MapReduce任务分析与讨论MapReduce job explained
In the last post we saw how to run a MapReduce job on Hadoop. Now we're going to analyze how a MapReduce program works. And, if you don't know what MapReduce is, the short answer is "MapReduce is a programming model for processing large data sets with a parallel, distributed algorithm on a cluster" (from Wikipedia).
Let's take a look at the source code: we can find a Java main method that is called from Hadoop, and two inner static classes, the mapper and the reducer. The code for the mapper is:
public static class TokenizerMapper extends Mapper<Object, Text, Text, IntWritable> { private final static IntWritable one = new IntWritable(1);
private Text word = new Text(); @Override
public void map(Object key, Text value, Context context) throws IOException, InterruptedException {
StringTokenizer itr = new StringTokenizer(value.toString());
while (itr.hasMoreTokens()) {
word.set(itr.nextToken());
context.write(word, one);
}
}
}
As we can see, this class extends Mapper, which - as its JavaDoc says - maps input key/value pairs to a set of intermediate key/value pairs; when the job starts, the Hadoop framework passes to the mapper a chunk of data (a subset of the whole dataset) to process. The output of the mapper will be the input of the reducers (it's not the complete story, but we'll arrive there in another post). The Mapper uses Java generics to specify what kind of data will process; in this example, we use a class that extends Mapper and specifies Object and Text as the classes of key/value pairs in input, and Text and IntWritable as the classes of key/value pairs for the output to the reducers (we'll see the details of those classes in a moment).
Let's examine the code: there's only one overridden method, the map() that takes the key/value pair as arguments and the Hadoop context; every time this method is called by Hadoop, the method receives an offset of the file where the value is as the key, and a line of the text file we're reading as the value.
Hadoop has some basic types that ore optimized for network serialization; here is a table with a few of them:
Java type | Hadoop type |
---|---|
Integer | IntWritable |
Long | LongWritable |
Double | DoubleWritable |
String | TextWritable |
Map | MapWritable |
Array | ArrayWritable |
Now it's easy to understand what this method does: for every line of the book it receives, it uses a StringTokenizer to split the line into every single word; then it sets the word in the Textobject and maps it the the value of 1; then writes it to the mappers via the Hadoop context.
Let's now look at the reducer:
public static class IntSumReducer extends Reducer<Text,IntWritable,Text,IntWritable> {
private IntWritable result = new IntWritable(); @Override
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException {
int sum = 0;
for (IntWritable val : values) {
sum += val.get();
}
result.set(sum);
context.write(key, result);
}
}
This time we have the first two arguments of the overridden method reduce that are the same type of the last two of the TokenizerMapper class; that's because - as we said - the mapper outputs the data that the reducer will use as an input. The Hadoop framework takes care of calling this method for every key that comes from the mappers; as we saw before, the keys are the words of the file we're counting the words of.
The reduce method now has to sum all the occurrences of every single word, so it initializes a sum variable to 0 and then loops over all the values for that specific key that it receives from the mappers. For every word it updates the sum variable with the value mapped to that key. At the end of the loop, when all the occurrences of that word are counted, the method sets the value obtained into an IntWritable object and gives it to the Hadoop context to be outputted to the user.
We're now at the main method of the class, which is the one that is called by Hadoop when it's executed as a JAR file.
public static void main(String[] args) throws Exception {
Configuration conf = new Configuration();
String[] otherArgs = new GenericOptionsParser(conf, args).getRemainingArgs();
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);
job.setMapperClass(TokenizerMapper.class);
job.setCombinerClass(IntSumReducer.class);
job.setReducerClass(IntSumReducer.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(IntWritable.class);
FileInputFormat.addInputPath(job, new Path(otherArgs[0]));
FileOutputFormat.setOutputPath(job, new Path(otherArgs[1]));
System.exit(job.waitForCompletion(true) ? 0 : 1);
}
}
In the method, we first setup a Configuration object, then we check for the number of arguments passed to it; If the number of arguments is correct, we create a Job object and we set a few values for making it work. Let's dive into the details:
- setJarByClass: sets the Jar by finding where a given class came from; this needs an explanation: Hadoop distributes the code to execute to the cluster as a JAR file; instead of specifying the name of the JAR, we tell Hadoop the name of the class that every instance on the cluster has to look for inside its classpath
- setMapperClass: sets the class that will be executed as the mapper
- setCombinerClass: sets the class that will be executed as the combiner (we'll explain what is a combiner in a future post)
- setReducerClass: sets the class that will be executed as the reducer
- setOutputKeyClass: sets the class that will be used as the key for outputting data to the user
- setOutputValueClass: sets the class that will be used as the value for outputting data to the user
Then we say to Hadoop where it can find the input with the FileInputFormat.addInputPath() method and where it has to write the output with the FileOutputFormat.setOutputPath()method. The last method call is the waitForCompletion(), that submits the job to the cluster and waits for it to finish.
