Hadoop如何实现关联计算

    选择Hadoop，低成本和高扩展性是主要原因，但但它的开发效率实在无法让人满意。
    以关联计算为例。
    假设：HDFS上有2个文件，分别是客户信息和订单信息，customerID是它们之间的关联字段。如何进行关联计算，以便将客户名称添加到订单列表中？
    一般方法是：输入2个源文件。根据文件名在Map中处理每条数据，如果是Order，则在foreign key上加标记”O”，形成combined key；如果是Customer则做标记”C”。Map之后的数据按照key分区，再按照combined key分组排序。最后在reduce中合并结果再输出。
实现代码：
public static class JMapper extends Mapper<LongWritable, Text, TextPair, Text> {
    //mark every row with "O" or "C" according to file name
    @Override
    protected void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException {
    String pathName = ((FileSplit) context.getInputSplit()).getPath().toString();
    if (pathName.contains("order.txt")) {//identify order by file name
            String values[] = value.toString().split("\t");
            TextPair tp = new TextPair(new Text(values[1]), new Text("O"));//mark with "O"
            context.write(tp, new Text(values[0] + "\t" + values[2]));
        }
   if (pathName.contains("customer.txt")) {//identify customer by file name
           String values[] = value.toString().split("\t");
           TextPair tp = new TextPair(new Text(values[0]), new Text("C"));//mark with "C"
           context.write(tp, new Text(values[1]));
        }
    }
}
public static class JPartitioner extends Partitioner<TextPair, Text> {
    //partition by key, i.e. customerID
    @Override
    public int getPartition(TextPair key, Text value, int numParititon) {
        return Math.abs(key.getFirst().hashCode() * 127) % numParititon;
    }
}
public static class JComparator extends WritableComparator {
    //group by muti-key
    public JComparator() {
        super(TextPair.class, true);
    }
    @SuppressWarnings("unchecked")
    public int compare(WritableComparable a, WritableComparable b) {
        TextPair t1 = (TextPair) a;
        TextPair t2 = (TextPair) b;
        return t1.getFirst().compareTo(t2.getFirst());
    }
}
public static class JReduce extends Reducer<TextPair, Text, Text, Text> {
    //merge and output
    protected void reduce(TextPair key, Iterable<Text> values, Context context) throws IOException,InterruptedException {
    Text pid = key.getFirst();
    String desc = values.iterator().next().toString();
    while (values.iterator().hasNext()) {
        context.write(pid, new Text(values.iterator().next().toString() + "\t" + desc));
   }
    }
}
public class TextPair implements WritableComparable<TextPair> {
    //make muti-key
    private Text first;
    private Text second;
    public TextPair() {
        set(new Text(), new Text());
    }
    public TextPair(String first, String second) {
        set(new Text(first), new Text(second));
    }
    public TextPair(Text first, Text second) {
        set(first, second);
    }
    public void set(Text first, Text second) {
  this.first = first;
  this.second = second;
    }
    public Text getFirst() {
  return first;
    }
    public Text getSecond() {
  return second;
    }
    public void write(DataOutput out) throws IOException {
  first.write(out);
  second.write(out);
    }
    public void readFields(DataInput in) throws IOException {
  first.readFields(in);
  second.readFields(in);
    }
    public int compareTo(TextPair tp) {
  int cmp = first.compareTo(tp.first);
  if (cmp != 0) {
       return cmp;
  }
    return second.compareTo(tp.second);
    }
}
public static void main(String agrs[]) throws IOException, InterruptedException, ClassNotFoundException {
    //job entrance
    Configuration conf = new Configuration();
    GenericOptionsParser parser = new GenericOptionsParser(conf, agrs);
    String[] otherArgs = parser.getRemainingArgs();
    if (agrs.length < 3) {
   System.err.println("Usage: J <in_path_one> <in_path_two> <output>");
   System.exit(2);
    }
    Job job = new Job(conf, "J");
    job.setJarByClass(J.class);//Join class
    job.setMapperClass(JMapper.class);//Map class
    job.setMapOutputKeyClass(TextPair.class);//Map output key class
    job.setMapOutputValueClass(Text.class);//Map output value class
    job.setPartitionerClass(JPartitioner.class);//partition class
    job.setGroupingComparatorClass(JComparator.class);//condition group class after partition
    job.setReducerClass(Example_Join_01_Reduce.class);//reduce class
    job.setOutputKeyClass(Text.class);//reduce output key class
    job.setOutputValueClass(Text.class);//reduce ouput value class
    FileInputFormat.addInputPath(job, new Path(otherArgs[0]));//one of source files
    FileInputFormat.addInputPath(job, new Path(otherArgs[1]));//another file
    FileOutputFormat.setOutputPath(job, new Path(otherArgs[2]));//output path
    System.exit(job.waitForCompletion(true) ? 0 : 1);//run untill job ends
}

    不能直接使用原始数据，而是要搞一堆代码处理标记，并绕过MapReduce原本的架构，最后从底层设计并计算数据之间的关联关系。这还是最简单的关联计算，如果用MapReduce进行多表关联或逻辑更复杂的关联计算，复杂度会呈几何级数递增。