hadoop reduce 阶段遍历 Iterable 的 2 个“坑”

01 package com.test;
02  
03 import java.util.ArrayList;
04 import java.util.Iterator;
05 import java.util.List;
06  
07 public class T {
08  
09     public static void main(String[] args) {
10  
11         // 只要实现了Iterable接口的对象都可以使用for-each循环。
12         // Iterable接口只由iterator方法构成,
13         // iterator()方法是java.lang.Iterable接口,被Collection继承。
14         /*public interface Iterable<T> {
15             Iterator<T> iterator();
16         }*/
17         Iterable<String> iter = new Iterable<String>() {
18             public Iterator<String> iterator() {
19                 List<String> l = new ArrayList<String>();
20                 l.add("aa");
21                 l.add("bb");
22                 l.add("cc");
23                 return l.iterator();
24             }
25         };
26         for(int count : new int[] {12}){
27             for (String item : iter) {
28                 System.out.println(item);
29             }
30             System.out.println("---------->> " + count + " END.");
31         }
32     }
33 }
结果当然是很正常的完整无误的打印了两遍  Iterable 的值。那究竟是什么原因导致了 reduce 阶段的  Iterable 只能被遍历一次呢?

我们先看一段测试代码:

测试数据:

1 3
2 4
3 50
4 60
5 70
6 8
7 9
01 import java.io.IOException;
02 import java.util.ArrayList;
03 import java.util.List;
04  
05 import org.apache.hadoop.conf.Configuration;
06 import org.apache.hadoop.fs.FileSystem;
07 import org.apache.hadoop.fs.Path;
08 import org.apache.hadoop.io.Text;
09 import org.apache.hadoop.mapreduce.Job;
10 import org.apache.hadoop.mapreduce.Mapper;
11 import org.apache.hadoop.mapreduce.Reducer;
12 import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
13 import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
14 import org.apache.hadoop.util.GenericOptionsParser;
15  
16 public class TestIterable {
17  
18     public static class M1 extends Mapper<Object, Text, Text, Text> {
19         private Text oKey = new Text();
20         private Text oVal = new Text();
21         String[] lineArr;
22  
23         public void map(Object key, Text value, Context context) throws IOException, InterruptedException {
24             lineArr = value.toString().split(" ");
25             oKey.set(lineArr[0]);
26             oVal.set(lineArr[1]);
27             context.write(oKey, oVal);
28         }
29     }
30  
31     public static class R1 extends Reducer<Text, Text, Text, Text> {
32         List<String> valList = new ArrayList<String>();
33         List<Text> textList = new ArrayList<Text>();
34         String strAdd;
35         public void reduce(Text key, Iterable<Text> values, Context context) throws IOException,
36                 InterruptedException {
37             valList.clear();
38             textList.clear();
39             strAdd = "";
40             for (Text val : values) {
41                 valList.add(val.toString());
42                 textList.add(val);
43             }
44              
45             // 坑之 1 :为神马输出的全是最后一个值?why?
46             for(Text text : textList){
47                 strAdd += text.toString() + ", ";
48             }
49             System.out.println(key.toString() + "\t" + strAdd);
50             System.out.println(".......................");
51              
52             // 我这样干呢?对了吗?
53             strAdd = "";
54             for(String val : valList){
55                 strAdd += val + ", ";
56             }
57             System.out.println(key.toString() + "\t" + strAdd);
58             System.out.println("----------------------");
59              
60             // 坑之 2 :第二次遍历的时候为什么得到的都是空?why?
61             valList.clear();
62             strAdd = "";
63             for (Text val : values) {
64                 valList.add(val.toString());
65             }
66             for(String val : valList){
67                 strAdd += val + ", ";
68             }
69             System.out.println(key.toString() + "\t" + strAdd);
70             System.out.println(">>>>>>>>>>>>>>>>>>>>>>");
71         }
72     }
73  
74     public static void main(String[] args) throws Exception {
75         Configuration conf = new Configuration();
76         conf.set("mapred.job.queue.name""regular");
77         String[] otherArgs = new GenericOptionsParser(conf, args).getRemainingArgs();
78         if (otherArgs.length != 2) {
79             System.err.println("Usage: wordcount <in> <out>");
80             System.exit(2);
81         }
82         System.out.println("------------------------");
83         Job job = new Job(conf, "TestIterable");
84         job.setJarByClass(TestIterable.class);
85         job.setMapperClass(M1.class);
86         job.setReducerClass(R1.class);
87         job.setOutputKeyClass(Text.class);
88         job.setOutputValueClass(Text.class);
89         // 输入输出路径
90         FileInputFormat.addInputPath(job, new Path(otherArgs[0]));
91         FileSystem.get(conf).delete(new Path(otherArgs[1]), true);
92         FileOutputFormat.setOutputPath(job, new Path(otherArgs[1]));
93         System.exit(job.waitForCompletion(true) ? 0 1);
94     }
95 }
在 Eclipse 控制台中的结果如下:
01 a   9999,
02 .......................
03 a   34709,
04 ----------------------
05 a  
06 >>>>>>>>>>>>>>>>>>>>>>
07 b   888,
08 .......................
09 b   50608,
10 ----------------------
11 b  
12 >>>>>>>>>>>>>>>>>>>>>>
关于第 1 个坑:对象重用( objects reuse 

