如何在不知道spark scala中json列的模式的情况下动态解析数据帧中的json列

给定一个spark数据帧，该数据帧的列可能包含也可能不包含嵌套的json。这个嵌套的json是动态的。最终的要求是打破json，并为嵌套json中的每个键生成一个新的数据帧，其中包含新的列

json是动态的，因此生成的表是动态的。还请考虑数据文件由1亿个以上的记录组成。 乙二醇-

输入

------------------------------------------------------------------------
|id         |key     |type      |value 

|f9f          |BUSI    |off         |false                                                                                                                                                                                                          
|f96          |NAME    |50          |true                                                                                                                                                                                                           
|f9z          |BANK    |off         |{"Name":"United School","admNumber":"197108","details":{"code":"WEREFFW32","studentName":"Abhishek kumar","doc":"certificate","admId":"3424325328","stat":0,"studentDetails":false} }|

输出：-

--------------------------------------------------------------------------------------------------------------------------
|id   |key    |type     |value  |Name    | admNumber   |code    | studentName  | doc   |admId    |stat   | studentDetails
+------------------------------------+-----------------+-------------+----------------------------------------------------                              

|f9f  |BUSI    |off     |false  |NULL    |NULL         |NULL    |NULL          |NULL   |NULL     |NULL   |NULL |                                            
|f96  |NAME    |50      |true   |NULL    |NULL         |NULL    |NULL          |NULL   |NULL     |NULL   |NULL            |                                                                                                                
|f9z  |BANK    |off     |NULL   |United School    |197108         |WEREFFW32    |Abhishek kumar          |certificate   |3424325328     |0   |false |    

---

初始数据帧是否只有一个json列？是否希望所有嵌套键都成为列？（展平？）给出一个输入和输出示例。如果您不知道模式，将对性能产生很大影响。或者，如果您在某种程度上可以轻松提高性能之前就知道模式。@ziad.rida是的，初始数据帧有一个json列，所有嵌套键都需要成为列。嵌套json的深度不会超过3或4级。这是否也适用于嵌套json？

  val data = Seq(
    (77, "email1", """{"key1":38,"key3":39}"""),
    (78, "email2", """{"key1":38,"key4":39}"""),
    (178, "email21", """{"key1":"when string","key4":36, "key6":"test", "key10":false }"""),
    (179, "email8", """{"sub1":"qwerty","sub2":["42"]}"""),
      (180, "email8", """{"sub1":"qwerty","sub2":["42", "56", "test"]}""")
  ).toDF("id", "name", "colJson")

  data.show(false)
//  +---+-------+---------------------------------------------------------------+
//  |id |name   |colJson                                                        |
//  +---+-------+---------------------------------------------------------------+
//  |77 |email1 |{"key1":38,"key3":39}                                          |
//  |78 |email2 |{"key1":38,"key4":39}                                          |
//  |178|email21|{"key1":"when string","key4":36, "key6":"test", "key10":false }|
//  |178|email8 |{"sub1":"qwerty","sub2":"42"}                                  |
//  +---+-------+---------------------------------------------------------------+


  val schema = spark.read.json(data.select("colJson").as[String]).schema
  val res = data.select($"id", $"name", from_json($"colJson", schema).as("s")).select("id", "name", "s.*")
  res.show(false)
//  +---+-------+-----------+-----+----+----+----+------+----+
//  |id |name   |key1       |key10|key3|key4|key6|sub1  |sub2|
//  +---+-------+-----------+-----+----+----+----+------+----+
//  |77 |email1 |38         |null |39  |null|null|null  |null|
//  |78 |email2 |38         |null |null|39  |null|null  |null|
//  |178|email21|when string|false|null|36  |test|null  |null|
//  |178|email8 |null       |null |null|null|null|qwerty|42  |
//  +---+-------+-----------+-----+----+----+----+------+----+

  val  df1 = res.filter('sub1.equalTo("qwerty"))
  df1.show(false)
//  +---+------+----+-----+----+----+----+------+----+
//  |id |name  |key1|key10|key3|key4|key6|sub1  |sub2|
//  +---+------+----+-----+----+----+----+------+----+
//  |178|email8|null|null |null|null|null|qwerty|42  |
//  +---+------+----+-----+----+----+----+------+----+