首页 > 代码库 > Pig、Hive、MapReduce 解决分组 Top K 问题(转)

Pig、Hive、MapReduce 解决分组 Top K 问题(转)

问题:

有如下数据文件 city.txt (id, city, value)

cat city.txt 
1 wh 500
2 bj 600
3 wh 100
4 sh 400
5 wh 200
6 bj 100
7 sh 200
8 bj 300
9 sh 900
需要按 city 分组聚合,然后从每组数据中取出前两条value最大的记录。

1、这是实际业务中经常会遇到的 group TopK 问题,下面来看看 pig 如何解决:

1a = load ‘/data/city.txt‘  using PigStorage(‘ ‘as (id:chararray, city:chararray, value:int);
2b = group by city;
3c = foreach b {c1=order by value desc; c2=limit c1 2; generate group,c2.value;};
4d = stream c through `sed ‘s/[(){}]//g‘`;
5dump d;
结果:
1(bj,600,300)
2(sh,900,400)
3(wh,500,200)
这几行代码其实也实现了mysql中的 group_concat 函数的功能:
1a = load ‘/data/city.txt‘  using PigStorage(‘ ‘as (id:chararray, city:chararray, value:int);
2b = group by city;
3c = foreach b {c1=order by value desc;  generate group,c1.value;};
4d = stream c through `sed ‘s/[(){}]//g‘`;
5dump d;
结果:
1(bj,600,300,100)
2(sh,900,400,200)
3(wh,500,200,100)

2、下面我们再来看看hive如何处理group topk的问题:

本质上HSQL和sql有很多相同的地方,但HSQL目前功能还有很多缺失,至少不如原生态的SQL功能强大,

比起PIG也有些差距,如果SQL中这类分组topk的问题如何解决呢?

1select from city a where
22>(select count(1) from city where cname=a.cname and value>a.value)
3distribute by a.cname sort by a.cname,a.value desc;
http://my.oschina.net/leejun2005/blog/78904

但是这种写法在HQL中直接报语法错误了,下面我们只能用hive udf的思路来解决了:

排序city和value,然后对city计数,最后where过滤掉city列计数器大于k的行即可。

好了,上代码:

(1)定义UDF:

01package com.example.hive.udf;
02import org.apache.hadoop.hive.ql.exec.UDF;
03      
04public final class Rank extends UDF{
05    private int  counter;
06    private String last_key;
07    public int evaluate(final String key){
08      if ( !key.equalsIgnoreCase(this.last_key) ) {
09         this.counter = 0;
10         this.last_key = key;
11      }
12      return this.counter++;
13    }
14}
(2)注册jar、建表、导数据,查询:
1add jar Rank.jar;
2create temporary function rank as ‘com.example.hive.udf.Rank‘;
3create table city(id int,cname string,value int) row format delimited fields terminated by ‘ ‘;
4LOAD DATA LOCAL INPATH ‘city.txt‘ OVERWRITE INTO TABLE city;
5select cname, value from (
6    select cname,rank(cname) csum,value from (
7        select id, cname, value from city distribute by cname sort by cname,value desc
8    )a
9)b where csum < 2;

(3)结果:

 

1bj  600
2bj  300
3sh  900
4sh  400
5wh  500
6wh  200
可以看到,hive相比pig来说,处理起来稍微复杂了点,但随着hive的日渐完善,以后比pig更简洁也说不定。

REF:hive中分组取前N个值的实现

http://baiyunl.iteye.com/blog/1466343

 

3、最后我们来看一下原生态的MR:

 

