hive源码之explode函数

目录

一、函数介绍

二、使用案例

三、源码分析

四、总结

一、函数介绍

UDTF（User-Defined Table-Generating Functions）是一进多出函数，如hive中的explode()、posexplode()函数。explode()函数可以将数组（array类型）的元素分隔成多行，或将映射（map类型）的元素分隔为多行和多列。工作中经常会用到这个函数，今天我们这次来分析下explode()函数源码。

二、使用案例

查询sql以及结果

select explode(array(1,2,3));

三、源码分析

一个自定义的UDTF函数需要继承GenericUDTF抽象类，且实现initialize()、 process()、close()（可选）方法。initialize()方法会被hive调用去通知UDTF函数将要接收到的参数类型。该UDTF必须返回一个与UDTF函数输出相对应的对象检查器。一旦initialize()方法被调用，hive将通过process()方法把一行行数据传给UDTF。在process()方法中，UDTF可以通过调用forward()方法将数据传给其他的operator。最后，当把所有的数据都处理完以后hive会调用close()方法。
由于需要继承GenericUDTF抽象类，我们先来看下GenericUDTF抽象类源码。

/**
 * A Generic User-defined Table Generating Function (UDTF)
 *
 * Generates a variable number of output rows for a single input row. Useful for
 * explode(array)...
 */

public abstract class GenericUDTF {
  Collector collector = null;

  /**
   * Additionally setup GenericUDTF with MapredContext before initializing.
   * This is only called in runtime of MapRedTask.
   *
   * @param context context
   */
  public void configure(MapredContext mapredContext) {
  }

  public StructObjectInspector initialize(StructObjectInspector argOIs)
      throws UDFArgumentException {
    List<? extends StructField> inputFields = argOIs.getAllStructFieldRefs();
    ObjectInspector[] udtfInputOIs = new ObjectInspector[inputFields.size()];
    for (int i = 0; i < inputFields.size(); i++) {
      udtfInputOIs[i] = inputFields.get(i).getFieldObjectInspector();
    }
    return initialize(udtfInputOIs);
  }

  /**
   * Initialize this GenericUDTF. This will be called only once per instance.
   *
   * @param argOIs
   *          An array of ObjectInspectors for the arguments
   * @return A StructObjectInspector for output. The output struct represents a
   *         row of the table where the fields of the stuct are the columns. The
   *         field names are unimportant as they will be overridden by user
   *         supplied column aliases.
   */
  @Deprecated
  public StructObjectInspector initialize(ObjectInspector[] argOIs)
      throws UDFArgumentException {
    throw new IllegalStateException("Should not be called directly");
  }

  /**
   * Give a set of arguments for the UDTF to process.
   *
   * @param args
   *          object array of arguments
   */
  public abstract void process(Object[] args) throws HiveException;

  /**
   * Called to notify the UDTF that there are no more rows to process.
   * Clean up code or additional forward() calls can be made here.
   */
  public abstract void close() throws HiveException;

  /**
   * Associates a collector with this UDTF. Can't be specified in the
   * constructor as the UDTF may be initialized before the collector has been
   * constructed.
   *
   * @param collector
   */
  public final void setCollector(Collector collector) {
    this.collector = collector;
  }

  /**
   * Passes an output row to the collector.
   *
   * @param o
   * @throws HiveException
   */
  protected final void forward(Object o) throws HiveException {
    collector.collect(o);
  }

}

1. initialize初始化：UDTF首先会调用initialize方法，此方法返回UDTF的返回行的信息（返回个数，类型，名称）。其实就是把上一个operator返回的对象检查器传给UDTF，因为object inspector 对象检查器中保存了数据的类型，initialize针对任务调一次。
2. process：初始化完成后，会调用process方法，对传入的参数进行处理，可以通过forword()方法把结果写出。传入一行数据写出去多次，与mapreduce中的map方法很像，也是一行一行的数据传入，传入一行数据输出多行数据，process针对每行数据调用一次该方法。
3. close：最后close()方法调用，对需要清理的方法进行清理，close()方法针对整个任务调一次。

了解GenericUDTF抽象类之后，我们再来看下explode函数源码。

/**
 * GenericUDTFExplode.
 *
 */
@Description(name = "explode",
    value = "_FUNC_(a) - separates the elements of array a into multiple rows,"
      + " or the elements of a map into multiple rows and columns ")
public class GenericUDTFExplode extends GenericUDTF {

  private transient ObjectInspector inputOI = null;
  @Override
  public void close() throws HiveException {
  }

  @Override
  public StructObjectInspector initialize(ObjectInspector[] args) throws UDFArgumentException {
    if (args.length != 1) {
      throw new UDFArgumentException("explode() takes only one argument");
    }

    ArrayList<String> fieldNames = new ArrayList<String>();
    ArrayList<ObjectInspector> fieldOIs = new ArrayList<ObjectInspector>();

    switch (args[0].getCategory()) {
    case LIST:
      inputOI = args[0];
      fieldNames.add("col");
      fieldOIs.add(((ListObjectInspector)inputOI).getListElementObjectInspector());
      break;
    case MAP:
      inputOI = args[0];
      fieldNames.add("key");
      fieldNames.add("value");
      fieldOIs.add(((MapObjectInspector)inputOI).getMapKeyObjectInspector());
      fieldOIs.add(((MapObjectInspector)inputOI).getMapValueObjectInspector());
      break;
    default:
      throw new UDFArgumentException("explode() takes an array or a map as a parameter");
    }

    return ObjectInspectorFactory.getStandardStructObjectInspector(fieldNames,
        fieldOIs);
  }

  private transient final Object[] forwardListObj = new Object[1];
  private transient final Object[] forwardMapObj = new Object[2];

  @Override
  public void process(Object[] o) throws HiveException {
    switch (inputOI.getCategory()) {
    case LIST:
      ListObjectInspector listOI = (ListObjectInspector)inputOI;
      List<?> list = listOI.getList(o[0]);
      if (list == null) {
        return;
      }
      for (Object r : list) {
        forwardListObj[0] = r;
        forward(forwardListObj);
      }
      break;
    case MAP:
      MapObjectInspector mapOI = (MapObjectInspector)inputOI;
      Map<?,?> map = mapOI.getMap(o[0]);
      if (map == null) {
        return;
      }
      for (Entry<?,?> r : map.entrySet()) {
        forwardMapObj[0] = r.getKey();
        forwardMapObj[1] = r.getValue();
        forward(forwardMapObj);
      }
      break;
    default:
      throw new TaskExecutionException("explode() can only operate on an array or a map");
    }
  }

  @Override
  public String toString() {
    return "explode";
  }
}

有了对GenericUDTF抽象类的理解，就不难理解explode函数的源码了。由于explode函数可以把array数组或者映射map类型转成对行或者多列，所以会对数据类型进行判断，然后分别处理。有了以上源码经验，就可以尝试下自己去写一个自定义UDTF了。

四、总结

不管是hive还是mapreduece，都是比较复杂的一整块，我们在学习的时候不妨“浅尝辄止”。比如我们在继承这个类的时候，只需要关心它能实现什么功能、我们需要处理什么业务逻辑，而不去“深究”为什么它可以实现这样的功能。