如何使用Hadoop的Partitioner

[b][color=olive][size=large]Hadoop里面的MapReduce编程模型,非常灵活,大部分环节我们都可以重写它的API,来灵活定制我们自己的一些特殊需求。

今天散仙要说的这个分区函数Partitioner,也是一样如此,下面我们先来看下Partitioner的作用:
对map端输出的数据key作一个散列,使数据能够均匀分布在各个reduce上进行后续操作,避免产生热点区。
Hadoop默认使用的分区函数是Hash Partitioner,源码如下:
[/size][/color][/b]
/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

package org.apache.hadoop.mapreduce.lib.partition;

import org.apache.hadoop.mapreduce.Partitioner;

/** Partition keys by their {@link Object#hashCode()}. */
public class HashPartitioner extends Partitioner {

/** Use {@link Object#hashCode()} to partition. */
public int getPartition(K key, V value,
int numReduceTasks) {
//默认使用key的hash值与上int的最大值,避免出现数据溢出 的情况
return (key.hashCode() & Integer.MAX_VALUE) % numReduceTasks;
}

}

[b][color=green][size=large]大部分情况下,我们都会使用默认的分区函数,但有时我们又有一些,特殊的需求,而需要定制Partition来完成我们的业务,案例如下:
对如下数据,按字符串的长度分区,长度为1的放在一个,2的一个,3的各一个。
[/size][/color][/b]
河南省;1
河南;2
中国;3
中国人;4
大;1
小;3
中;11

[b][color=olive][size=large]这时候,我们使用默认的分区函数,就不行了,所以需要我们定制自己的Partition,首先分析下,我们需要3个分区输出,所以在设置reduce的个数时,一定要设置为3,其次在partition里,进行分区时,要根据长度具体分区,而不是根据字符串的hash码来分区。核心代码如下:[/size][/color][/b]
	/**
* Partitioner
*
*
* */
public static class PPartition extends Partitioner{
@Override
public int getPartition(Text arg0, Text arg1, int arg2) {
/**
* 自定义分区,实现长度不同的字符串,分到不同的reduce里面
*
* 现在只有3个长度的字符串,所以可以把reduce的个数设置为3
* 有几个分区,就设置为几
* */

String key=arg0.toString();
if(key.length()==1){
return 1%arg2;
}else if(key.length()==2){
return 2%arg2;
}else if(key.length()==3){
return 3%arg2;
}



return 0;
}




}


[b][color=olive][size=large]全部代码如下:[/size][/color][/b]
package com.partition.test;

import java.io.IOException;

import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Partitioner;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.db.DBConfiguration;
import org.apache.hadoop.mapreduce.lib.db.DBInputFormat;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.mapreduce.lib.output.MultipleOutputs;
import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat;

import com.qin.operadb.PersonRecoder;
import com.qin.operadb.ReadMapDB;


/**
* @author qindongliang
*
* 大数据交流群:376932160
*
*
* **/
public class MyTestPartition {

/**
* map任务
*
* */
public static class PMapper extends Mapper{

@Override
protected void map(LongWritable key, Text value,Context context)
throws IOException, InterruptedException {
// System.out.println("进map了");
//mos.write(namedOutput, key, value);
String ss[]=value.toString().split(";");

context.write(new Text(ss[0]), new Text(ss[1]));



}


}

/**
* Partitioner
*
*
* */
public static class PPartition extends Partitioner{
@Override
public int getPartition(Text arg0, Text arg1, int arg2) {
/**
* 自定义分区,实现长度不同的字符串,分到不同的reduce里面
*
* 现在只有3个长度的字符串,所以可以把reduce的个数设置为3
* 有几个分区,就设置为几
* */

String key=arg0.toString();
if(key.length()==1){
return 1%arg2;
}else if(key.length()==2){
return 2%arg2;
}else if(key.length()==3){
return 3%arg2;
}



return 0;
}




}


/***
* Reduce任务
*
* **/
public static class PReduce extends Reducer{
@Override
protected void reduce(Text arg0, Iterable arg1, Context arg2)
throws IOException, InterruptedException {