reduce方法的javadoc中已经说明了会出现的问题: 

The framework calls this method for each <key, (list of values)> pair in the grouped inputs. Output values must be of the same type as input values. Input keys must not be altered. The framework will reuse the key and value objects that are passed into the reduce, therefore the application should clone the objects they want to keep a copy of.

      也就是说虽然reduce方法会反复执行多次,但key和value相关的对象只有两个,reduce会反复重用这两个对象。所以如果要保存key或者value的结果,只能将其中的值取出另存或者重新clone一个对象(例如Text store = new Text(value) 或者 String a = value.toString()),而不能直接赋引用。因为引用从始至终都是指向同一个对象,你如果直接保存它们,那最后它们都指向最后一个输入记录。会影响最终计算结果而出错。 

看到这里,我想你会恍然大悟:这不是刚毕业找工作,面试官常问的问题:String 是不可变对象但为什么能相加呢?为什么字符串相加不提倡用 String,而用 StringBuilder ?如果你还不清楚这个问题怎么回答,建议你看看这篇深入理解 String, StringBuffer 与 StringBuilder 的区别http://my.oschina.net/leejun2005/blog/102377

关于第 2 个坑:http://stackoverflow.com/questions/6111248/iterate-twice-on-values

The Iterator you receive from that Iterable's iterator() method is special. The values may not all be in memory; Hadoop may be streaming them from disk. They aren't really backed by a Collection, so it's nontrivial to allow multiple iterations.

最后想说明的是:hadoop 框架的作者们真的是考虑很周全,在 hadoop 框架中,不仅有对象重用,还有 JVM 重用等,节约一切可以节约的资源,提高一切可以提高的性能。因为在这种海量数据处理的场景下,性能优化是非常重要的,你可能处理100条数据体现不出性能差别,但是你面对的是千亿、万亿级别的数据呢?

PS:

我的代码是在 Eclipse 中远程调试的,所以 reduce 是没有写 hdfs 的,直接在 eclipse 终端上可以看到结果,很方便,关于怎么在 windows 上远程调试 hadoop,请参考这里 《实战 windows7 下 eclipse 远程调试 linux hadoophttp://my.oschina.net/leejun2005/blog/122775

REF:

hadoop中迭代器的对象重用问题

http://paddy-w.iteye.com/blog/1514595

关于 hadoop 中 JVM 重用和对象重用的介绍

http://wikidoop.com/wiki/Hadoop/MapReduce/Reducer

posted @ 2014-05-18 19:07  baoendemao  阅读(629)  评论(0编辑  收藏  举报