Question

我需要更新delta数据的数据帧的行号列。我已经实现了基本加载的行号，如下所示：

输入数据：

val base = List(List("001", "a", "abc"), List("001", "a", "123"),List("003", "c", "456") ,List("002", "b", "dfr"), List("003", "c", "ytr"))
  .map(row => (row(0), row(1), row(2)))

val DS1 = base.toDF("KEY1", "KEY2" ,"VAL")

DS1.show()

+----+----+---+
|KEY1|KEY2|VAL|
+----+----+---+
| 001|   a|abc|
| 001|   a|123|
| 003|   c|456|
| 002|   b|dfr|
| 003|   c|ytr|
+----+----+---+

现在我使用窗口函数添加了行号，如下所示：

val baseDF =  DS1.select(col("KEY1"), col("KEY2"), col("VAL") ,row_number().over(Window.partitionBy(col("KEY1"), col("KEY2")).orderBy(col("KEY1"), col("KEY2").asc)).alias("Row_Num"))
baseDF.show()

+----+----+---+-------+
|KEY1|KEY2|VAL|Row_Num|
+----+----+---+-------+
|001 |a   |abc|1      |
|001 |a   |123|2      |
|002 |b   |dfr|1      |
|003 |c   |456|1      |
|003 |c   |ytr|2      |
+----+----+---+-------+

现在增量负荷如下：

val delta = List(List("001", "a", "y45") ,List("002", "b", "444"))
  .map(row => (row(0), row(1), row(2)))

val DS2 = delta.toDF("KEY1", "KEY2" ,"VAL")
DS2.show()

+----+----+---+
|KEY1|KEY2|VAL|
+----+----+---+
| 001|   a|y45|
| 002|   b|444|
+----+----+---+

所以预期的更新结果应该是：

baseDF.show()

|KEY1|KEY2|VAL|Row_Num|
+----+----+---+-------+
|001 |a   |abc|1      |
|001 |a   |123|2      |
| 001|   a|y45|3      | -----> Delta record
|002 |b   |dfr|1      |
| 002|   b|444|2      | -----> Delta record 
|003 |c   |456|1      |
|003 |c   |ytr|2      |
+----+----+---+-------+

使用数据框/数据集实施此解决方案的任何建议？我们能用spark rdd的zipWithIndex？

来实现上述解决方案

Answer 1

使用更新的行号添加增量的一种方法是：1）在Row_Num，2）union DS2中添加一个带有大数字的列baseDF，以及3）计算新的行号，如下所示：

import org.apache.spark.sql.functions._
import org.apache.spark.sql.expressions.Window

val combinedDF = baseDF.union(
  DS2.withColumn("Row_Num", lit(Long.MaxValue))
)

val resultDF = combinedDF.select(
  col("KEY1"), col("KEY2"), col("VAL"), row_number().over(
    Window.partitionBy(col("KEY1"), col("KEY2")).orderBy(col("Row_Num"))
  ).alias("New_Row_Num")
)

resultDF.show
+----+----+---+-----------+
|KEY1|KEY2|VAL|New_Row_Num|
+----+----+---+-----------+
| 003|   c|456|          1|
| 003|   c|ytr|          2|
| 002|   b|dfr|          1|
| 002|   b|444|          2|
| 001|   a|abc|          1|
| 001|   a|123|          2|
| 001|   a|y45|          3|
+----+----+---+-----------+

为Delta Data更新Spark Dataframe的窗口函数row_number列

1 个答案: