二次排序,从字面上可以理解为在对key排序的基础上对key所对应的值value排序,也叫辅助排序。一般情况下,MapReduce框架只对key排序,而不对key所对应的值排序,因此value的排序经常是不固定的。但是我们经常会遇到同时对key和value排序的需求,例如Hadoop权威指南中的求一年的高高气温,key为年份,value为最高气温,年份按照降序排列,气温按照降序排列。还有水果电商网站经常会有按天统计水果销售排行榜的需求等等,这些都是需要对key和value同时进行排序。如下图所示:
如何设计一个MapReduce程序解决对key和value同时排序的需求呢?这就需要用到组合键、分区、分组的概念。在这里又看到分区的影子,可知分区在MapReduce是多么的重要,一定要好好掌握,是优化的重点。
按照上图中数据流转的方向,我们首先设计一个Fruit类,有三个字段,分别是日期、水果名和销量,将日期、水果名和销量作为一个复合键;接着设计一个自定义Partition类,根据Fruit的日期字段分区,让相同日期的数据流向同一个partition分区中;最后定义一个分组类,实现同一个分区内的数据分组,然后按照销量字段进行二次排序。
具体实现思路:
1、定义Fruit类,实现WritableComparable接口,并且重写compareTo、equal和hashcode方法以及序列化和反序列化方法readFields和write方法。Java类要在网络上传输必须序列化和反序列化。在Map端的map函数中将Fruit对象当做key。compareTo方法用于比较两个key的大小,在本文中就是比较两个Fruit对象的排列顺序。
2、自定义第一次排序类,继承WritableComparable或者WritableComparator接口,重写compareTo或者compare方法,。就是在Map端对Fruit对象的第一个字段进行排序
3、自定义Partition类,实现Partitioner接口,并且重写getPartition方法,将日期相同的Fruit对象分发到同一个partition中。
4、定义分组类,继承WritableComparator接口,并且重写compare方法。用于比较同一分组内两个Fruit对象的排列顺序,根据销量字段比较。日期相同的Fruit对象会划分到同一个分组。通过setGroupingComparatorClass方法设置分组类。如果不设置分组类,则按照key默认的compare方法来对key进行排序。
代码如下:
1 import org.apache.hadoop.conf.Configured; 2 import org.apache.hadoop.io.WritableComparable; 3 import java.io.DataInput; 4 import java.io.DataOutput; 5 import java.io.IOException; 6 import org.apache.hadoop.io.*; 7 import org.apache.hadoop.mapreduce.Partitioner; 8 import org.apache.hadoop.mapreduce.Mapper; 9 import org.apache.hadoop.mapreduce.Reducer; 10 import org.apache.hadoop.conf.Configuration; 11 import org.apache.hadoop.fs.FileSystem; 12 import org.apache.hadoop.fs.Path; 13 import org.apache.hadoop.mapreduce.Job; 14 import org.apache.hadoop.mapreduce.lib.input.FileInputFormat; 15 import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat; 16 import org.apache.hadoop.mapreduce.lib.input.TextInputFormat; 17 import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat; 18 import org.apache.hadoop.util.Tool; 19 import org.apache.hadoop.util.ToolRunner; 20 import org.slf4j.Logger; 21 import org.slf4j.LoggerFactory; 22 23 public class SecondrySort extends Configured implements Tool { 24 25 static class Fruit implements WritableComparable<Fruit>{ 26 private static final Logger logger = LoggerFactory.getLogger(Fruit.class); 27 private String date; 28 private String name; 29 private Integer sales; 30 public Fruit(){ 31 } 32 public Fruit(String date,String name,Integer sales){ 33 this.date = date; 34 this.name = name; 35 this.sales = sales; 36 } 37 38 public String getDate(){ 39 return this.date; 40 } 41 42 public String getName(){ 43 return this.name; 44 } 45 46 public Integer getSales(){ 47 return this.sales; 48 } 49 50 @Override 51 public void readFields(DataInput in) throws IOException{ 52 this.date = in.readUTF(); 53 this.name = in.readUTF(); 54 this.sales = in.readInt(); 55 } 56 57 @Override 58 public void write(DataOutput out) throws IOException{ 59 out.writeUTF(this.date); 60 out.writeUTF(this.name); 61 out.writeInt(sales); 62 } 63 64 @Override 65 public int compareTo(Fruit other) { 66 int result1 = this.date.compareTo(other.getDate()); 67 if(result1 == 0) { 68 int result2 = this.sales - other.getSales(); 69 if (result2 == 0) { 70 double result3 = this.name.compareTo(other.getName()); 71 if(result3 > 0) return -1; 72 else if(result3 < 0) return 1; 73 else return 0; 74 }else if(result2 >0){ 75 return -1; 76 }else if(result2 < 0){ 77 return 1; 78 } 79 }else if(result1 > 0){ 80 return -1; 81 }else{ 82 return 1; 83 } 84 return 0; 85 } 86 87 @Override 88 public int hashCode(){ 89 return this.date.hashCode() * 157 + this.sales + this.name.hashCode(); 90 } 91 92 @Override 93 public boolean equals(Object object){ 94 if (object == null) 95 return false; 96 if (this == object) 97 return true; 98 if (object instanceof Fruit){ 99 Fruit r = (Fruit) object;100 // if(r.getDate().toString().equals(this.getDate().toString())){101 