String key=arg0.toString().split(",")[0];
System.out.println("key==> "+key);
for(Text t:arg1){
//System.out.println("Reduce: "+arg0.toString()+" "+t.toString());
arg2.write(arg0, t);
}


}



}


public static void main(String[] args) throws Exception{
JobConf conf=new JobConf(ReadMapDB.class);
//Configuration conf=new Configuration();
conf.set("mapred.job.tracker","192.168.75.130:9001");
//读取person中的数据字段
conf.setJar("tt.jar");
//注意这行代码放在最前面,进行初始化,否则会报


/**Job任务**/
Job job=new Job(conf, "testpartion");
job.setJarByClass(MyTestPartition.class);
System.out.println("模式: "+conf.get("mapred.job.tracker"));;
// job.setCombinerClass(PCombine.class);
job.setPartitionerClass(PPartition.class);

job.setNumReduceTasks(3);//设置为3
job.setMapperClass(PMapper.class);
// MultipleOutputs.addNamedOutput(job, "hebei", TextOutputFormat.class, Text.class, Text.class);
// MultipleOutputs.addNamedOutput(job, "henan", TextOutputFormat.class, Text.class, Text.class);
job.setReducerClass(PReduce.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);

String path="hdfs://192.168.75.130:9000/root/outputdb";
FileSystem fs=FileSystem.get(conf);
Path p=new Path(path);
if(fs.exists(p)){
fs.delete(p, true);
System.out.println("输出路径存在,已删除!");
}
FileInputFormat.setInputPaths(job, "hdfs://192.168.75.130:9000/root/input");
FileOutputFormat.setOutputPath(job,p );
System.exit(job.waitForCompletion(true) ? 0 : 1);


}



}

[b][size=large][color=green]运行情况如下:[/color][/size][/b]

模式:  192.168.75.130:9001
输出路径存在,已删除!
WARN - JobClient.copyAndConfigureFiles(746) | Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
INFO - FileInputFormat.listStatus(237) | Total input paths to process : 1
WARN - NativeCodeLoader.(52) | Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
WARN - LoadSnappy.(46) | Snappy native library not loaded
INFO - JobClient.monitorAndPrintJob(1380) | Running job: job_201404101853_0005
INFO - JobClient.monitorAndPrintJob(1393) | map 0% reduce 0%
INFO - JobClient.monitorAndPrintJob(1393) | map 100% reduce 0%
INFO - JobClient.monitorAndPrintJob(1393) | map 100% reduce 11%
INFO - JobClient.monitorAndPrintJob(1393) | map 100% reduce 22%
INFO - JobClient.monitorAndPrintJob(1393) | map 100% reduce 55%
INFO - JobClient.monitorAndPrintJob(1393) | map 100% reduce 100%
INFO - JobClient.monitorAndPrintJob(1448) | Job complete: job_201404101853_0005
INFO - Counters.log(585) | Counters: 29
INFO - Counters.log(587) | Job Counters
INFO - Counters.log(589) | Launched reduce tasks=3
INFO - Counters.log(589) | SLOTS_MILLIS_MAPS=7422
INFO - Counters.log(589) | Total time spent by all reduces waiting after reserving slots (ms)=0
INFO - Counters.log(589) | Total time spent by all maps waiting after reserving slots (ms)=0
INFO - Counters.log(589) | Launched map tasks=1
INFO - Counters.log(589) | Data-local map tasks=1
INFO - Counters.log(589) | SLOTS_MILLIS_REDUCES=30036
INFO - Counters.log(587) | File Output Format Counters
INFO - Counters.log(589) | Bytes Written=61
INFO - Counters.log(587) | FileSystemCounters
INFO - Counters.log(589) | FILE_BYTES_READ=93
INFO - Counters.log(589) | HDFS_BYTES_READ=179
INFO - Counters.log(589) | FILE_BYTES_WRITTEN=218396
INFO - Counters.log(589) | HDFS_BYTES_WRITTEN=61
INFO - Counters.log(587) | File Input Format Counters
INFO - Counters.log(589) | Bytes Read=68
INFO - Counters.log(587) | Map-Reduce Framework
INFO - Counters.log(589) | Map output materialized bytes=93
INFO - Counters.log(589) | Map input records=7
INFO - Counters.log(589) | Reduce shuffle bytes=93
INFO - Counters.log(589) | Spilled Records=14
INFO - Counters.log(589) | Map output bytes=61
INFO - Counters.log(589) | Total committed heap usage (bytes)=207491072
INFO - Counters.log(589) | CPU time spent (ms)=2650
INFO - Counters.log(589) | Combine input records=0
INFO - Counters.log(589) | SPLIT_RAW_BYTES=111
INFO - Counters.log(589) | Reduce input records=7
INFO - Counters.log(589) | Reduce input groups=7
INFO - Counters.log(589) | Combine output records=0
INFO - Counters.log(589) | Physical memory (bytes) snapshot=422174720
INFO - Counters.log(589) | Reduce output records=7
INFO - Counters.log(589) | Virtual memory (bytes) snapshot=2935713792
INFO - Counters.log(589) | Map output records=7

