Hadoop Combiner 操作

近期看了一本书：Data-intensive Text Processing with MapReduce，是讲如何设计MR程序的，看到一个例子是Combiner的设计模式，然后就动手实现了下。具体问题如下：

现有输入数据如下：

one	3.9
one	4.0
one	3.8
two	44
two	44
two	44
three	9898
four	2323
four	2323
five	2323
six	23
six	2323
four	232
five	2323

第一列代表用户，第二列代表用户在一个网站上所停留的时间，现在想求每个用户在这个网站的平均停留时间。如果不用combine操作的话，那么其MR伪代码如下（复制书上的内容）：

class Mapper
method Map(string t, integer r)
Emit(string t, integer r)
class Reducer
method Reduce(string t, integers [r1 , r2 , . . .])
sum ← 0
cnt ← 0
for all integer r ∈ integers [r1 , r2 , . . .] do
sum ← sum + r
cnt ← cnt + 1
ravg ← sum/cnt
Emit(string t, integer ravg )

如果要加combine怎么操作呢？Combiner和Reducer一样么(求最大气温的例子或许是一样的，但这里却不是，而且现实中的很多例子都不是一样的)，如果一样的话那么就会变成下面的错误操作了：

Mean(1, 2, 3, 4, 5) = Mean(Mean(1, 2), Mean(3, 4, 5))

正确的伪代码如下（书上摘录）：

class Mapper
method Map(string t, integer r)
Emit(string t, pair (r, 1))
class Combiner
method Combine(string t, pairs [(s1 , c1 ), (s2 , c2 ) . . .])
sum ← 0
cnt ← 0
for all pair (s, c) ∈ pairs [(s1 , c1 ), (s2 , c2 ) . . .] do
sum ← sum + s
cnt ← cnt + c
Emit(string t, pair (sum, cnt))
class Reducer
method Reduce(string t, pairs [(s1 , c1 ), (s2 , c2 ) . . .])
sum ← 0
cnt ← 0
for all pair (s, c) ∈ pairs [(s1 , c1 ), (s2 , c2 ) . . .] do
sum ← sum + s
cnt ← cnt + c
ravg ← sum/cnt
Emit(string t, integer ravg )

由于Combiner的输入和输出格式要一样，即Combiner的输入要和Mapper的输出格式一样，Combiner的输出要和Reducer的输入格式一样。所以上面有pairs。参考上面的伪代码编写的代码如下：

Driver:

package org.fansy.date922;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.KeyValueTextInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.util.GenericOptionsParser;

public class AverageDriver3 {
	public static void main(String[] args) throws Exception{
		// TODO Auto-generated method stub	
		Configuration conf1 = new Configuration();	
	    String[] otherArgs = new GenericOptionsParser(conf1, args).getRemainingArgs();
	    if (otherArgs.length != 2) {
	      System.err.println("Usage: AverageDriver<in> <out>");
	      System.exit(2);
	    }
	    Job job1 = new Job(conf1, "AverageDriver  job ");
	    job1.setInputFormatClass(KeyValueTextInputFormat.class);    
	    job1.setNumReduceTasks(1);
	    job1.setJarByClass(AverageDriver3.class); 
	    job1.setMapperClass(AverageM2.class);
	    job1.setMapOutputKeyClass(Text.class);
		job1.setMapOutputValueClass(TextPair.class);
		job1.setCombinerClass(AverageC3.class);
	    job1.setReducerClass(AverageR2.class);
	    job1.setOutputKeyClass(Text.class);
	    job1.setOutputValueClass(DoubleWritable.class);
	    KeyValueTextInputFormat.addInputPath(job1, new Path(otherArgs[0]));
	    FileOutputFormat.setOutputPath(job1, new Path(otherArgs[1]));    
	    if(!job1.waitForCompletion(true)){
	    	System.exit(1); // run error then exit
	    }  
		System.out.println("************************");
	}
}

Mapper:

package org.fansy.date922;

import java.io.IOException;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;

public class AverageM3 extends Mapper<Text,Text,Text,TextPair>{

//	private Text newkey=new Text();
	private TextPair newvalue=new TextPair();
	private DoubleWritable r=new DoubleWritable();
	private IntWritable number=new IntWritable(1);
	public  void map(Text key,Text value,Context context)throws IOException,InterruptedException {
		// TODO Auto-generated method stub
		System.out.println(key.toString());
		double shuzhi=Double.parseDouble(value.toString());
		r.set(shuzhi);
		newvalue.set(r, number);
		context.write(key, newvalue);
	}
}

Combiner:

package org.fansy.date922;

import java.io.IOException;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;

public class AverageC3 extends Reducer<Text,TextPair,Text,TextPair>{
	private DoubleWritable newvalued=new DoubleWritable();
	private IntWritable newvaluei=new IntWritable();
	private TextPair newvalue=new TextPair();
	public  void reduce(Text key,Iterable<TextPair> values,Context context) throws IOException,InterruptedException{
		// TODO Auto-generated method stub
		double sum= 0.0;
		int num=0;
		for(TextPair val:values){
			sum+=val.getFirst().get();
			num+=val.getSecond().get();
		}
		newvalued.set(sum);
		newvaluei.set(num);
		newvalue.set(newvalued,newvaluei);
		context.write(key, newvalue);
	}
}

