JDBC批量Insert深度優化(有事務)

JDBC批量Insert深度優化(有事務)
 
環境:
MySQL 5.1
RedHat Linux AS 5
JavaSE 1.5
DbConnectionBroker 微型數據庫鏈接池
 
測試的方案:
執行10萬次Insert語句,使用不一樣方式。
 
A組:靜態SQL,自動提交,沒事務控制(MyISAM引擎)
一、逐條執行10萬次
二、分批執行將10萬分紅m批,每批n條,分多種分批方案來執行。
 
B組:預編譯模式SQL,自動提交,沒事務控制(MyISAM引擎)
一、逐條執行10萬次
二、分批執行將10萬分紅m批,每批n條,分多種分批方案來執行。
-------------------------------------------------------------------------------------------
C組:靜態SQL,不自動提交,有事務控制(InnoDB引擎)
一、逐條執行10萬次
二、分批執行將10萬分紅m批,每批n條,分多種分批方案來執行。
 
D組:預編譯模式SQL,不自動提交,有事務控制(InnoDB引擎)
一、逐條執行10萬次
二、分批執行將10萬分紅m批,每批n條,分多種分批方案來執行。
 
本次主要測試C、D組,並得出測試結果。
 
SQL代碼
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組測試代碼:
package testbatch;

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

/**
* JDBC批量Insert優化(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();
                }
        }

        /**
         * 初始化測試環境
         *
         * @throws SQLException 異常時拋出
         */

        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條靜態SQL插入
         *
         * @throws Exception 異常時拋出
         */

        public static void testInsert() throws Exception {
                init();         //初始化環境
                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("單條執行100000條Insert操做,共耗時:" + (end - start) / 1000f + "秒!");
        }

        /**
         * 批處理執行靜態SQL測試
         *
         * @param m 批次
         * @param n 每批數量
         * @throws Exception 異常時拋出
         */

        public static void testInsertBatch(int m, int n) throws Exception {
                init();             //初始化環境
                Long start = System.currentTimeMillis();
                for (int i = 0; i < m; i++) {
                        //從池中獲取鏈接
                        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" +
                                                ")";
                                //加入批處理
                                stmt.addBatch(sql);
                        }
                        stmt.executeBatch();    //執行批處理
                        conn.commit();
//                        stmt.clearBatch();        //清理批處理
                        stmt.close();
                        myBroker.freeConnection(conn); //鏈接歸池
                }
                Long end = System.currentTimeMillis();
                System.out.println("批量執行" + m + "*" + n + "=" + m * n + "條Insert操做,共耗時:" + (end - start) / 1000f + "秒!");
        }

        /**
         * 100000條預約義SQL插入
         *
         * @throws Exception 異常時拋出
         */

        public static void testInsert2() throws Exception {     //單條執行100000條Insert操做,共耗時:40.422秒!
                init();         //初始化環境
                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("單條執行100000條Insert操做,共耗時:" + (end - start) / 1000f + "秒!");
        }

        /**
         * 批處理執行預處理SQL測試
         *
         * @param m 批次
         * @param n 每批數量
         * @throws Exception 異常時拋出
         */

        public static void testInsertBatch2(int m, int n) throws Exception {
                init();             //初始化環境
                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++) {
                        //從池中獲取鏈接
                        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()));
                                //加入批處理
                                pstmt.addBatch();
                        }
                        pstmt.executeBatch();    //執行批處理
                        conn.commit();
//                        pstmt.clearBatch();        //清理批處理
                        pstmt.close();
                        myBroker.freeConnection(conn); //鏈接歸池
                }
                Long end = System.currentTimeMillis();
                System.out.println("批量執行" + m + "*" + n + "=" + m * n + "條Insert操做,共耗時:" + (end - start) / 1000f + "秒!");
        }

        public static void main(String[] args) throws Exception {
                init();
                Long start = System.currentTimeMillis();
                System.out.println("--------C組測試----------");
                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組測試過程結束,所有測試耗時:" + (end1 - start) / 1000f + "秒!");

