效率较高,在有实物的情况下比静态SQL稍有不及。但预定义SQL还有个好处就是消耗的内存较少,静态SQL串会占用大量的内存资源,容易导致内存溢出的问题。因此批量执行时候可以优先选择预定义SQL。
 
5、在批处理执行的时候,每批执行完成后,最好显式的调用pstmt.close()或stmt.close()方法,以便尽快释放执行过的SQL语句,提高内存利用率。
 
6、对于有大量SELECT操作,MyISAM是更好的选择;对于有大量INSERT和UPDATE操作的表,InnoDB效率更好。
 
7、虽然测试结果只能反映特定情况下的一些事实,以上的优化策略是普遍策略,可以明显缩短寻找最优策略的时间,对于效率要求很高的程序,还应该做并发性等测试。
 
8、测试是件很辛苦的事情,你需要有大量的事实来证明你的优化是有效的,而不能单单凭经验,因为每个机器的环境都不一样,使用的方式也不同。
 

你可能感兴趣的:(JDBC批量Insert深度优化(有事务))