Apache Spark 中支持的七种 Join 类型

文章目录

1 INNER JOIN
2 CROSS JOIN
3 LEFT OUTER JOIN
4 RIGHT OUTER JOIN
5 FULL OUTER JOIN
6 LEFT SEMI JOIN
7 LEFT ANTI JOIN

数据分析中将两个数据集进行 Join 操作是很常见的场景。我在这篇文章中介绍了 Spark 支持的五种 Join 策略，本文我将给大家介绍一下 Apache Spark 中支持的 Join 类型（Join Type）。

目前 Apache Spark 3.0 版本中，一共支持以下七种 Join 类型：

INNER JOIN
CROSS JOIN
LEFT OUTER JOIN
RIGHT OUTER JOIN
FULL OUTER JOIN
LEFT SEMI JOIN
LEFT ANTI JOIN

在实现上，这七种 Join 对应的实现类分别如下：

object JoinType {
  def apply(typ: String): JoinType = typ.toLowerCase(Locale.ROOT).replace("_", "") match {
    case "inner" => Inner
    case "outer" | "full" | "fullouter" => FullOuter
    case "leftouter" | "left" => LeftOuter
    case "rightouter" | "right" => RightOuter
    case "leftsemi" | "semi" => LeftSemi
    case "leftanti" | "anti" => LeftAnti
    case "cross" => Cross
    case _ =>
      val supported = Seq(
        "inner",
        "outer", "full", "fullouter", "full_outer",
        "leftouter", "left", "left_outer",
        "rightouter", "right", "right_outer",
        "leftsemi", "left_semi", "semi",
        "leftanti", "left_anti", "anti",
        "cross")

      throw new IllegalArgumentException(s"Unsupported join type '$typ'. " +
        "Supported join types include: " + supported.mkString("'", "', '", "'") + ".")
  }
}

今天，我并不打算从底层代码来介绍这七种 Join 类型的实现，而是从数据分析师的角度来介绍这几种 Join 的含义和使用。在介绍下文之前，假设我们有顾客（customer）和订单（order）相关的两张表，如下：

scala> val order = spark.sparkContext.parallelize(Seq(
     |   (1, 101,2500), (2,102,1110), (3,103,500), (4 ,102,400)
     | )).toDF("paymentId", "customerId","amount")
order: org.apache.spark.sql.DataFrame = [paymentId: int, customerId: int ... 1 more field]

scala> order.show
+---------+----------+------+
|paymentId|customerId|amount|
+---------+----------+------+
|        1|       101|  2500|
|        2|       102|  1110|
|        3|       103|   500|
|        4|       102|   400|
+---------+----------+------+


scala> val customer = spark.sparkContext.parallelize(Seq(
     |     (101,"iteblog") ,(102,"iteblog_hadoop") ,(103,"iteblog001"), (104,"iteblog002"), (105,"iteblog003"), (106,"iteblog004")
     | )).toDF("customerId", "name")
customer: org.apache.spark.sql.DataFrame = [customerId: int, name: string]

scala> customer.show
+----------+--------------+
|customerId|          name|
+----------+--------------+
|       101|       iteblog|
|       102|iteblog_hadoop|
|       103|    iteblog001|
|       104|    iteblog002|
|       105|    iteblog003|
|       106|    iteblog004|
+----------+--------------+

准备好数据之后，现在我们来一一介绍这些 Join 类型。

INNER JOIN

在 Spark 中，如果没有指定任何 Join 类型，那么默认就是 INNER JOIN。INNER JOIN 只会返回满足 Join 条件（ join condition）的数据，这个大家用的应该比较多，具体如下：

scala> val df = customer.join(order,"customerId")
df: org.apache.spark.sql.DataFrame = [customerId: int, name: string ... 2 more fields]

scala> df.show
+----------+--------------+---------+------+
|customerId|          name|paymentId|amount|
+----------+--------------+---------+------+
|       101|       iteblog|        1|  2500|
|       103|    iteblog001|        3|   500|
|       102|iteblog_hadoop|        2|  1110|
|       102|iteblog_hadoop|        4|   400|
+----------+--------------+---------+------+

从上面可以看出，当我们没有指定任何 Join 类型时，默认就是 INNER JOIN；在生成的结果中， Spark 自动为我们删除了两张表都存在的 customerId。如果用图来表示的话， INNER JOIN 可以如下表示：

如果想及时了解Spark、Hadoop或者HBase相关的文章，欢迎关注微信公众号：iteblog_hadoop

上图粉色部分就是 INNER JOIN 的结果。

CROSS JOIN

这种类型的 Join 也称为笛卡儿积（Cartesian Product），Join 左表的每行数据都会跟右表的每行数据进行 Join，产生的结果行数为 m*n，所以在生产环境下尽量不要用这种 Join。下面是 CROSS JOIN 的使用例子：

scala> val df = customer.crossJoin(order)
df: org.apache.spark.sql.DataFrame = [customerId: int, name: string ... 3 more fields]

