多个MapReduce之间的嵌套
[TOC]
多个MapReduce之间的嵌套
在很多实际工作中，单个MR不能满足逻辑需求，而是需要多个MR之间的相互嵌套。很多场景下，一个MR的输入依赖于另一个MR的输出。结合案例实现一下两个MR的嵌套。
Tip：如果只关心多个MR嵌套的实现，可以直接跳到下面《多个MR嵌套源码》章节查看
案例描述
根据log日志计算log中不同的IP地址数量是多少。测试数据如下图所示：
image.png
1363157985060 13726230503 30-FD-07-A4-72-B8:CMCC 120.196.100.82
1363157985061 13726230504 30-FC-07-A4-72-6A:CMCC 120.196.100.83
1363157985062 13726230505 12-FA-07-A5-72-B9:CMCC 120.196.100.84
1363157985063 13726230506 44-FE-07-A3-72-B8:CMCC 120.196.100.85
1363157985064 13726230507 56-FA-07-A4-72-B8:CMCC 120.196.100.86
1363157985065 13726230508 78-FD-07-A2-72-B8:CMCC 120.196.100.87
1363157985066 13726230545 98-FD-07-A7-72-B8:CMCC 120.196.100.87
该日志中每个字段都是用Tab建分割的。
将上述文件内容上传到HDFS：bin/hdfs dfs -mkdir /chain/，bin/hdfs dfs -put chain.txt /chain/input。
运行命令：bin/hadoop jar hadoopstudy-1.0-SNAPSHOT.jar com.ipcount.mrmr.Driver /chain/input /chain/midoutput /chain/output。
查看运行结果：bin/hdfs dfs -cat /chain/output/*
案例分析
本次任务的目的是计算该日志不同的IP地址一共有多少。实现这个目的的方式有很多种，但是本文的目的是借助改案例对两个MapReduce之间的嵌套进行总结的。
实现方法
该任务分为两个MR过程，第一个MR（命名为MR1）负责将重复的ip地址去掉，然后将无重复的ip地址进行输出。第二个MR（命名为MR2）负责将MR1输出的ip地址文件进行汇总，然后将计算总数输出。
MR1阶段
map过程
public class IpFilterMapper extends Mapper<LongWritable, Text, Text, NullWritable> {
@Override
protected void map(LongWritable key, Text value,
Mapper<LongWritable, Text, Text, NullWritable>.Context context)
throws IOException, InterruptedException {
String line = value.toString();
String[] splits = line .split("\\s+");
String ip = splits[3];
context.write(new Text(ip), NullWritable.get());
}
}
输入的key和value是文本的行号和每行的内容。
输出的key是ip地址，输出的value为空类型。
shuffle过程
主要是针对map阶段输出的key进行排序和分组，将相同的key分为一组，并且将相同key的value放到同一个集合里面，所以不同的组绝对不会出现相同的ip地址，分好组之后将值传递给reduce。注：该阶段是hadoop系统自动完成的，不需要程序员编程
reduce过程
public class IpFilterReducer extends Reducer<Text, NullWritable, Text, NullWritable> {
@Override
protected void reduce(Text key, Iterable<NullWritable> values, Context context)
throws IOException, InterruptedException {
context.write(key, NullWritable.get());
}
}
由于经过shuffle阶段之后所有输入的key都是不同的，也就是ip地址是无重复的，所以可以直接输出。
MR2阶段
map过程
public class IpCountMapper extends Mapper<LongWritable, Text, Text, NullWritable> {
@Override
protected void map(LongWritable key, Text value, Mapper<LongWritable, Text, Text, NullWritable>.Context context)
throws IOException, InterruptedException {
//输出的key为字符串"ip",这个可以随便设置，只要保证每次输出的key都一样就行
//目的是为了在shuffle阶段分组
context.write(new Text("ip"), NullWritable.get());
}
}
shuffle过程
按照相同的key进行分组，由于map阶段所有的key都一样，所以最后只有一组。
reduce过程
public class IpCountReducer extends Reducer<Text, NullWritable, Text, NullWritable> {
@Override
protected void reduce(Text key, Iterable<NullWritable> values,
Reducer<Text, NullWritable, Text, NullWritable>.Context context) throws IOException, InterruptedException {
//用于存放ip地址总数量
int count = 0;
for (NullWritable v : values) {
count ++;
}
context.write(new Text(count+""), NullWritable.get());
}
}
流程图
image.png
源码
该案例所有源码都在下面。
MR1 map源码
//MR1 map源码
package com.ipcount.mrmr;
import java.io.IOException;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.NullWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;
public class IpFilterMapper extends Mapper<LongWritable, Text, Text, NullWritable> {
@Override
protected void map(LongWritable key, Text value,
Mapper<LongWritable, Text, Text, NullWritable>.Context context)
throws IOException, InterruptedException {
String line = value.toString();
String[] splits = line .split("\\s+");
String ip = splits[3];
context.write(new Text(ip), NullWritable.get());
}
}
MR1 reduce源码