return r.getDate().equals(this.getDate()) && r.getName().equals(this.getName())102 && this.getSales() == r.getSales();103 }else{104 return false;105 }106 }107 108 public String toString() {109 return this.date + " " + this.name + " " + this.sales;110 }111 112 }113 114 static class FruitPartition extends Partitioner<Fruit, NullWritable>{115 @Override116 public int getPartition(Fruit key, NullWritable value,int numPartitions){117 return Math.abs(Integer.parseInt(key.getDate()) * 127) % numPartitions;118 }119 }120 121 public static class GroupingComparator extends WritableComparator{122 protected GroupingComparator(){123 super(Fruit.class, true);124 }125 126 @Override127 public int compare(WritableComparable w1, WritableComparable w2){128 Fruit f1 = (Fruit) w1;129 Fruit f2 = (Fruit) w2;130 131 if(!f1.getDate().equals(f2.getDate())){132 return f1.getDate().compareTo(f2.getDate());133 }else{134 return f1.getSales().compareTo(f2.getSales());135 }136 }137 }138 139 public static class Map extends Mapper<LongWritable, Text, Fruit, NullWritable> {140 141 public void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException {142 String line = value.toString();143 String str[] = line.split(" ");144 Fruit fruit = new Fruit(str[0],str[1],new Integer(str[2]));145 //Fruit fruit = new Fruit();146 //fruit.set(str[0],str[1],new Integer(str[2]));147 context.write(fruit, NullWritable.get());148 }149 }150 151 public static class Reduce extends Reducer<Fruit, NullWritable, Text, NullWritable> {152 153 public void reduce(Fruit key, Iterable<NullWritable> values, Context context) throws IOException, InterruptedException {154 String str = key.getDate() + " " + key.getName() + " " + key.getSales();155 context.write(new Text(str), NullWritable.get());156 }157 }158 159 @Override160 public int run(String[] args) throws Exception {161 Configuration conf = new Configuration();162 // 判断路径是否存在,如果存在,则删除163 Path mypath = new Path(args[1]);164 FileSystem hdfs = mypath.getFileSystem(conf);165 if (hdfs.isDirectory(mypath)) {166 hdfs.delete(mypath, true);167 }168 169 Job job = Job.getInstance(conf, "Secondry Sort app");170 // 设置主类171 job.setJarByClass(SecondrySort.class);172 173 // 输入路径174 FileInputFormat.setInputPaths(job, new Path(args[0]));175 // 输出路径176 FileOutputFormat.setOutputPath(job, new Path(args[1]));177 178 // Mapper179 job.setMapperClass(Map.class);180 // Reducer181 job.setReducerClass(Reduce.class);182 183 // 分区函数184 job.setPartitionerClass(FruitPartition.class);185 186 // 分组函数187 job.setGroupingComparatorClass(GroupingComparator.class);188 189 // map输出key类型190 job.setMapOutputKeyClass(Fruit.class);191 // map输出value类型192 job.setMapOutputValueClass(NullWritable.class);193 194 // reduce输出key类型195 job.setOutputKeyClass(Text.class);196 // reduce输出value类型197 job.setOutputValueClass(NullWritable.class);198 199 // 输入格式200 job.setInputFormatClass(TextInputFormat.class);201 // 输出格式202 job.setOutputFormatClass(TextOutputFormat.class);203 204 return job.waitForCompletion(true) ? 0 : 1;205 }206 207 public static void main(String[] args) throws Exception{208 int exitCode = ToolRunner.run(new SecondrySort(), args);209 System.exit(exitCode);210 }211 }
测试数据:
20180906 Apple 200
20180904 Apple 200
20180905 Banana 100
20180906 Orange 300
20180906 Banana 400
20180904 Orange 100
20180905 Apple 400
20180904 Banana 300
20180905 Orange 500
运行结果:
20180906 Banana 400
20180906 Orange 300
20180906 Apple 200
20180905 Orange 500
20180905 Apple 400
20180905 Banana 100
20180904 Banana 300
20180904 Apple 200
20180904 Orange 100
总结:
1、在使用实现WritableComparable接口的方式实现自定义比较器时,必须有一个无参的构造函数。否则会报Unable to initialize any output collector的错误。
2、readFields和write方法中处理字段的顺序必须一致,否则会报MapReduce Error: java.io.EOFException at java.io.DataInputStream.readFully(DataInputStream.java:197)的错误。
原文出处:https://www.cnblogs.com/airnew/p/9631718.html