01import java.io.IOException;
02import java.util.TreeSet;
03 
04import org.apache.hadoop.conf.Configuration;
05import org.apache.hadoop.fs.Path;
06import org.apache.hadoop.io.IntWritable;
07import org.apache.hadoop.io.LongWritable;
08import org.apache.hadoop.io.Text;
09import org.apache.hadoop.mapreduce.Job;
10import org.apache.hadoop.mapreduce.Mapper;
11import org.apache.hadoop.mapreduce.Reducer;
12import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
13import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
14import org.apache.hadoop.util.GenericOptionsParser;
15 
16public class GroupTopK {
17    // 这个 MR 将会取得每组年龄中 id 最大的前 3 个
18    // 测试数据由脚本生成:http://my.oschina.net/leejun2005/blog/76631
19    public static class GroupTopKMapper extends
20            Mapper<LongWritable, Text, IntWritable, LongWritable> {
21        IntWritable outKey = new IntWritable();
22        LongWritable outValue =http://www.mamicode.com/ new LongWritable();
23        String[] valArr = null;
24 
25        public void map(LongWritable key, Text value, Context context)
26                throws IOException, InterruptedException {
27            valArr = value.toString().split("\t");
28            outKey.set(Integer.parseInt(valArr[2]));// age int
29            outValue.set(Long.parseLong(valArr[0]));// id long
30            context.write(outKey, outValue);
31        }
32    }
33 
34    public static class GroupTopKReducer extends
35            Reducer<IntWritable, LongWritable, IntWritable, LongWritable> {
36 
37        LongWritable outValue =http://www.mamicode.com/ new LongWritable();
38 
39        public void reduce(IntWritable key, Iterable<LongWritable> values,
40                Context context) throws IOException, InterruptedException {
41            TreeSet<Long> idTreeSet = new TreeSet<Long>();
42            for (LongWritable val : values) {
43                idTreeSet.add(val.get());
44                if (idTreeSet.size() > 3) {
45                    idTreeSet.remove(idTreeSet.first());
46                }
47            }
48            for (Long id : idTreeSet) {
49                outValue.set(id);
50                context.write(key, outValue);
51            }
52        }
53    }
54 
55    public static void main(String[] args) throws Exception {
56        Configuration conf = new Configuration();
57        String[] otherArgs = new GenericOptionsParser(conf, args)
58                .getRemainingArgs();
59 
60        System.out.println(otherArgs.length);
61        System.out.println(otherArgs[0]);
62        System.out.println(otherArgs[1]);
63 
64        if (otherArgs.length != 3) {
65            System.err.println("Usage: GroupTopK <in> <out>");
66            System.exit(2);
67        }
68        Job job = new Job(conf, "GroupTopK");
69        job.setJarByClass(GroupTopK.class);
70        job.setMapperClass(GroupTopKMapper.class);
71        job.setReducerClass(GroupTopKReducer.class);
72        job.setNumReduceTasks(1);
73        job.setOutputKeyClass(IntWritable.class);
74        job.setOutputValueClass(LongWritable.class);
75        FileInputFormat.addInputPath(job, new Path(otherArgs[1]));
76        FileOutputFormat.setOutputPath(job, new Path(otherArgs[2]));
77        System.exit(job.waitForCompletion(true) ? 0 1);
78    }
79}

hadoop jar GroupTopK.jar GroupTopK /tmp/decli/record_new.txt /tmp/1

结果:

 

hadoop fs -cat /tmp/1/part-r-00000
0       12869695
0       12869971
0       12869976
1       12869813
1       12869870
1       12869951

......

数据验证:

awk ‘$3==0{print $1}‘ record_new.txt|sort -nr|head -3
12869976
12869971
12869695

可以看到结果没有问题。 

注:测试数据由以下脚本生成:

http://my.oschina.net/leejun2005/blog/76631

 

PS:

如果说hive类似sql的话,那pig就类似plsql存储过程了:程序编写更自由,逻辑能处理的更强大了。

pig中还能直接通过反射调用java的静态类中的方法,这块内容请参考之前的相关pig博文。

附几个HIVE UDAF链接,有兴趣的同学自己看下:

Hive UDAF和UDTF实现group by后获取top值 http://blog.csdn.net/liuzhoulong/article/details/7789183
hive中自定义函数(UDAF)实现多行字符串拼接为一行 http://blog.sina.com.cn/s/blog_6ff05a2c0100tjw4.html
编写Hive UDAF http://www.fuzhijie.me/?p=118
Hive UDAF开发 http://richiehu.blog.51cto.com/2093113/386113

Pig、Hive、MapReduce 解决分组 Top K 问题(转)