                System.out.println("--------D組測試----------");
                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組測試過程結束,所有測試耗時:" + (end2 - end1) / 1000f + "秒!");
        }
}
 
執行結果:
--------C組測試----------
單條執行100000條Insert操做,共耗時:103.656秒!
批量執行100*1000=100000條Insert操做,共耗時:31.328秒!
批量執行250*400=100000條Insert操做,共耗時:31.406秒!
批量執行400*250=100000條Insert操做,共耗時:31.75秒!
批量執行500*200=100000條Insert操做,共耗時:31.438秒!
批量執行1000*100=100000條Insert操做,共耗時:31.968秒!
批量執行2000*50=100000條Insert操做,共耗時:32.938秒!
批量執行2500*40=100000條Insert操做,共耗時:33.141秒!
批量執行5000*20=100000條Insert操做,共耗時:35.265秒!
C組測試過程結束,所有測試耗時:363.656秒!
--------D組測試----------
單條執行100000條Insert操做,共耗時:107.61秒!
批量執行100*1000=100000條Insert操做,共耗時:32.64秒!
批量執行250*400=100000條Insert操做,共耗時:32.641秒!
批量執行400*250=100000條Insert操做,共耗時:33.109秒!
批量執行500*200=100000條Insert操做,共耗時:32.859秒!
批量執行1000*100=100000條Insert操做,共耗時:33.547秒!
批量執行2000*50=100000條Insert操做,共耗時:34.312秒!
批量執行2500*40=100000條Insert操做,共耗時:34.672秒!
批量執行5000*20=100000條Insert操做,共耗時:36.672秒!
D組測試過程結束,所有測試耗時:378.922秒!
 
 
測試結果意想不到吧,最短期居然超過上篇。觀察整個測試結果,發現總時間很長,緣由是逐條執行的效率過低了。
 
結論:
 
在本測試條件下,得出結論:
 
數據庫鏈接池控制下,不自動提交,事務控制(InnoDB引擎)
 
一、逐條執行的效率很低很低,儘量避免逐條執行。
二、事務控制下,靜態SQL的效率超過預處理SQL。
三、分批的大小對效率影響挺大的,通常來講,事務控制下,分批大小在100-1000之間比較合適。
四、談到優化方式,上面的批處理就是很好的優化策略。
 
 
大總結:
 
對比上篇沒事務的測試結果,得出一個全面的結論:
 
一、鏈接池最基本的也是最重要的優化策略,總能大幅提升性能。
 
二、批處理在效率上老是比逐條處理有優點,要處理的數據的記錄條數越大,批處理的優點越明顯,批處理還有一個好處就是減小了對數據庫的連接次數,從而減輕數據庫的壓力。
 
三、批處理執行SQL的時候,批處理的分批的大小與數據庫的吞吐量以及硬件配置有很大關係,須要經過測試找到最佳的分批大小,通常在50-1000之間。
 
四、預處理SQL在沒事務的表上效率較高,在有實物的狀況下比靜態SQL稍有不及。但預約義SQL還有個好處就是消耗的內存較少,靜態SQL串會佔用大量的內存資源,容易致使內存溢出的問題。所以批量執行時候能夠優先選擇預約義SQL。
 
五、在批處理執行的時候,每批執行完成後,最好顯式的調用pstmt.close()或stmt.close()方法,以便儘快釋放執行過的SQL語句,提升內存利用率。
 
六、對於有大量SELECT操做,MyISAM是更好的選擇;對於有大量INSERT和UPDATE操做的表,InnoDB效率更好。
 
七、雖然測試結果只能反映特定狀況下的一些事實,以上的優化策略是廣泛策略,能夠明顯縮短尋找最優策略的時間,對於效率要求很高的程序,還應該作併發性等測試。
 
八、測試是件很辛苦的事情,你須要有大量的事實來證實你的優化是有效的,而不能單單憑經驗,由於每一個機器的環境都不同,使用的方式也不一樣。
相關文章
相關標籤/搜索