java实现高性能的数据同步

最近在做一个银行的生产数据脱敏系统,今天写代码时遇到了一个“瓶颈”,脱敏系统需要将生产环境上Infoxmix里的数据原封不动的Copy到另一台 Oracle数据库服务器上,然后对Copy后的数据作些漂白处理。为了将人为干预的因素降到最低,在系统设计时采用Java代码对数据作Copy,思路

    首 先在代码与生产库间建立一个Connection,将读取到的数据放在ResultSet对象,然后再与开发库建立一个Connection。从 ResultSet取出数据后通过TestConnection插入到开发库,以此来实现Copy。代码写完后运行程序,速度太慢了,一秒钟只能Copy 一千条数据,生产库上有上亿条数据,按照这个速度同步完要到猴年马月呀,用PreparedStatement批处理速度也没有提交多少。我想能不能用多 线程处理,多个人干活总比一个人干活速度要快。
    假设生产库有1万条数据,我开5个线程,每个线程分2000条数据,同时向开发库里插数据,Oracle支持高并发这样的话速度至少会提高好多倍,按照这 个思路重新进行了编码,批处理设置为1万条一提交,统计插入数量的变量使用 java.util.concurrent.atomic.AtomicLong,程序一运行,传输速度飞快CPU利用率在70%~90%,现在一秒钟可 以拷贝50万条记录,没过几分钟上亿条数据一条不落地全部Copy到目标库。

在查询的时候我用了如下语句
  1. String queryStr = "SELECT * FROM xx";  
  2. ResultSet coreRs = PreparedStatement.executeQuery(queryStr);  

实习生问如果xx表里有上千万条记录,你全部查询出来放到ResultSet, 那内存不溢出了么?Java在设计的时候已经考虑到这个问题了,并没有查询出所有的数据,而是只查询了一部分数据放到ResultSet,数据“用完”它 会自动查询下一批数据,你可以用setFetchSize(int rows)方法设置一个建议值给ResultSet,告诉它每次从数据库Fetch多少条数据。但我不赞成,因为JDBC驱动会根据实际情况自动调整 Fetch的数量。另外性能也与网线的带宽有直接的关系。
相关代码
package com.dlbank.domain;  
   
 import java.sql.Connection;  
 import java.sql.PreparedStatement;  
 import java.sql.ResultSet;  
 import java.sql.Statement;  
 import java.util.List;  
 import java.util.concurrent.atomic.AtomicLong;  
   
 import org.apache.log4j.Logger;  
   
 /**
  * <p>
  * title: 数据同步类
  * </p>
  * <p>
  * Description: 该类用于将生产核心库数据同步到开发库
  * </p>
  * 
  * @author Tank Zhang
  */  
 public class CoreDataSyncImpl implements CoreDataSync {  
       
     private List<String> coreTBNames; // 要同步的核心库表名
     private ConnectionFactory connectionFactory;  
     private Logger log = Logger.getLogger(getClass());  
       
