hadoop 里执行 MapReduce 任务的几种常见方式

说明：

测试文件：

echo -e "aa\tbb \tcc\nbb\tcc\tdd" > 3.txt

hadoop fs -put 3.txt /tmp/3.txt

全文的例子均以该文件做测试用例，统计单词出现的次数（WordCount）。

1、原生态的方式：java 源码编译打包成jar包后，由 hadoop 脚本调度执行，举例：

import java.io.IOException;
import java.util.StringTokenizer;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.util.GenericOptionsParser;

public class WordCount {

	public static class TokenizerMapper extends Mapper<Object, Text, Text, IntWritable> {
		/**
		 * LongWritable, IntWritable, Text 均是 Hadoop 中实现的用于封装 Java 数据类型的类，
		 * 这些类实现了WritableComparable接口， 都能够被串行化从而便于在分布式环境中进行数据交换，
		 * 你可以将它们分别视为long,int,String 的替代品。
		 */
		// IntWritable one 相当于 java 原生类型 int 1
		private final static IntWritable one = new IntWritable(1);
		private Text word = new Text();

		public void map(Object key, Text value, Context context) throws IOException, InterruptedException {
			// 每行记录都会调用 map 方法处理，此处是每行都被分词
			StringTokenizer itr = new StringTokenizer(value.toString());
			while (itr.hasMoreTokens()) {
				word.set(itr.nextToken());
				// 输出每个词及其出现的次数 1，类似 <word1,1><word2,1><word1,1>
				context.write(word, one);
			}
		}
	}

	public static class IntSumReducer extends Reducer<Text, IntWritable, Text, IntWritable> {
		private IntWritable result = new IntWritable();

		public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException,
				InterruptedException {
			// key 相同的键值对会被分发到同一个 reduce中处理
			// 例如 <word1,<1,1>>在 reduce1 中处理，而<word2,<1>> 会在 reduce2 中处理
			int sum = 0;
			// 相同的key（单词）的出现次数会被 sum 累加
			for (IntWritable val : values) {
				sum += val.get();
			}
			result.set(sum);
			// 1个 reduce 处理完1 个键值对后，会输出其 key（单词）对应的结果（出现次数）
			context.write(key, result);
		}
	}

	public static void main(String[] args) throws Exception {
		Configuration conf = new Configuration();
		// 多队列hadoop集群中，设置使用的队列
		conf.set("mapred.job.queue.name", "regular");
		// 之所以此处不直接用 argv[1] 这样的，是为了排除掉运行时的集群属性参数，例如队列参数，
		// 得到用户输入的纯参数，如路径信息等
		String[] otherArgs = new GenericOptionsParser(conf, args).getRemainingArgs();
		for (String argsStr : otherArgs) {
			System.out.println("-->> " + argsStr);
		}
		if (otherArgs.length < 2) {
			System.err.println("Usage: wordcount <in> <out>");
			System.exit(2);
		}
		Job job = new Job(conf, "word count");
		job.setJarByClass(WordCount.class);
		// map、reduce 输入输出类
		job.setMapperClass(TokenizerMapper.class);
		job.setCombinerClass(IntSumReducer.class);
		job.setReducerClass(IntSumReducer.class);
		job.setOutputKeyClass(Text.class);
		job.setOutputValueClass(IntWritable.class);
		// 输入输出路径
		FileInputFormat.addInputPath(job, new Path(otherArgs[1]));
		FileOutputFormat.setOutputPath(job, new Path(otherArgs[2]));
		// 多子job的类中，可以保证各个子job串行执行
		System.exit(job.waitForCompletion(true) ? 0 : 1);
	}
}

执行：

bin/hadoop jar /tmp/wordcount.jar WordCount /tmp/3.txt /tmp/5

结果：

hadoop fs -cat /tmp/5/*
aa      1
bb      2
cc      2
dd      1

参考资料：

Hadoop - Map/Reduce 通过WordCount例子的变化来了解新版hadoop接口的变化

http://blog.csdn.net/derekjiang/article/details/6836209

Hadoop示例程序WordCount运行及详解

http://samuschen.iteye.com/blog/763940

官方的 wordcount v1.0 例子

http://hadoop.apache.org/docs/r1.1.1/mapred_tutorial.html#Example%3A+WordCount+v1.0