Reducer:

package org.fansy.date922;

import java.io.IOException;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;

public class AverageR3 extends Reducer<Text,TextPair,Text,DoubleWritable>{
	private DoubleWritable newvalue=new DoubleWritable();
	public  void reduce(Text key,Iterable<TextPair> values,Context context) throws IOException,InterruptedException{
		// TODO Auto-generated method stub
		double sum= 0.0;
		int num=0;
		for(TextPair val:values){
			sum+=val.getFirst().get();
			num+=val.getSecond().get();
		}
		double aver=sum/num;
		newvalue.set(aver);
		context.write(key, newvalue);
	}
}

TextPair:

package org.fansy.date922;

import java.io.DataInput;
import java.io.DataOutput;
import java.io.IOException;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.WritableComparable;

public class TextPair implements WritableComparable<TextPair> {
	private DoubleWritable first;
	private IntWritable second;
	public TextPair(){
		set(new DoubleWritable(),new IntWritable());
	}
	public  void set(DoubleWritable longWritable, IntWritable intWritable) {
		// TODO Auto-generated method stub
		this.first=longWritable;
		this.second=intWritable;
	}
	public DoubleWritable getFirst(){
		return first;
	}
	public IntWritable getSecond(){
		return second;
	}
	@Override
	public void readFields(DataInput arg0) throws IOException {
		// TODO Auto-generated method stub
		first.readFields(arg0);
		second.readFields(arg0);
	}
	@Override
	public void write(DataOutput arg0) throws IOException {
		// TODO Auto-generated method stub
		first.write(arg0);
		second.write(arg0);
	}
	@Override
	public int compareTo(TextPair o) {
		// TODO Auto-generated method stub
		int cmp=first.compareTo(o.first);
		if(cmp!=0){
			return cmp;
		}
		return second.compareTo(o.second);
	}	
}

查看终端中的显示也可以看出的确是有combine操作的：

12/09/22 15:55:45 INFO mapred.JobClient: Job complete: job_local_0001
12/09/22 15:55:45 INFO mapred.JobClient: Counters: 22
12/09/22 15:55:45 INFO mapred.JobClient:   File Output Format Counters 
12/09/22 15:55:45 INFO mapred.JobClient:     Bytes Written=65
12/09/22 15:55:45 INFO mapred.JobClient:   FileSystemCounters
12/09/22 15:55:45 INFO mapred.JobClient:     FILE_BYTES_READ=466
12/09/22 15:55:45 INFO mapred.JobClient:     HDFS_BYTES_READ=244
12/09/22 15:55:45 INFO mapred.JobClient:     FILE_BYTES_WRITTEN=82758
12/09/22 15:55:45 INFO mapred.JobClient:     HDFS_BYTES_WRITTEN=65
12/09/22 15:55:45 INFO mapred.JobClient:   File Input Format Counters 
12/09/22 15:55:45 INFO mapred.JobClient:     Bytes Read=122
12/09/22 15:55:45 INFO mapred.JobClient:   Map-Reduce Framework
12/09/22 15:55:45 INFO mapred.JobClient:     Map output materialized bytes=118
12/09/22 15:55:45 INFO mapred.JobClient:     Map input records=14
12/09/22 15:55:45 INFO mapred.JobClient:     Reduce shuffle bytes=0
12/09/22 15:55:45 INFO mapred.JobClient:     Spilled Records=12
12/09/22 15:55:45 INFO mapred.JobClient:     Map output bytes=231
12/09/22 15:55:45 INFO mapred.JobClient:     Total committed heap usage (bytes)=301727744
12/09/22 15:55:45 INFO mapred.JobClient:     CPU time spent (ms)=0
12/09/22 15:55:45 INFO mapred.JobClient:     SPLIT_RAW_BYTES=108
12/09/22 15:55:45 INFO mapred.JobClient:     Combine input records=14
12/09/22 15:55:45 INFO mapred.JobClient:     Reduce input records=6
12/09/22 15:55:45 INFO mapred.JobClient:     Reduce input groups=6
12/09/22 15:55:45 INFO mapred.JobClient:     Combine output records=6
12/09/22 15:55:45 INFO mapred.JobClient:     Physical memory (bytes) snapshot=0
12/09/22 15:55:45 INFO mapred.JobClient:     Reduce output records=6
12/09/22 15:55:45 INFO mapred.JobClient:     Virtual memory (bytes) snapshot=0
12/09/22 15:55:45 INFO mapred.JobClient:     Map output records=14
************************

那本书上面其实最后还有提到一个 in-Mapper Combining的一个编程，但是看的不是很明白，伪代码如下：

class Mapper
method Initialize
S ← new AssociativeArray
C ← new AssociativeArray
method Map(string t, integer r)
S{t} ← S{t} + r
C{t} ← C{t} + 1
method Close
for all term t ∈ S do
Emit(term t, pair (S{t}, C{t}))

继续学习 MR编程中。。