如何在 spark sql 中合并地图列？

Question

我在 Dataframe 中有两个 Map 类型的列。有没有一种方法可以创建一个新的 Map 列，使用 .withColumn 在 spark Sql 中合并这两列？

val sampleDF = Seq(
 ("Jeff", Map("key1" -> "val1"), Map("key2" -> "val2"))
).toDF("name", "mapCol1", "mapCol2")

sampleDF.show()

+----+-----------------+-----------------+
|name|          mapCol1|          mapCol2|
+----+-----------------+-----------------+
|Jeff|Map(key1 -> val1)|Map(key2 -> val2)|
+----+-----------------+-----------------+

Answer 1

您可以使用withColumn 编写一个udf 函数将两列合并为一个，如下所示

import org.apache.spark.sql.functions._
def mergeUdf = udf((map1: Map[String, String], map2: Map[String, String])=> map1 ++ map2)

sampleDF.withColumn("merged", mergeUdf(col("mapCol1"), col("mapCol2"))).show(false)

这应该给你

+----+-----------------+-----------------+-------------------------------+
|name|mapCol1          |mapCol2          |merged                         |
+----+-----------------+-----------------+-------------------------------+
|Jeff|Map(key1 -> val1)|Map(key2 -> val2)|Map(key1 -> val1, key2 -> val2)|
+----+-----------------+-----------------+-------------------------------+

希望回答对你有帮助

Answer 2

仅当由于性能原因您的用例没有内置函数时才使用 UDF。

Spark 2.4 及以上版本

import org.apache.spark.sql.functions.{map_concat, col}

sampleDF.withColumn("map_concat", map_concat(col("mapCol1"), col("mapCol2"))).show(false)

输出

+----+-----------------+-----------------+-------------------------------+
|name|mapCol1          |mapCol2          |map_concat                     |
+----+-----------------+-----------------+-------------------------------+
|Jeff|Map(key1 -> val1)|Map(key2 -> val2)|Map(key1 -> val1, key2 -> val2)|
+----+-----------------+-----------------+-------------------------------+

Spark 版本 2.4 以下

按照 @RameshMaharjan answer in this question 创建一个 UDF，但我添加了一个空检查以避免在运行时出现 NPE，如果不添加最终会导致作业失败。

import org.apache.spark.sql.functions.{udf, col}

val map_concat = udf((map1: Map[String, String],
                      map2: Map[String, String]) =>
  if (map1 == null) {
    map2
  } else if (map2 == null) {
    map1
  } else {
    map1 ++ map2
  })

sampleDF.withColumn("map_concat", map_concat(col("mapCol1"), col("mapCol2")))
 .show(false)

Answer 3

你可以使用struct来实现。

val sampleDF = Seq(
 ("Jeff", Map("key1" -> "val1"), Map("key2" -> "val2"))
).toDF("name", "mapCol1", "mapCol2")

sampleDF.show()

+----+-----------------+-----------------+
|name|          mapCol1|          mapCol2|
+----+-----------------+-----------------+
|Jeff|Map(key1 -> val1)|Map(key2 -> val2)|
+----+-----------------+-----------------+

sampleDF.withColumn("NewColumn",struct(sampleDF("mapCol1"), sampleDF("mapCol2"))).take(2)
    res17: Array[org.apache.spark.sql.Row] = Array([Jeff,Map(key1 -> val1),Map(key2 -> val2),[Map(key1 -> val1),Map(key2 -> val2)]])

+----+-----------------+-----------------+--------------------+
|name|          mapCol1|          mapCol2|           NewColumn|
+----+-----------------+-----------------+--------------------+
|Jeff|Map(key1 -> val1)|Map(key2 -> val2)|[Map(key1 -> val1...|
+----+-----------------+-----------------+--------------------+

参考：How to merge two columns of a `Dataframe` in Spark into one 2-Tuple?