1

我有一个方法，它以 Parquet 形式编写我的一个类Foo，它被定义为 Thrift。

  import Foo
  import org.apache.spark.rdd.RDD
  import org.apache.thrift.TBase
  import org.apache.hadoop.mapreduce.Job
  import org.apache.parquet.hadoop.ParquetOutputFormat
  import org.apache.parquet.hadoop.thrift.ParquetThriftOutputFormat

  def writeThriftParquet(rdd: RDD[Foo], outputPath: String): Unit = {
    val job = Job.getInstance()
    ParquetThriftOutputFormat.setThriftClass(job, classOf[Foo])
    ParquetOutputFormat.setWriteSupportClass(job, classOf[Foo])

    rdd
      .map(x => (null, x))
      .saveAsNewAPIHadoopFile(
        outputPath,
        classOf[Void],
        classOf[Foo],
        classOf[ParquetThriftOutputFormat[Foo]],
        job.getConfiguration)
  }

这很好用，但我更愿意编写一个更通用的方法。我尝试了（相对）简单的：

  def writeThriftParquetGeneral[A <: TBase[_, _]](rdd: RDD[A], outputPath: String): Unit = {
    val job = Job.getInstance()
    ParquetThriftOutputFormat.setThriftClass(job, classOf[A])
    ParquetOutputFormat.setWriteSupportClass(job, classOf[A])

    rdd
      .map(x => (null, x))
      .saveAsNewAPIHadoopFile(
        outputPath,
        classOf[Void],
        classOf[A],
        classOf[ParquetThriftOutputFormat[A]],
        job.getConfiguration)
  }

但失败并出现以下错误：

 class type required but A found ParquetThriftOutputFormat.setThriftClass(job, classOf[A])
 class type required but A found ParquetOutputFormat.setWriteSupportClass(job, classOf[A])

为了解决这个问题，我使用了 a ClassTag，但还没有编译。

  import scala.reflect._
  implicit val ct = ClassTag[Foo](classOf[Foo])

  def writeThriftParquetGeneral[A <: TBase[_, _]](rdd: RDD[A], outputPath: String)(
    implicit tag: ClassTag[A]): Unit = {
    val job = Job.getInstance()

    // The problem line
    ParquetThriftOutputFormat.setThriftClass(job, tag.runtimeClass)

    // Seems OK from here
    ParquetOutputFormat.setWriteSupportClass(job, tag.runtimeClass)

    rdd
      .map(x => (null, x))
      .saveAsNewAPIHadoopFile(
        outputPath,
        classOf[Void],
        tag.runtimeClass,
        classOf[ParquetThriftOutputFormat[A]],
        job.getConfiguration)
  }

这失败了：ParquetThriftOutputFormat.setThriftClass(job, tag.runtimeClass)

[error]  found   : Class[_$1] where type _$1
[error]  required: Class[_ <: org.apache.thrift.TBase[_, _]]

我很惊讶编译器（Scala 2.11）没有认识到tag.runtimeClass必须是 a classOf[A]，并且A满足定义绑定的类型。

4

1 回答 1

2

ClassTag#runtimeClass只返回一个Class[_]

https://github.com/scala/scala/blob/2.13.x/src/library/scala/reflect/ClassTag.scala#L55

Class[_ <: TBase[_, _]]是一种不同于Class[_]（实际上是它的子类型）的存在类型

implicitly[Class[_ <: TBase[_, _]] <:< Class[_]]

尝试将问题行替换为

ParquetThriftOutputFormat.setThriftClass(job, classTag.runtimeClass.asSubclass(classOf[TBase[_, _]]))
于 2020-01-25T23:28:16.943 回答