package com.ipcount.mrmr;
import java.io.IOException;
import org.apache.hadoop.io.NullWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;
public class IpFilterReducer extends Reducer<Text, NullWritable, Text, NullWritable> {
@Override
protected void reduce(Text key, Iterable<NullWritable> values, Context context)
throws IOException, InterruptedException {
context.write(key, NullWritable.get());
}
}
MR2 map源码
package com.ipcount.mrmr;
import java.io.IOException;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.NullWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;
public class IpCountMapper extends Mapper<LongWritable, Text, Text, NullWritable> {
@Override
protected void map(LongWritable key, Text value, Mapper<LongWritable, Text, Text, NullWritable>.Context context)
throws IOException, InterruptedException {
context.write(new Text("ip"), NullWritable.get());
}
}
MR2 reduce源码
package com.ipcount.mrmr;
import java.io.IOException;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.NullWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;
public class IpCountReducer extends Reducer<Text, NullWritable, Text, NullWritable> {
@Override
protected void reduce(Text key, Iterable<NullWritable> values,
Reducer<Text, NullWritable, Text, NullWritable>.Context context) throws IOException, InterruptedException {
int count = 0;
for (NullWritable v : values) {
count ++;
}
context.write(new Text(count+""), NullWritable.get());
}
}
多个MR嵌套源码
package com.ipcount.mrmr;
import java.io.IOException;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.NullWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.jobcontrol.ControlledJob;
import org.apache.hadoop.mapreduce.lib.jobcontrol.JobControl;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
public class Driver {
public static void main(String[] args) throws Exception {
JobConf conf = new JobConf(Driver.class);
//job1设置
Job job1 = new Job(conf, "job1");
job1.setJarByClass(Driver.class);
job1.setMapperClass(IpFilterMapper.class);
job1.setMapOutputKeyClass(Text.class);
job1.setMapOutputValueClass(NullWritable.class);
job1.setReducerClass(IpFilterReducer.class);
job1.setOutputKeyClass(Text.class);
job1.setOutputValueClass(NullWritable.class);
FileInputFormat.setInputPaths(job1, new Path(args[0]));
FileOutputFormat.setOutputPath(job1, new Path(args[1]));
//job1加入控制器
ControlledJob ctrlJob1 = new ControlledJob(conf);
ctrlJob1.setJob(job1);
//job2设置
Job job2 = new Job(conf, "job2");
job2.setJarByClass(Driver.class);
job2.setMapperClass(IpCountMapper.class);
job2.setMapOutputKeyClass(Text.class);
job2.setMapOutputValueClass(NullWritable.class);
job2.setReducerClass(IpCountReducer.class);
job2.setOutputKeyClass(Text.class);
job2.setOutputValueClass(NullWritable.class);
FileInputFormat.setInputPaths(job2, new Path(args[1]));
FileOutputFormat.setOutputPath(job2, new Path(args[2]));
//job2加入控制器
ControlledJob ctrlJob2 = new ControlledJob(conf);
ctrlJob2.setJob(job2);
//设置作业之间的以来关系，job2的输入以来job1的输出
ctrlJob2.addDependingJob(ctrlJob1);
//设置主控制器，控制job1和job2两个作业
JobControl jobCtrl = new JobControl("myCtrl");
//添加到总的JobControl里，进行控制
jobCtrl.addJob(ctrlJob1);
jobCtrl.addJob(ctrlJob2);
//在线程中启动，记住一定要有这个
Thread thread = new Thread(jobCtrl);
thread.start();
while (true) {
if (jobCtrl.allFinished()) {
System.out.println(jobCtrl.getSuccessfulJobList());
jobCtrl.stop();
break;
}
}
}
}
https://blog.csdn.net/u010521842/article/details/75042771
http://7de2eea1.wiz03.com/share/s/1ZUKWx0LjArB2oIU8d3nvWg42PsE6c3-gkXN2Ke6DX2DHb-4
