如何将我的java复杂对象转换为spark数据帧

Question

我正在使用 java spark，下面是我的代码

JavaRDD<MyComplexEntity> myObjectJavaRDD = resultJavaRDD.flatMap(result -> result.getMyObjects());

DataFrame df = sqlContext.createDataFrame(myObjectJavaRDD, MyComplexEntity.class);

df.saveAsParquetFile("s3a://mybucket/test.parquet");

MyComplexEntity.java

public MyComplexEntity implements Serializable {
     private Identifier identifier;
     private boolean isSwitch1True;
     private String note;
     private java.util.ArrayList<Identifier> secodaryIds;
     ......
} 

标识符.java

public Identifier implements Serializable {
     private int id;
     private String uuid;
     ......
}

问题是我在从 myObjectJavaRDD 创建数据帧时在第 2 步失败了。 如何将复杂的 java 对象列表转换为数据框。 谢谢

Answer 1

无论如何，您可以将其转换为 Scala 吗？

Scala 支持这种case class

对于您的情况，挑战是您有一个Seq/Array of Inner case 类为 => private java.util.ArrayList<Identifier> secodaryIds;

所以它可以通过以下方式完成

// inner case class Identifier
case class Identifier(Id : Integer , uuid : String)
val innerVal = Seq(Identifier(1,"gsgsg"),Identifier(2,"dvggwgwg"))

// Outer case class MyComplexEntity
case class MyComplexEntity(notes : String, identifierArray : Seq[Identifier])
val outerVal = MyComplexEntity("Hello", innerVal)

请注意=>

externalVal ：MyComplexEntity 包含一个标识符对象列表，如下所示

outerVal: MyComplexEntity = MyComplexEntity(Hello,List(Identifier(1,gsgsg), Identifier(2,dvggwgwg)))

现在使用Dataset的实际火花方式

import spark.implicits._
// Convert Our Input Data in Same Structure as your MyComplexEntity
// Only Trick is To 'Reflect' A Seq[(Int,String)] => Seq[Identifier]
// Hence we have to do 2 Mapping once for Outer Case class (MyComplexEntity) And Once For Inner Seq of Identifier
// If We Just Take this Input Data and Convert To DataSet ( without any Schema Inference)
// This is How It looks 

val inputData = Seq(("Some DAY",Seq((210,"wert67"),(310,"bill123"))),
                    ("I WILL BE", Seq((420,"henry678"),(1000,"baba123"))),
                    ("Saturday Night",Seq((1000,"Roger123"),(2000,"God345")))
                    )
                    
val unMappedDs = inputData.toDS

给我们=>

// See how it is Infered
// unMappedDs: org.apache.spark.sql.Dataset[(String, Seq[(Int, String)])] = [_1: string, _2: array<struct<_1:int,_2:string>>]

但是如果我们“正确地”映射它=>

As => // Second element is a Seq[(Int,String)] and We map it into Seq[Identifier] as x._2.map(y => Identifier(y._1,y._2))

像下面：

val resultDs = inputData.toDS.map(x =>MyComplexEntity(x._1,x._2.map(y => Identifier(y._1,y._2))))
resultDs.show(20,false)

我们得到一个结构，如=>

resultDs: org.apache.spark.sql.Dataset[MyComplexEntity] = [notes: string, identifierArray: array<struct<Id:int,uuid:string>>]

和数据为：

+--------------+--------------------------------+
|notes         |identifierArray                 |
+--------------+--------------------------------+
|Some DAY      |[[210,wert67], [310,bill123]]   |
|I WILL BE     |[[420,henry678], [1000,baba123]]|
|Saturday Night|[[1000,Roger123], [2000,God345]]|
+--------------+--------------------------------+

使用Scala很容易。 谢谢。