spark的rdd是一个二维表，二维表可以直接写入mongo，但mongo可以更灵活方便，可以存储分层结构的json类型数据。

比如rdd二维表：

[
{student:"张三",subject:"语文",score:"90"},
{student:"张三",subject:"数学",score:"90"}
]

期望得到层次结构：

[
{student:"张三",score:{语文:"90",数学:"90"}}
]

spark dataframe 二维表将列转为map类型：

def toJsonRdd(df:DataFrame,groupField:String,attributeField:Seq[String],valueField:Map[String,String]) ={
    var resultDf=df

    var currentGroupFields:Seq[String]=attributeField.+:(groupField)

    while(currentGroupFields.length>1){
      val currentAttributeName=currentGroupFields.last
      currentGroupFields = currentGroupFields.dropRight(1)

      var valueColumn:Seq[Column]=Seq.empty
      for( (name,alias) <- valueField){
        valueColumn=valueColumn:+
          map_from_entries(
            collect_list(struct(col(currentAttributeName),col(name)))
          ).as(name)
      }
      resultDf=resultDf.groupBy(currentGroupFields.map(col(_)):_*).agg(
        valueColumn(0),
        valueColumn.drop(1):_*
      )
    }

    for( (name,alias) <- valueField){
      if(StringUtils.isNotEmpty(alias)){
        resultDf=resultDf.withColumnRenamed(name,alias)
      }
    }

    resultDf
  }