1 import java.io.IOException; 2 import org.apache.hadoop.conf.Configuration; 3 import org.apache.hadoop.fs.Path; 4 import org.apache.hadoop.io.IntWritable; 5 import org.apache.hadoop.io.Text; 6 import org.apache.hadoop.mapreduce.Job; 7 import org.apache.hadoop.mapreduce.Mapper; 8 import org.apache.hadoop.mapreduce.Reducer; 9 import org.apache.hadoop.mapreduce.lib.input.FileInputFormat; 10 import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat; 11 import org.apache.hadoop.util.GenericOptionsParser; 12 13 public class Sort { 14 // map将输入中的value化成IntWritable类型,作为输出的key 15 public static class Map extends 16 Mapper<Object, Text, IntWritable, IntWritable> { 17 private static IntWritable data = new IntWritable(); 18 19 // 实现map函数 20 public void map(Object key, Text value, Context context) 21 throws IOException, InterruptedException { 22 23 String line = value.toString(); 24 25 data.set(Integer.parseInt(line)); 26 27 context.write(data, new IntWritable(1)); 28 } 29 } 30 31 // reduce将输入中的key复制到输出数据的key上, 32 33 // 然后根据输入的value-list中元素的个数决定key的输出次数 34 35 // 用全局linenum来代表key的位次 36 37 public static class Reduce extends 38 39 Reducer<IntWritable, IntWritable, IntWritable, IntWritable> { 40 private static IntWritable linenum = new IntWritable(1); 41 42 // 实现reduce函数 43 public void reduce(IntWritable key, Iterable<IntWritable> values, 44 Context context) 45 46 throws IOException, InterruptedException { 47 48 for (IntWritable val : values) { 49 50 context.write(linenum, key); 51 52 linenum = new IntWritable(linenum.get() + 1); 53 54 } 55 } 56 } 57 58 public static void main(String[] args) throws Exception { 59 60 Configuration conf = new Configuration(); 61 conf.set("mapred.job.tracker", "localhost:9000"); 62 String[] ioArgs = new String[] { "hdfs://localhost:9000/input/sort", 63 "hdfs://localhost:9000/output/sortout" }; 64 65 String[] otherArgs = new GenericOptionsParser(conf, ioArgs) 66 .getRemainingArgs(); 67 68 if (otherArgs.length != 2) { 69 70 System.err.println("Usage: Data Sort <in> <out>"); 71 72 System.exit(2); 73 74 } 75 // Job job = new Job(conf, "Data Sort"); 76 Job job = Job.getInstance(conf, "Data Sort"); 77 78 job.setJarByClass(Sort.class); 79 // 设置Map和Reduce处理类 80 job.setMapperClass(Map.class); 81 82 job.setReducerClass(Reduce.class); 83 // 设置输出类型 84 job.setOutputKeyClass(IntWritable.class); 85 86 job.setOutputValueClass(IntWritable.class); 87 // 设置输入和输出目录 88 FileInputFormat.addInputPath(job, new Path(otherArgs[0])); 89 90 FileOutputFormat.setOutputPath(job, new Path(otherArgs[1])); 91 92 System.exit(job.waitForCompletion(true) ? 0 : 1); 93 } 94 }