Array of JSON to Dataframe in Spark received by Kafka

问题

I'm writing a Spark application in Scala using Spark Structured Streaming that receive some data formatted in JSON style from Kafka. This application could receive both a single or multiple JSON object formatted in this way:

[{"key1":"value1","key2":"value2"},{"key1":"value1","key2":"value2"},...,{"key1":"value1","key2":"value2"}]

I tried to define a StructType like:

var schema = StructType(
                  Array(
                        StructField("key1",DataTypes.StringType),
                        StructField("key2",DataTypes.StringType)
             ))

But it doesn't work. My actual code for parsing JSON:

var data = (this.stream).getStreamer().load()
  .selectExpr("CAST (value AS STRING) as json")
  .select(from_json($"json",schema=schema).as("data"))

I would like to get this JSON objects in a dataframe like

+----------+---------+
|      key1|     key2|
+----------+---------+
|    value1|   value2|
|    value1|   value2|
        ........
|    value1|   value2|
+----------+---------+

Anyone can help me please? Thank you!

回答1:

As your incoming string is Array of JSON, one way is to write a UDF to parse the Array, then explode the parsed Array. Below is the complete code with each steps explained. I have written it for batch but same can be used for streaming with minimal changes.

object JsonParser{

  //case class to parse the incoming JSON String
  case class JSON(key1: String, key2: String)

  def main(args: Array[String]): Unit = {
    val spark = SparkSession.
      builder().
      appName("JSON").
      master("local").
      getOrCreate()

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

    //sample JSON array String coming from kafka
    val str = Seq("""[{"key1":"value1","key2":"value2"},{"key1":"value3","key2":"value4"}]""")

    //UDF to parse JSON array String
    val jsonConverter = udf { jsonString: String =>
      val mapper = new ObjectMapper()
      mapper.registerModule(DefaultScalaModule)
      mapper.readValue(jsonString, classOf[Array[JSON]])
    }

    val df = str.toDF("json") //json String column
      .withColumn("array", jsonConverter($"json")) //parse the JSON Array
      .withColumn("json", explode($"array")) //explode the Array
      .drop("array") //drop unwanted columns
      .select("json.*") //explode the JSON to separate columns

    //display the DF
    df.show()
    //+------+------+
    //|  key1|  key2|
    //+------+------+
    //|value1|value2|
    //|value3|value4|
    //+------+------+

  }
}

回答2:

This worked fine for me in Spark 3.0.0 and Scala 2.12.10. I used schema_of_json to get the schema of the data in a suitable format for from_json, and applied explode and * operator in the last step of the chain to expand accordingly.

// TO KNOW THE SCHEMA
scala> val str = Seq("""[{"key1":"value1","key2":"value2"},{"key1":"value3","key2":"value4"}]""")
str: Seq[String] = List([{"key1":"value1","key2":"value2"},{"key1":"value3","key2":"value4"}])

scala> val df = str.toDF("json")
df: org.apache.spark.sql.DataFrame = [json: string]

scala> df.show()
+--------------------+
|                json|
+--------------------+
|[{"key1":"value1"...|
+--------------------+

scala> val schema = df.select(schema_of_json(df.select(col("json")).first.getString(0))).as[String].first
schema: String = array<struct<key1:string,key2:string>>

Use the resulting string as your schema: 'array<structkey1:string,key2:string>', as follows:

// TO PARSE THE ARRAY OF JSON's
scala> val parsedJson1 = df.selectExpr("from_json(json, 'array<struct<key1:string,key2:string>>') as parsed_json")
parsedJson1: org.apache.spark.sql.DataFrame = [parsed_json: array<struct<key1:string,key2:string>>]

scala> parsedJson1.show()
+--------------------+
|         parsed_json|
+--------------------+
|[[value1, value2]...|
+--------------------+

scala> val data = parsedJson1.selectExpr("explode(parsed_json) as json").select("json.*")
data: org.apache.spark.sql.DataFrame = [key1: string, key2: string]

scala> data.show()
+------+------+
|  key1|  key2|
+------+------+
|value1|value2|
|value3|value4|
+------+------+

Just FYI, without the star expansion the intermediate result looks as follows:

scala> val data = parsedJson1.selectExpr("explode(parsed_json) as json")
data: org.apache.spark.sql.DataFrame = [json: struct<key1: string, key2: string>]

scala> data.show()
+----------------+
|            json|
+----------------+
|[value1, value2]|
|[value3, value4]|
+----------------+

来源：https://stackoverflow.com/questions/53800906/array-of-json-to-dataframe-in-spark-received-by-kafka