create a spark dataframe from a nested json file i

问题:

This question already has an answer here:

• How to access sub-entities in JSON file? 1 answer

I have a json file that looks like this

{
"group" : {},
"lang" : [ 
    [ 1, "scala", "functional" ], 
    [ 2, "java","object" ], 
    [ 3, "py","interpreted" ]
]
}

I tried to create a dataframe using

val path = "some/path/to/jsonFile.json"
val df = sqlContext.read.json(path)
df.show()

when I run this I get

df: org.apache.spark.sql.DataFrame = [_corrupt_record: string]

How do we create a df based on contents of "lang" key? I do not care about group{} all I need is, pull data out of "lang" and apply case class like this

case class ProgLang (id: Int, lang: String, type: String )

I have read this post Reading JSON with Apache Spark - `corrupt_record` and understand that each record needs to be on a newline but in my case I cannot change the file structure

回答1:

The json format is wrong. The the json api of sqlContext is reading it as corrupt record. Correct form is

{"group":{},"lang":[[1,"scala","functional"],[2,"java","object"],[3,"py","interpreted"]]}

and supposing you have it in a file ("/home/test.json"), then you can use following method to get the dataframe you want

import org.apache.spark.sql.functions._
import sqlContext.implicits._

val df = sqlContext.read.json("/home/test.json")

val df2 = df.withColumn("lang", explode($"lang"))
    .withColumn("id", $"lang"(0))
    .withColumn("langs", $"lang"(1))
    .withColumn("type", $"lang"(2))
    .drop("lang")
    .withColumnRenamed("langs", "lang")
    .show(false)

You should have

+---+-----+-----------+
|id |lang |type       |
+---+-----+-----------+
|1  |scala|functional |
|2  |java |object     |
|3  |py   |interpreted|
+---+-----+-----------+

Updated

If you don't want to change your input json format as mentioned in your comment below, you can use wholeTextFiles to read the json file and parse it as below

import sqlContext.implicits._
import org.apache.spark.sql.functions._

val readJSON = sc.wholeTextFiles("/home/test.json")
  .map(x => x._2)
  .map(data => data.replaceAll("\n", ""))

val df = sqlContext.read.json(readJSON)

val df2 = df.withColumn("lang", explode($"lang"))
  .withColumn("id", $"lang"(0).cast(IntegerType))
  .withColumn("langs", $"lang"(1))
  .withColumn("type", $"lang"(2))
  .drop("lang")
  .withColumnRenamed("langs", "lang")

df2.show(false)
df2.printSchema

It should give you dataframe as above and schema as

root
 |-- id: integer (nullable = true)
 |-- lang: string (nullable = true)
 |-- type: string (nullable = true)

回答2:

As of Spark 2.2 you can use multiLine option to deal with the case of multi-line JSONs.

scala> spark.read.option("multiLine", true).json("jsonFile.json").printSchema
root
 |-- lang: array (nullable = true)
 |    |-- element: array (containsNull = true)
 |    |    |-- element: string (containsNull = true)

Before Spark 2.2 see How to access sub-entities in JSON file? or Read multiline JSON in Apache Spark.