2、基于 MR 的数据流 Like SQL 脚本开发语言：pig

A1 = load '/data/3.txt';
A = stream A1 through `sed "s/\t/ /g"`;
B = foreach A generate flatten(TOKENIZE((chararray)$0)) as word;
C = filter B by word matches '\\w+';
D = group C by word;
E = foreach D generate COUNT(C), group;
dump E;

注意：不同分隔符对load及后面的$0的影响。

详情请见：

https://gist.github.com/186460

http://www.slideshare.net/erikeldridge/a-brief-handson-introduction-to-hadoop-pig

3、构建数据仓库的类 SQL 开发语言：hive

create table textlines(text string);
load data inpath '/data/3.txt' overwrite into table textlines;
SELECT wordColumn, count(1) FROM textlines LATERAL VIEW explode(split(text,'\t+')) wordTable AS wordColumn GROUP BY wordColumn;

详情请见：

http://my.oschina.net/leejun2005/blog/83045

http://blog.csdn.net/techdo/article/details/7433222

4、跨平台的脚本语言：python

map：

#!/usr/bin/python
import os,re,sys
for line in sys.stdin:
	for i in line.strip().split("\t"):
		print i

reduce:

#!/usr/bin/python
import os,re,sys
arr = {}
for words in sys.stdin:
	word = words.strip()
	if word not in arr:
		arr[word] = 1
	else:
		arr[word] += 1
for k, v in arr.items():
	print str(k) + ": " + str(v)

最后在shell下执行：

hadoop jar $HADOOP_HOME/contrib/streaming/hadoop-streaming-0.20.203.0.jar -file map.py -file reduce.py  -mapper map.py -reducer reduce.py -input /data/3.txt -output /data/py

注意：脚本开头需要显示指定何种解释器以及赋予脚本执行权限

详情请见：

http://blog.csdn.net/jiedushi/article/details/7390015

5、Linux 下的瑞士军刀：shell 脚本

map:

#!/bin/bash
tr '\t' '\n'

reduce:

#!/bin/bash
sort|uniq -c

最后在shell下执行：

june@deepin:~/hadoop/hadoop-0.20.203.0/tmp>
hadoop jar $HADOOP_HOME/contrib/streaming/hadoop-streaming-0.20.203.0.jar -file map.py -file reduce.py  -mapper map.py -reducer reduce.py -input /data/3.txt -output /data/py
packageJobJar: [map.py, reduce.py, /home/june/data_hadoop/tmp/hadoop-unjar2676221286002400849/] [] /tmp/streamjob8722854685251202950.jar tmpDir=null
12/10/14 21:57:00 INFO mapred.FileInputFormat: Total input paths to process : 1
12/10/14 21:57:00 INFO streaming.StreamJob: getLocalDirs(): [/home/june/data_hadoop/tmp/mapred/local]
12/10/14 21:57:00 INFO streaming.StreamJob: Running job: job_201210141552_0041
12/10/14 21:57:00 INFO streaming.StreamJob: To kill this job, run:
12/10/14 21:57:00 INFO streaming.StreamJob: /home/june/hadoop/hadoop-0.20.203.0/bin/../bin/hadoop job  -Dmapred.job.tracker=localhost:9001 -kill job_201210141552_0041
12/10/14 21:57:00 INFO streaming.StreamJob: Tracking URL: http://localhost:50030/jobdetails.jsp?jobid=job_201210141552_0041
12/10/14 21:57:01 INFO streaming.StreamJob:  map 0%  reduce 0%
12/10/14 21:57:13 INFO streaming.StreamJob:  map 67%  reduce 0%
12/10/14 21:57:19 INFO streaming.StreamJob:  map 100%  reduce 0%
12/10/14 21:57:22 INFO streaming.StreamJob:  map 100%  reduce 22%
12/10/14 21:57:31 INFO streaming.StreamJob:  map 100%  reduce 100%
12/10/14 21:57:37 INFO streaming.StreamJob: Job complete: job_201210141552_0041
12/10/14 21:57:37 INFO streaming.StreamJob: Output: /data/py
june@deepin:~/hadoop/hadoop-0.20.203.0/tmp>
hadoop fs -cat /data/py/part-00000
      1 aa	
      1 bb 	
      1 bb	
      2 cc	
      1 dd	
june@deepin:~/hadoop/hadoop-0.20.203.0/tmp>

特别提示：上述有些方法对字段后的空格忽略或计算，请注意仔细甄别。

说明：列举了上述几种方法主要是给大家一个不同的思路，

在解决问题的过程中，开发效率、执行效率都是我们需要考虑的，不要太局限某一种方法了。