     private AtomicLong currentSynCount = new AtomicLong(0L); // 当前已同步的条数
       
     private int syncThreadNum;  // 同步的线程数
   
     @Override  
     public void syncData(int businessType) throws Exception {  
           
         for (String tmpTBName : coreTBNames) {  
             log.info("开始同步核心库" + tmpTBName + "表数据");  
             // 获得核心库连接
             Connection coreConnection = connectionFactory.getDMSConnection(4);  
             Statement coreStmt = coreConnection.createStatement();  
             // 为每个线程分配结果集
             ResultSet coreRs = coreStmt.executeQuery("SELECT count(*) FROM "+tmpTBName);  
             coreRs.next();  
             // 总共处理的数量
             long totalNum = coreRs.getLong(1);  
             // 每个线程处理的数量
             long ownerRecordNum =(long) Math.ceil((totalNum / syncThreadNum));   
             log.info("共需要同步的数据量:"+totalNum);  
             log.info("同步线程数量:"+syncThreadNum);  
             log.info("每个线程可处理的数量:"+ownerRecordNum);  
             // 开启五个线程向目标库同步数据
             for(int i=0; i < syncThreadNum; i ++){  
                 StringBuilder sqlBuilder = new StringBuilder();  
                 // 拼装后SQL示例
                 // Select * From dms_core_ds Where id between 1 And 657398
                 // Select * From dms_core_ds Where id between 657399 And
     // 1314796
                 // Select * From dms_core_ds Where id between 1314797 And
     // 1972194
                 // Select * From dms_core_ds Where id between 1972195 And
     // 2629592
                 // Select * From dms_core_ds Where id between 2629593 And
     // 3286990
                 // ..
                 sqlBuilder.append("Select * From ").append(tmpTBName)  
                         .append(" Where id between " ).append(i * ownerRecordNum +1)  
                         .append( " And ")  
                         .append((i * ownerRecordNum + ownerRecordNum));  
                 Thread workThread = new Thread(  
                         new WorkerHandler(sqlBuilder.toString(),businessType,tmpTBName));  
                 workThread.setName("SyncThread-"+i);  
                 workThread.start();  
             }  
             while (currentSynCount.get() < totalNum);  
             // 休眠一会儿让数据库有机会commit剩余的批处理(只针对JUnit单元测试,因为单元测试完成后会关闭虚拟器,使线程里的代码没有机会作提交操作);
             // Thread.sleep(1000 * 3);
             log.info( "核心库"+tmpTBName+"表数据同步完成,共同步了" + currentSynCount.get() + "条数据");  
         }  
     }// end for loop
       
     public void setCoreTBNames(List<String> coreTBNames) {  
         this.coreTBNames = coreTBNames;  
     }  
   
     public void setConnectionFactory(ConnectionFactory connectionFactory) {  
         this.connectionFactory = connectionFactory;  
     }  
       
     public void setSyncThreadNum(int syncThreadNum) {  
         this.syncThreadNum = syncThreadNum;  
     }  
       
     // 数据同步线程
     final class WorkerHandler implements Runnable {  
         ResultSet coreRs;  
         String queryStr;  
         int businessType;  
         String targetTBName;  
         public WorkerHandler(String queryStr,int businessType,String targetTBName) {  
             this.queryStr = queryStr;  
             this.businessType = businessType;  
             this.targetTBName = targetTBName;  
         }  
         @Override  
         public void run() {  
             try {  
                 // 开始同步
                 launchSyncData();  
             } catch(Exception e){  
                 log.error(e);  
                 e.printStackTrace();  
             }  
         }  
         // 同步数据方法
         void launchSyncData() throws Exception{  
             // 获得核心库连接
             Connection coreConnection = connectionFactory.getDMSConnection(4);  
             Statement coreStmt = coreConnection.createStatement();  
             // 获得目标库连接
             Connection targetConn = connectionFactory.getDMSConnection(businessType);  
             targetConn.setAutoCommit(false);// 设置手动提交
             PreparedStatement targetPstmt = targetConn.prepareStatement("INSERT INTO " + targetTBName+" VALUES (?,?,?,?,?)");  
             ResultSet coreRs = coreStmt.executeQuery(queryStr);  
             log.info(Thread.currentThread().getName()+"'s Query SQL::"+queryStr);  
             int batchCounter = 0; // 累加的批处理数量
             while (coreRs.next()) {  
                 targetPstmt.setString(1, coreRs.getString(2));  
                 targetPstmt.setString(2, coreRs.getString(3));  
                 targetPstmt.setString(3, coreRs.getString(4));  
                 targetPstmt.setString(4, coreRs.getString(5));  
                 targetPstmt.setString(5, coreRs.getString(6));  
                 targetPstmt.addBatch();  
                 batchCounter++;  
                 currentSynCount.incrementAndGet();// 递增
                 if (batchCounter % 10000 == 0) { // 1万条数据一提交
                     targetPstmt.executeBatch();  
                     targetPstmt.clearBatch();  
                     targetConn.commit();  
                 }  
             }  
             // 提交剩余的批处理
             targetPstmt.executeBatch();  
             targetPstmt.clearBatch();  
             targetConn.commit();  
             // 释放连接
             connectionFactory.release(targetConn, targetPstmt,coreRs);  
         }  
     }  
 }

相关内容推荐