Now that the mechanism of a MapReduce job is more clear, we can start playing with it.
from: http://andreaiacono.blogspot.com/2014/02/mapreduce-job-explained.html
MapReduce任务分析与讨论MapReduce job explained的更多相关文章
- MapReduce教程(一)基于MapReduce框架开发<转>
1 MapReduce编程 1.1 MapReduce简介 MapReduce是一种编程模型,用于大规模数据集(大于1TB)的并行运算,用于解决海量数据的计算问题. MapReduce分成了两个部分: ...
- Migrating from MapReduce 1 (MRv1) to MapReduce 2 (MRv2, YARN)...
This is a guide to migrating from Apache MapReduce 1 (MRv1) to the Next Generation MapReduce (MRv2 o ...
- 使用Cloudera Manager搭建MapReduce集群及MapReduce HA
使用Cloudera Manager搭建MapReduce集群及MapReduce HA 作者:尹正杰 版权声明:原创作品,谢绝转载!否则将追究法律责任. 一.通过CM部署MapReduce On ...
- 【MapReduce】一、MapReduce简介与实例
(一)MapReduce介绍 1.MapReduce简介 MapReduce是Hadoop生态系统的一个重要组成部分,与分布式文件系统HDFS.分布式数据库HBase一起合称为传统Hadoop的三 ...
- hadoop2.2编程:从default mapreduce program 来理解mapreduce
下面写一个default mapreduce 的程序: import org.apache.hadoop.mapreduce.Mapper; import org.apache.hadoop.mapr ...
- Top N之MapReduce程序加强版Enhanced MapReduce for Top N items
In the last post we saw how to write a MapReduce program for finding the top-n items of a dataset. T ...
- Python实现MapReduce,wordcount实例,MapReduce实现两表的Join
Python实现MapReduce 下面使用mapreduce模式实现了一个简单的统计日志中单词出现次数的程序: from functools import reduce from multiproc ...
- yarn/mapreduce工作机制及mapreduce客户端代码编写
首先需要知道的就是在老版本的hadoop中是没有yarn的,mapreduce既负责资源分配又负责业务逻辑处理.为了解耦,把资源分配这块抽了出来,形成了yarn,这样不仅mapreudce可以用yar ...
- 【MapReduce】三、MapReduce运行机制
通过前面对map端.reduce端以及整个shuffle端工作流程的介绍,我们已经了解了MapReduce的并行运算模型,基本可以使用MapReduce进行编程,那么MapRecude究竟是如何执 ...
随机推荐
- poj1979 Red And Black(DFS)
题目链接 http://poj.org/problem?id=1979 思路 floodfill问题,使用dfs解决 代码 #include <iostream> #include < ...
- 常见的mysql数据库sql语句的编写和运行结果
省份城市试题#省份表 -> select * from province;+----+----------+| id | province |+----+----------+| 1 | ...
- php读取文件内容的4钟常用方法函数
这四种方法根据不同情况使用,可以实现对文件的任何操作,下面有详细介绍. 1.把整个文件读入一个字符串中 file_get_contents(); 2.把整个文件读入一个数组中,一行就是一个数组元素 f ...
- Android手机系统设置页面跳转
android.provider.Settings. 1. ACTION_ACCESSIBILITY_SETTINGS : // 跳转系统的辅助功能界面 Intent intent = ne ...
- 【知了堂学习笔记】java 正则表达式
本文参考网络上面别人的博客知识产出 正则表达式基础 1.句号 假设你想要找出三个字母的单词,而且这些单词必须以“t”字母开头,以“n”字母结束.另外,假设有一本英文字典,你可以用正则表达式搜索它的全部 ...
- 不改源代码,修改服务器配置,解决全局Get中文乱码问题
- vdp配置
转:http://jiangjianlong.blog.51cto.com/3735273/1902879 本文将介绍VDP 6.1.2的部署与配置,主要内容包括部署VDP的OVA模板.初始化配置VD ...
- #pragma region、{}
定义一个region,这个region内部的代码你可以把它折叠起来是用于组织代码的,没有其他特别重要的意义. 而{}定义了作用域 { int a = 0; } { int a = 0; }
- luoguP2303 [SDOI2012]Longge的问题 化式子
求\(\sum \limits_{i = 1}^n gcd(i, n)\) \(\sum \limits_{i = 1}^n gcd(i, n)\) \(=\sum \limits_{i = 1}^n ...
- [Agc011F] Train Service Planning
[Agc011F] Train Service Planning 题目大意: 有n+1个车站,n条轨道,第i条轨道联通i-1和i车站,通过它要花a[i]时间,这条轨道有b[i]=1或2条车道,也就是说 ...