Scala-使用Spark将JSON文件作为单个字符串读取

Question

I have JSON files describing a table structure. I want to read each file from S3 as a single String in order to then apply a fromJson() method of apache.spark.sql.types.DataType

DataType.fromJson(jsonString).asInstanceOf[StructType] 

But for now I only managed to read the files into a DataFrame:

 val testJsonData = sqlContext.read.option("multiline", "true").json("/s3Bucket/metrics/metric1.json")

But I don't need a df.schema , instead I need to parse the contents of a JSON string to a StructType.

The contents of a JSON file:

{
  "type" : "struct",
  "fields" : [ {
    "name" : "metric_name",
    "type" : "string",
    "nullable" : true,
    "metadata" : { }
  }, {
    "name" : "metric_time",
    "type" : "long",
    "nullable" : true,
    "metadata" : { }
  }, {
    "name" : "metric_value",
    "type" : "string",
    "nullable" : true,
    "metadata" : { }
  }]
}

Answer 1

It looks like what you want to use is sc.wholeTextFiles (sc is a SparkContext in this case).

This results in an RDD[(String, String)] where ._1 is the file name, and ._2 is the entire file content. Maybe you can try:

val files = sc.wholeTextFiles("/s3Bucket/metrics/", 16).toDS()
files.map(DataType.fromJson(_._2).asInstanceOf[StructType])

Which, in theory, would give you an Dataset[StructType] . Unfortunately, I'm not finding a similar function in the pure spark sql API, but this may work.