在 Spark 数据帧中分解嵌套结构

Question

我正在研究 Databricks 示例。数据框的架构如下所示：

> parquetDF.printSchema
root
|-- department: struct (nullable = true)
|    |-- id: string (nullable = true)
|    |-- name: string (nullable = true)
|-- employees: array (nullable = true)
|    |-- element: struct (containsNull = true)
|    |    |-- firstName: string (nullable = true)
|    |    |-- lastName: string (nullable = true)
|    |    |-- email: string (nullable = true)
|    |    |-- salary: integer (nullable = true)

在示例中，他们展示了如何将 employees 列分解为 4 个额外的列：

val explodeDF = parquetDF.explode($"employees") { 
case Row(employee: Seq[Row]) => employee.map{ employee =>
  val firstName = employee(0).asInstanceOf[String]
  val lastName = employee(1).asInstanceOf[String]
  val email = employee(2).asInstanceOf[String]
  val salary = employee(3).asInstanceOf[Int]
  Employee(firstName, lastName, email, salary)
 }
}.cache()
display(explodeDF)

我如何对部门列做类似的事情（即向数据框添加两个额外的列，称为“id”和“name”）？这些方法并不完全相同，我只能弄清楚如何使用以下方法创建全新的数据框：

val explodeDF = parquetDF.select("department.id","department.name")
display(explodeDF)

如果我尝试：

val explodeDF = parquetDF.explode($"department") { 
  case Row(dept: Seq[String]) => dept.map{dept => 
  val id = dept(0) 
  val name = dept(1)
  } 
}.cache()
display(explodeDF)

我收到警告和错误：

<console>:38: warning: non-variable type argument String in type pattern Seq[String] is unchecked since it is eliminated by erasure
            case Row(dept: Seq[String]) => dept.map{dept => 
                           ^
<console>:37: error: inferred type arguments [Unit] do not conform to    method explode's type parameter bounds [A <: Product]
  val explodeDF = parquetDF.explode($"department") { 
                                   ^

Answer 1

你可以使用类似的东西：

var explodeDF = explodeDF.withColumn("id", explodeDF("department.id"))
explodeDeptDF = explodeDeptDF.withColumn("name", explodeDeptDF("department.name"))

你帮我解决了哪些问题以及这些问题：

Answer 2

这似乎可行（尽管可能不是最优雅的解决方案）。

var explodeDF2 = explodeDF.withColumn("id", explodeDF("department.id"))
explodeDF2 = explodeDF2.withColumn("name", explodeDF2("department.name"))

Answer 3

在我看来，最优雅的解决方案是使用 select 运算符对 Struct 进行星形扩展，如下所示：

var explodedDf2 = explodedDf.select("department.*","*")

https://docs.databricks.com/spark/latest/spark-sql/complex-types.html

在 Spark 数据帧中分解嵌套结构

Exploding nested Struct in Spark dataframe

scala

distributed-computing

apache-spark

apache-spark-sql

databricks