Scala:RDD上的地图和平面图

时间:2016-08-30 09:49:43

标签: scala apache-spark

我有一个这种结构的RDD

     RDD[((String, String), List[(Int, Timestamp, String)])]

和数据

    ((D2,Saad Arif),List((4,2011-10-05 00:00:00.0,C101), (5,2010-01-27 00:00:00.0,C101)))
    ((D3,Faran Abid),List((7,2016-10-05 00:00:00.0,C101)))
    ((D1,Atif Shahzad),List((1,2012-04-15 00:00:00.0,C101), (2,2011-10-05 00:00:00.0,C101), (3,2006-12-25 00:00:00.0,C101)))

将此视为表格意味着

   '(D2,Saad Arif)' 

就像键和

    'List((4,2011-10-05 00:00:00.0,C101), (5,2010-01-27 00:00:00.0,C101)' 

就像这个键的行。 现在我想检查每一行,如果在两年或更长时间之前有代码'C101'的记录(历史),则将level设置为2,否则设置为1.因此生成的RDD应如下所示

((D2,Saad Arif),List((4,2011-10-05 00:00:00.0,C101, 1), (5,2010-01-27 00:00:00.0,C101, 1)))
((D3,Faran Abid),List((7,2016-10-05 00:00:00.0,C101, 1)))
((D1,Atif Shahzad),List((1,2012-04-15 00:00:00.0,C101, 2), (2,2011-10-05 00:00:00.0,C101, 2), (3,2006-12-25 00:00:00.0,C101, 1)))

注意时间戳后的新级别。如何使用地图或平面图进行此操作?

1 个答案:

答案 0 :(得分:1)

import java.time.LocalDate
import java.time.format.DateTimeFormatter
import java.time.Period    


val df1 = DateTimeFormatter.ofPattern("yyyy-MM-dd HH:mm:ss.S")

val futureDate = LocalDate.parse("2100-01-01 00:00:00.0", df1)

val yourRequiredRdd = yourRdd
  .map({
    case (t, list) => {
      val list1 = list.map({
        case (id, dateStr, id2) => (id, LocalDate.parse(dateStr, df1), id2) 
      })

      val oldestDate = list1
        .filter({ case (id, date, id2) => id2.equals("C101") })
        .map(_._2)
        .foldLeft(futureDate)((oldestDate, date) => {
          val period = Period.between(oldestDate, date)
          if (!period.isNegative()) oldestDate else date
        })

      val newList = list1
        .map({
          case (id, date, "C101") => {
            val periodFromOldestDate = Period.between(oldestDate, date)
            val extraNumber = if (periodFromOldestDate.getYears() >= 2) 2 else 1
            (id, date, "C101", extraNumber)
          }
          case (id, date, id2) => {
            (id, date, id2, 1)
          }
        })

      (t, newList)
    }
  })
  .flatMap({
    case ((pid, name), list) => list.map({
      case (id, date, code, level) => (id, name, code, pid, date, level)
    })
  })