[b][color=green][size=large]运行后的结果文件如下:[/size][/color][/b]
[img]http://dl2.iteye.com/upload/attachment/0095/9674/9c2db39f-1df8-3f04-89a2-ff699e9a4a8d.jpg[/img]

[b][color=green][size=large]其中,part-r-000000里面的数据[/size][/color][/b]
中国人	4
河南省 1


[b][color=green][size=large]其中,part-r-000001里面的数据[/size][/color][/b]
中	11
大 1
小 3



[b][color=green][size=large]其中,part-r-000002里面的数据[/size][/color][/b]
中国	3
河南 2

[b][color=olive][size=large]至此,我们使用自定义的分区策略完美的实现了,数据分区了。


总结:引用一段话

(Partition)分区出现的必要性,如何使用Hadoop产生一个全局排序的文件?最简单的方法就是使用一个分区,但是该方法在处理大型文件时效率极低,因为一台机器必须处理所有输出文件,从而完全丧失了MapReduce所提供的并行架构的优势。事实上我们可以这样做,首先创建一系列排好序的文件;其次,串联这些文件(类似于归并排序);最后得到一个全局有序的文件。主要的思路是使用一个partitioner来描述全局排序的输出。比方说我们有1000个1-10000的数据,跑10个ruduce任务, 如果我们运行进行partition的时候,能够将在1-1000中数据的分配到第一个reduce中,1001-2000的数据分配到第二个reduce中,以此类推。即第n个reduce所分配到的数据全部大于第n-1个reduce中的数据。这样,每个reduce出来之后都是有序的了,我们只要cat所有的输出文件,变成一个大的文件,就都是有序的了

基本思路就是这样,但是现在有一个问题,就是数据的区间如何划分,在数据量大,还有我们并不清楚数据分布的情况下。一个比较简单的方法就是采样,假如有一亿的数据,我们可以对数据进行采样,如取10000个数据采样,然后对采样数据分区间。在Hadoop中,patition我们可以用TotalOrderPartitioner替换默认的分区。然后将采样的结果传给他,就可以实现我们想要的分区。在采样时,我们可以使用hadoop的几种采样工具,RandomSampler,InputSampler,IntervalSampler。

这样,我们就可以对利用分布式文件系统进行大数据量的排序了,我们也可以重写Partitioner类中的compare函数,来定义比较的规则,从而可以实现字符串或其他非数字类型的排序,也可以实现二次排序乃至多次排序。
[/size][/color][/b]

你可能感兴趣的:(Hadoop)