左反加入Spark？

Question

我已经定义了两个这样的表:

 val tableName = "table1"
    val tableName2 = "table2"

    val format = new SimpleDateFormat("yyyy-MM-dd")
      val data = List(
        List("mike", 26, true),
        List("susan", 26, false),
        List("john", 33, true)
      )
    val data2 = List(
        List("mike", "grade1", 45, "baseball", new java.sql.Date(format.parse("1957-12-10").getTime)),
        List("john", "grade2", 33, "soccer", new java.sql.Date(format.parse("1978-06-07").getTime)),
        List("john", "grade2", 32, "golf", new java.sql.Date(format.parse("1978-06-07").getTime)),
        List("mike", "grade2", 26, "basketball", new java.sql.Date(format.parse("1978-06-07").getTime)),
        List("lena", "grade2", 23, "baseball", new java.sql.Date(format.parse("1978-06-07").getTime))
      )

      val rdd = sparkContext.parallelize(data).map(Row.fromSeq(_))
      val rdd2 = sparkContext.parallelize(data2).map(Row.fromSeq(_))
      val schema = StructType(Array(
        StructField("name", StringType, true),
        StructField("age", IntegerType, true),
        StructField("isBoy", BooleanType, false)
      ))
    val schema2 = StructType(Array(
        StructField("name", StringType, true),
        StructField("grade", StringType, true),
        StructField("howold", IntegerType, true),
        StructField("hobby", StringType, true),
        StructField("birthday", DateType, false)
      ))

      val df = sqlContext.createDataFrame(rdd, schema)
      val df2 = sqlContext.createDataFrame(rdd2, schema2)
      df.createOrReplaceTempView(tableName)
      df2.createOrReplaceTempView(tableName2)

我正在尝试构建查询以返回table1中没有table2中匹配行的行.我尝试使用此查询来执行此操作:

Select * from table1 LEFT JOIN table2 ON table1.name = table2.name AND table1.age = table2.howold AND table2.name IS NULL AND table2.howold IS NULL

但这只是给了我table1的所有行:

列表({"name":"john","age":33,"isBoy":true},{"name":"susan","age":26,"isBoy":false},{"name" : "迈克", "年龄":26, "isBoy":真})

如何有效地在Spark中进行这种类型的连接？

我正在寻找一个SQL查询,因为我需要能够指定要在两个表之间进行比较的列,而不是像在其他推荐的问题中那样逐行进行比较.喜欢使用减法,除了等.

Answer 1

您可以使用"左反"连接类型 - 使用DataFrame API或SQL(DataFrame API支持SQL支持的所有内容,包括您需要的任何连接条件):

DataFrame API:

df.as("table1").join(
  df2.as("table2"),
  $"table1.name" === $"table2.name" && $"table1.age" === $"table2.howold",
  "leftanti"
)

SQL:

sqlContext.sql(
  """SELECT table1.* FROM table1
    | LEFT ANTI JOIN table2
    | ON table1.name = table2.name AND table1.age = table2.howold
  """.stripMargin)

注意:还值得注意的是,有一种更简洁,更简洁的方法来创建示例数据,而无需单独指定架构,使用元组和隐式toDF方法,然后在需要时"修复"自动推断的架构:

import spark.implicits._
val df = List(
  ("mike", 26, true),
  ("susan", 26, false),
  ("john", 33, true)
).toDF("name", "age", "isBoy")

val df2 = List(
  ("mike", "grade1", 45, "baseball", new java.sql.Date(format.parse("1957-12-10").getTime)),
  ("john", "grade2", 33, "soccer", new java.sql.Date(format.parse("1978-06-07").getTime)),
  ("john", "grade2", 32, "golf", new java.sql.Date(format.parse("1978-06-07").getTime)),
  ("mike", "grade2", 26, "basketball", new java.sql.Date(format.parse("1978-06-07").getTime)),
  ("lena", "grade2", 23, "baseball", new java.sql.Date(format.parse("1978-06-07").getTime))
).toDF("name", "grade", "howold", "hobby", "birthday").withColumn("birthday", $"birthday".cast(DateType))

Answer 2

您可以使用内置函数来做到这一点except （我会使用您提供的代码，但是您没有包含导入，所以我不能只对它进行c / p :(）

val a = sc.parallelize(Seq((1,"a",123),(2,"b",456))).toDF("col1","col2","col3")
val b= sc.parallelize(Seq((4,"a",432),(2,"t",431),(2,"b",456))).toDF("col1","col2","col3")

scala> a.show()
+----+----+----+
|col1|col2|col3|
+----+----+----+
|   1|   a| 123|
|   2|   b| 456|
+----+----+----+


scala> b.show()
+----+----+----+
|col1|col2|col3|
+----+----+----+
|   4|   a| 432|
|   2|   t| 431|
|   2|   b| 456|
+----+----+----+

scala> a.except(b).show()
+----+----+----+
|col1|col2|col3|
+----+----+----+
|   1|   a| 123|
+----+----+----+