在Spark SQL中使用循环引用处理模型吗?

时间:2018-09-22 07:48:10

标签: scala apache-spark apache-spark-sql apache-spark-dataset

昨天(实际上是完整的日记),我试图找出一种优雅的方式来表示带有 Scala / Spark SQL 2.2.1 的循环引用的模型。

让我们说这是原始的模型方法,它当然不起作用(请记住,实际模型具有数十个属性):

case class Branch(id: Int, branches: List[Branch] = List.empty)
case class Tree(id: Int, branches: List[Branch])

val trees = Seq(Tree(1, List(Branch(2, List.empty), Branch(3, List(Branch(4, List.empty))))))

val ds = spark.createDataset(trees)
ds.show

这是它引发的错误:

java.lang.UnsupportedOperationException: cannot have circular references in class, but got the circular reference of class Branch

我知道我们拥有的最高等级为5 。因此,作为一种解决方法,我虽然使用以下方法:

case class BranchLevel5(id: Int)
case class BranchLevel4(id: Int, branches: List[BranchLevel5] = List.empty)
case class BranchLevel3(id: Int, branches: List[BranchLevel4] = List.empty)
case class BranchLevel2(id: Int, branches: List[BranchLevel3] = List.empty)
case class BranchLevel1(id: Int, branches: List[BranchLevel2] = List.empty)
case class Tree(id: Int, branches: List[BranchLevel1])

当然可以了。但这一点都不优雅,您可以想象实现过程中的痛苦(可读性,耦合性,维护性,可用性,代码重复性等)

问题是,如何处理模型中带有循环引用的案例?

1 个答案:

答案 0 :(得分:0)

如果您可以使用私有API,那么我发现了一种可行的方法:将整个自引用结构视为用户定义的类型。我正在遵循以下答案中的方法:https://stackoverflow.com/a/51957666/1823254

package org.apache.spark.custom.udts // we're calling some private API so need to be under 'org.apache.spark'

import java.io._
import org.apache.spark.sql.types.{DataType, UDTRegistration, UserDefinedType}

class BranchUDT extends UserDefinedType[Branch] {

  override def sqlType: DataType = org.apache.spark.sql.types.BinaryType
  override def serialize(obj: Branch): Any = {
    val bos = new ByteArrayOutputStream()
    val oos = new ObjectOutputStream(bos)
    oos.writeObject(obj)
    bos.toByteArray
  }
  override def deserialize(datum: Any): Branch = {
    val bis = new ByteArrayInputStream(datum.asInstanceOf[Array[Byte]])
    val ois = new ObjectInputStream(bis)
    val obj = ois.readObject()
    obj.asInstanceOf[Branch]
  }

  override def userClass: Class[Branch] = classOf[Branch]
}

object BranchUDT {
  def register() = UDTRegistration.register(classOf[Branch].getName, classOf[BranchUDT].getName)
}

只需创建并注册一个自定义的UDT,就是这样!

BranchUDT.register()
val trees = Seq(Tree(1, List(Branch(2, List.empty), Branch(3, List(Branch(4, List.empty))))))

val ds = spark.createDataset(trees)
ds.show(false)

//+---+----------------------------------------------------+
//|id |branches                                            |
//+---+----------------------------------------------------+
//|1  |[Branch(2,List()), Branch(3,List(Branch(4,List())))]|
//+---+----------------------------------------------------+