什么情况下，hive只会产生一个reduceTask，而没有mapTask

正常逻辑下，mapTask的职责就是负责读数据，做ETL，也可以利用combiner局部聚合；ReduceTask输入严重依赖于mapper输出，所以‘一直’的逻辑仅有reducer无法执行的。

hql常规解析后依赖的mapreduce主要还是ExecReducer/ExecMapper--通过实现hadoop的mapper和reducer来实现具体执行逻辑。

hive自己的mapreducer有一个实现案例，就是GenericMR，实现源码如下：

public final class GenericMR {
  public void map(final InputStream in, final OutputStream out,
      final Mapper mapper) throws Exception {
    map(new InputStreamReader(in), new OutputStreamWriter(out), mapper);
  }
 
 
  public void map(final Reader in, final Writer out, final Mapper mapper) throws Exception {
    handle(in, out, new RecordProcessor() {
      @Override
      public void processNext(RecordReader reader, Output output) throws Exception {
        mapper.map(reader.next(), output);
      }
    });
  }
 
 
  public void reduce(final InputStream in, final OutputStream out,
      final Reducer reducer) throws Exception {
    reduce(new InputStreamReader(in), new OutputStreamWriter(out), reducer);
  }
 
 
  public void reduce(final Reader in, final Writer out, final Reducer reducer) throws Exception {
    handle(in, out, new RecordProcessor() {
      @Override
      public void processNext(RecordReader reader, Output output) throws Exception {
        reducer.reduce(reader.peek()[0], new KeyRecordIterator(
            reader.peek()[0], reader), output);
      }
    });
  }
 
 
  private void handle(final Reader in, final Writer out,
      final RecordProcessor processor) throws Exception {
    final RecordReader reader = new RecordReader(in);
    final OutputStreamOutput output = new OutputStreamOutput(out);
 
 
    try {
      while (reader.hasNext()) {
        processor.processNext(reader, output);
      }
    } finally {
      try {
        output.close();
      } finally {
        reader.close();
      }
    }
  }
 
 
  private static interface RecordProcessor {
    void processNext(final RecordReader reader, final Output output) throws Exception;
  }
 
 
  private static final class KeyRecordIterator implements Iterator<String[]> {
    private final String key;
    private final RecordReader reader;
 
 
    private KeyRecordIterator(final String key, final RecordReader reader) {
      this.key = key;
      this.reader = reader;
    }
 
 
    @Override
    public boolean hasNext() {
      return (reader.hasNext() && key.equals(reader.peek()[0]));
    }
 
 
    @Override
    public String[] next() {
      if (!hasNext()) {
        throw new NoSuchElementException();
      }
 
 
      return reader.next();
    }
 
 
    @Override
    public void remove() {
      throw new UnsupportedOperationException();
    }
  }
 
 
  private static final class RecordReader {
    private final BufferedReader reader;
    private String[] next;
 
 
    private RecordReader(final InputStream in) {
      this(new InputStreamReader(in));
    }
 
 
    private RecordReader(final Reader in) {
      reader = new BufferedReader(in);
      next = readNext();
    }
 
 
    private String[] next() {
      final String[] ret = next;
 
 
      next = readNext();
 
 
      return ret;
    }
 
 
    private String[] readNext() {
      try {
        final String line = reader.readLine();
        return (line == null ? null : line.split("\t"));
      } catch (final Exception e) {
        throw new RuntimeException(e);
      }
    }
 
 
    private boolean hasNext() {
      return next != null;
    }
 
 
    private String[] peek() {
      return next;
    }
 
 
    private void close() throws Exception {
      reader.close();
      
    }
  }
 
 
  private static final class OutputStreamOutput implements Output {
    private final PrintWriter out;
 
 
    private OutputStreamOutput(final OutputStream out) {
      this(new OutputStreamWriter(out));
    }
 
 
    private OutputStreamOutput(final Writer out) {
      this.out = new PrintWriter(out);
    }
 
 
    public void close() throws Exception {
      out.close();
    }
 
 
    @Override
    public void collect(String[] record) throws Exception {
      out.println(_join(record, "\t"));
    }
 
 
    private static String _join(final String[] record, final String separator) {
      if (record == null || record.length == 0) {
        return "";
      }
      final StringBuilder sb = new StringBuilder();
      for (int i = 0; i < record.length; i++) {
        if (i > 0) {
          sb.append(separator);
        }
        sb.append(record[i]);
      }
      return sb.toString();
    }
  }
}
 
 

重点：常规的mapreducer的reducer输入依赖于mapper的输出的，所以无法单独执行。但是GenericMR的实现reducer里可以直接支持InputStream/Reader，所以就可以直接生成java的指定输入流或者reader即可

new GenericMR().reduce(new StringReader("a\tb\tc"), new StringWriter(),
          new Reducer() {
        public void reduce(String key, Iterator<String[]> records,
            Output output) throws Exception {
          while (true) {
            records.next();
          }
        }
});