请教spark通过jdbc读取大量数据

发布网友发布时间：2022-05-01 09:04

共2个回答

懂视网时间：2022-05-01 13:26

@Test 2 public void test() throws IOException { 3 BufferedReader reader=new BufferedReader(new InputStreamReader(new FileInputStream("C:\Users\yhzh\Desktop\zh_20160913"))); 4 String tmp=null; 5 List<String> nos=new ArrayList<String>(); 6 while((tmp=reader.readLine()) !=null) 7 nos.add(tmp); 8 9 Connection con = null;// 创建一个数据库连接 10 PreparedStatement pre = null;// 创建预编译语句对象，一般都是用这个而不用Statement 11 PreparedStatement pre2 = null; 12 ResultSet result = null;// 创建一个结果集对象 13 BufferedWriter csvWriter=null; 14 try 15 { 16 String tag=(new SimpleDateFormat("hhmmss")).format(new Date()); 17 csvWriter = new BufferedWriter(new OutputStreamWriter(new FileOutputStream(new File("C:\Users\yhzh\Desktop\贷后还款计划_"+tag+".csv")), "utf-8")); 18 19 Class.forName("oracle.jdbc.driver.OracleDriver"); 20 //:6006/hotfix 21 String url = "jdbc:oracle:thin:@//*.*.*.*:16030/zcgl",user = "*",password = "*"; 22 con = DriverManager.getConnection(url, user, password);// 获取连接 23 String sql="select max(lr_id) from t_loan_request where LR_REQUESTSTATUS =‘2‘ and lr_applyid=?"; 24 /*String sql = 27 "select req.LR_APPLYID,CURR_PERIODS,REPAY_DAY, " + 28 "(NEEDREPAY_PRINCIPAL+NEEDREPAY_INTEREST+NEEDREPAY_ADMIFEE+nvl(RISK_AMOUNT,0)+nvl(SERVICEFEE,0)+nvl(PARKINGFEE,0))NEEDREPAY_TOTAL, " + 29 "PERIOD_REPAY_AMOUNT " + 30 "from( " + 31 " select lr_id,LR_APPLYID from t_loan_request t " + 32 " where t.LR_REQUESTSTATUS =‘2‘ and t.creater=‘PostLoanOuterAction‘ " + 33 " order by lr_id desc)req " + 34 "left join t_repay_plan rp " + 35 "on req.lr_id=rp.lr_id " + 36 "order by req.lr_id,CURR_PERIODS ";// 预编译语句，“？”代表参数*/ 37 pre = con.prepareStatement(sql); 38 pre2=con.prepareStatement("select CURR_PERIODS,REPAY_DAY, " + 39 "(NEEDREPAY_PRINCIPAL+NEEDREPAY_INTEREST+NEEDREPAY_ADMIFEE+nvl(RISK_AMOUNT,0)+nvl(SERVICEFEE,0)+nvl(PARKINGFEE,0))NEEDREPAY_TOTAL, " + 40 "PERIOD_REPAY_AMOUNT " + 41 "from t_repay_plan " + 42 "where lr_id=? " + 43 "order by CURR_PERIODS"); 44 for(String no:nos){ 45 pre.setString(1,no); 46 result = pre.executeQuery(); 47 if(result.next()) { 48 long lrId=result.getLong(1); 49 pre2.setLong(1,lrId); 50 result = pre2.executeQuery(); 51 while (result.next()) { 52 csvWriter.write(no); 53 csvWriter.write(","); 54 csvWriter.write(result.getString(1)); 55 csvWriter.write(","); 56 csvWriter.write(result.getString(2)); 57 csvWriter.write(","); 58 csvWriter.write(result.getString(3)); 59 csvWriter.write(","); 60 csvWriter.write(result.getString(4)); 61 csvWriter.newLine(); 62 } 63 } 64 } 65 66 csvWriter.flush(); 67 } 68 catch (Exception e) 69 { 70 e.printStackTrace(); 71 } 72 finally 73 { 74 try 75 { 76 if(csvWriter !=null) 77 csvWriter.close(); 78 if (result != null) 79 result.close(); 80 if (pre != null) 81 pre.close(); 82 if (con != null) 83 con.close(); 84 System.out.println("数据库连接已关闭！"); 85 } 86 catch (Exception e) 87 { 88 e.printStackTrace(); 89 } 90 } 91 }

先读取所有编号形成List，后遍历这个List，先查出id再查详细数据。这样的数据csv文件中大约8万多条

一条条的来肯定慢，如果不按照编号，直接一次查出，数据是9万多条。速度都很慢！！！

后来想提高下，至少要有个明显的提升呀。写文件这块基本排除了，剩下的疑问就是ResultSet是否拿到了所有结果呢？

根据网上查到的资料和实际调试，得出答案：ResultSet默认一次取10条数据，怪不得要慢，如果一次全部读入内存再写入文件就一定很快了。

怎样一次读取所有数据呢？

sql改为读取全部

 1 String sql="select req.LR_APPLYID,CURR_PERIODS,REPAY_DAY,
" +
 2 "(NEEDREPAY_PRINCIPAL+NEEDREPAY_INTEREST+NEEDREPAY_ADMIFEE+nvl(RISK_AMOUNT,0)+nvl(SERVICEFEE,0)+nvl(PARKINGFEE,0))NEEDREPAY_TOTAL,
" +
 3 "PERIOD_REPAY_AMOUNT 
" +
 4 "from(
" +
 5 " select lr_id,LR_APPLYID from t_loan_request t 
" +
 6 " where t.LR_REQUESTSTATUS =‘2‘ and t.creater=‘PostLoanOuterAction‘
" +
 7 " order by lr_id desc)req
" +
 8 "left join t_repay_plan rp
" +
 9 "on req.lr_id=rp.lr_id
" +
10 "order by req.lr_id,CURR_PERIODS";

设置PreparedStatement：

1 pre = con.prepareStatement(sql);
2 pre.setFetchSize(100000);
3 result = pre.executeQuery();
4 //result.setFetchSize(100000);

主要是PreparedStatement的 setFetchSize 方法，

后来发现ResultSet也有个setFetchSize 方法，也是可行的，只是这个时候resultset中已经有了10条记录直到循环10次后，再次使用result.next()才去取100000，fetchSize才起作用

这样设置后速度飞快！！！

jdbc--取大量数据

标签：

热心网友时间：2022-05-01 10:34

Hadoop（大数据分析领域无可争辩的王者）专注于批处理。这种模型对许多情形（比如为网页建立索引）已经足够，但还存在其他一些使用模型，它们需要来自高度动态的来源的实时信息。