编写scala版hive的自定义函数

背景

最近，在数仓使用过程中，遇到一些数据处理过程中需要特殊处理，然而hive内置的函数确无法满足需求，因此需要通过自定义函数来扩充其内置函数，达到便捷处理数据的最终目的。hive支持Java版和Scala版自定义函数，本文要介绍的是编写scala版的自定义函数。

自定义函数类型

类型	特点
UDF	输入输出是一对一
UDTF	输入输出是一对多
UDAF	输入输出是多对一

编写UDF

UDF的编写需要从继承hive的特定类开始，目前支持两种方式，方式一编写简单，但仅支持操作基本数据类型，而方法二编写略微复杂，但支持所有数据类型，包括复杂类型，比如List，Map等，本文采用方式二，以输入输出数据类型均为List为例。
方式一：继承org.apache.hadoop.hive.ql.exec.UDF
方式二：继承org.apache.hadoop.hive.ql.udf.generic.GenericUDF

1. pom文件中引入以下依赖

    <dependencies>
        <dependency>
            <groupId>org.scala-lang</groupId>
            <artifactId>scala-library</artifactId>
            <version>${scala.version}</version>
            <scope>provided</scope>
        </dependency>
        <dependency>
            <groupId>org.apache.hive</groupId>
            <artifactId>hive-exec</artifactId>
            <version>3.1.1</version>
            <scope>provided</scope>
        </dependency>
    </dependencies>

2. 用scala语言编写UDF（List输入，基本类型输出）

@Description(name = "udf_name", value = "_FUNC_(arg) - description of function.")
class XXX extends GenericUDF {
  @transient private var listOI: ListObjectInspector = _
  @transient private var elementOI: BinaryObjectInspector = _

  //输入数据类型校验，并指定输出数据类型
  @throws(classOf[UDFArgumentException])
  override def initialize(arguments: Array[ObjectInspector]): ObjectInspector = {
    if (arguments.length != 1)
      throw new UDFArgumentLengthException("The operator 'udf_name' accepts 1 arguments, " +
        s"but get ${arguments.length}")

    val arg01 = arguments(0)

    if (!arg01.isInstanceOf[ListObjectInspector])
      throw new UDFArgumentTypeException(0, s"The argument type of function 'udf_name' should be list, " +
        s"but get ${arg01.getTypeName}")

    listOI = arg01.asInstanceOf[ListObjectInspector]
    val elementType = listOI.getListElementObjectInspector

    if (!elementType.isInstanceOf[BinaryObjectInspector])
      throw new UDFArgumentTypeException(0, s"The element type of list should be binary，but get ${elementType.getTypeName}")

    elementOI = elementType.asInstanceOf[BinaryObjectInspector]

    PrimitiveObjectInspectorFactory.writableBinaryObjectInspector
  }

  //具体执行逻辑
  override def evaluate(arguments: Array[GenericUDF.DeferredObject]): AnyRef = {
    val arg0 = arguments(0).get()
    val arg0_length = listOI.getListLength(arg0)

    if (arg0_length == -1) return null

    val wkbs = new util.ArrayList[Array[Byte]](arg0_length)
    var i = 0
    while (i < arg0_length) {
      wkbs.add(i, elementOI.getPrimitiveJavaObject(listOI.getListElement(arg0, i)))
      i += 1
    }

    new BytesWritable(todo(wkbs))
  }
  
  override def getDisplayString(children: Array[String]): String = {
    getStandardDisplayString("udf_name", children)
  }
}

3. 用scala语言编写UDF（基本类型输入，List输出）

@Description(name = "udf_name", value = "_FUNC_(arg1, arg2) - description of function.")
class GridInternalGrids extends GenericUDF{
  @transient private var stringOI: StringObjectInspector = _
  @transient private var intOI: IntObjectInspector = _

  @throws(classOf[UDFArgumentException])
  override def initialize(arguments: Array[ObjectInspector]): ObjectInspector = {
    if (arguments.length != 2)
      throw new UDFArgumentLengthException(s"The operator 'udf_name' accept 2 arguments, " +
        s"but get ${arguments.length}")
    val arg01 = arguments(0).asInstanceOf[PrimitiveObjectInspector]
    val arg02 = arguments(1).asInstanceOf[PrimitiveObjectInspector]
    if (arg01.getPrimitiveCategory != PrimitiveCategory.STRING)
      throw new UDFArgumentTypeException(0, s"The 1st arg of function 'udf_name' should be string, " +
        s"but get ${arg01.getTypeName}")
    if (arg02.getPrimitiveCategory != PrimitiveCategory.INT)
      throw new UDFArgumentTypeException(1, s"The 2nd arg of function 'udf_name' should be int, " +
        s"but get ${arg02.getTypeName}")
    stringOI = arg01.asInstanceOf[StringObjectInspector]
    intOI = arg02.asInstanceOf[IntObjectInspector]
    ObjectInspectorFactory.getStandardListObjectInspector(PrimitiveObjectInspectorFactory.writableStringObjectInspector)
  }

  @throws(classOf[UDFArgumentException])
  override def evaluate(arguments: Array[GenericUDF.DeferredObject]): AnyRef = {
    val arg01 = stringOI.getPrimitiveJavaObject(arguments(0).get())
    val arg02 = intOI.get(arguments(1).get())

    val result = new util.ArrayList[Text]()
    val temp_array = todo(arg01, arg02)
    var index = 0
    while (index < temp_array.length) {
      result.add(new Text(temp_array(index)))
      index += 1
    }
    result
  }

  override def getDisplayString(children: Array[String]): String = {
    getStandardDisplayString("udf_name", children)
  }
}

• UDTF，UDTF待写

自定义函数开发中异常处理

1. 继承hive的UDF类后，必须用scala的class定义子类，而非object类
2. 方法内用到split方法时，默认会对结果做隐式转换，这会导致udf运行时，
   触发报错running query: java.lang.NoSuchMethodError: scala.Predef，
   解决方案是import scala.Predef.{refArrayOps => _}，禁止做隐式转换
3. UDFArgumentTypeException第一个参数注意从0开始，否则无法正常抛异常
4. running query: java.lang.BootstrapMethodError: java.lang.NoClassDefFoundError: scala/runtime/java8/JFunction1$mcVI$sp (state=,code=0)
   循环遍历数组只能用while

部署

1. 上传jar包至hdfs

hdfs dfs -put -f /location_path/xx-xxx-xx.jar /hdfs_path/hive

2. 登录hive客户端beeline
3. 永久注册方法

create function udf_name as 'com.xxx.hive.xxx.Encode' using jar 'hdfs://xxx/hdfs_path/hive/xx-xxx-xx.jar';

4. 使用

select id, udf_name(arg01,arg02,arg03) as col_name from db_name.table_name limit 10;