scala> df.show
+----------+--------------+---------+----------+------+
|customerId|          name|paymentId|customerId|amount|
+----------+--------------+---------+----------+------+
|       101|       iteblog|        1|       101|  2500|
|       101|       iteblog|        2|       102|  1110|
|       101|       iteblog|        3|       103|   500|
|       101|       iteblog|        4|       102|   400|
|       102|iteblog_hadoop|        1|       101|  2500|
|       102|iteblog_hadoop|        2|       102|  1110|
|       102|iteblog_hadoop|        3|       103|   500|
|       102|iteblog_hadoop|        4|       102|   400|
|       103|    iteblog001|        1|       101|  2500|
|       103|    iteblog001|        2|       102|  1110|
|       103|    iteblog001|        3|       103|   500|
|       103|    iteblog001|        4|       102|   400|
|       104|    iteblog002|        1|       101|  2500|
|       104|    iteblog002|        2|       102|  1110|
|       104|    iteblog002|        3|       103|   500|
|       104|    iteblog002|        4|       102|   400|
|       105|    iteblog003|        1|       101|  2500|
|       105|    iteblog003|        2|       102|  1110|
|       105|    iteblog003|        3|       103|   500|
|       105|    iteblog003|        4|       102|   400|
+----------+--------------+---------+----------+------+
only showing top 20 rows

LEFT OUTER JOIN

LEFT OUTER JOIN 等价于 LEFT JOIN，这个 Join 的返回的结果相信大家都知道，我就不介绍了。下面三种写法都是等价的：

val leftJoinDf = customer.join(order,Seq("customerId"), "left_outer")
val leftJoinDf = customer.join(order,Seq("customerId"), "leftouter")
val leftJoinDf = customer.join(order,Seq("customerId"), "left")

scala> leftJoinDf.show
+----------+--------------+---------+------+
|customerId|          name|paymentId|amount|
+----------+--------------+---------+------+
|       101|       iteblog|        1|  2500|
|       103|    iteblog001|        3|   500|
|       102|iteblog_hadoop|        2|  1110|
|       102|iteblog_hadoop|        4|   400|
|       105|    iteblog003|     null|  null|
|       106|    iteblog004|     null|  null|
|       104|    iteblog002|     null|  null|
+----------+--------------+---------+------+

如果用图表示的话，LEFT OUTER JOIN 可以如下所示：可以看出，参与 Join 的左表数据都会显示出来，而右表只有关联上的才会显示。

如果想及时了解Spark、Hadoop或者HBase相关的文章，欢迎关注微信公众号：iteblog_hadoop

RIGHT OUTER JOIN

和 LEFT OUTER JOIN 类似，RIGHT OUTER JOIN 等价于 RIGHT JOIN，下面三种写法也是等价的：

val rightJoinDf = order.join(customer,Seq("customerId"), "right")
val rightJoinDf = order.join(customer,Seq("customerId"), "right_outer")
val rightJoinDf = order.join(customer,Seq("customerId"), "rightouter")

scala> rightJoinDf.show
+----------+---------+------+--------------+
|customerId|paymentId|amount|          name|
+----------+---------+------+--------------+
|       101|        1|  2500|       iteblog|
|       103|        3|   500|    iteblog001|
|       102|        2|  1110|iteblog_hadoop|
|       102|        4|   400|iteblog_hadoop|
|       105|     null|  null|    iteblog003|
|       106|     null|  null|    iteblog004|
|       104|     null|  null|    iteblog002|
+----------+---------+------+--------------+

如果用图表示的话，RIGHT OUTER JOIN 可以如下所示：可以看出，参与 Join 的右表数据都会显示出来，而左表只有关联上的才会显示。

如果想及时了解Spark、Hadoop或者HBase相关的文章，欢迎关注微信公众号：iteblog_hadoop

FULL OUTER JOIN

FULL OUTER JOIN 的含义大家应该也都熟悉，我就不介绍其含义了。FULL OUTER JOIN 有以下四种写法：

val fullJoinDf = order.join(customer,Seq("customerId"), "outer")
val fullJoinDf = order.join(customer,Seq("customerId"), "full")
val fullJoinDf = order.join(customer,Seq("customerId"), "full_outer")
val fullJoinDf = order.join(customer,Seq("customerId"), "fullouter")

scala> fullJoinDf.show
+----------+---------+------+--------------+
|customerId|paymentId|amount|          name|
+----------+---------+------+--------------+
|       101|        1|  2500|       iteblog|
|       103|        3|   500|    iteblog001|
|       102|        2|  1110|iteblog_hadoop|
|       102|        4|   400|iteblog_hadoop|
|       105|     null|  null|    iteblog003|
|       106|     null|  null|    iteblog004|
|       104|     null|  null|    iteblog002|
+----------+---------+------+--------------+

FULL OUTER JOIN 可以用如下图表示：

如果想及时了解Spark、Hadoop或者HBase相关的文章，欢迎关注微信公众号：iteblog_hadoop

LEFT SEMI JOIN

LEFT SEMI JOIN 这个大家应该知道的人相对少些，LEFT SEMI JOIN 只会返回匹配右表的数据，而且 LEFT SEMI JOIN 只会返回左表的数据，右表的数据是不会显示的，下面三种写法都是等价的：

val leftSemiJoinDf = order.join(customer,Seq("customerId"), "leftsemi")
val leftSemiJoinDf = order.join(customer,Seq("customerId"), "left_semi")
val leftSemiJoinDf = order.join(customer,Seq("customerId"), "semi")

scala> leftSemiJoinDf.show
+----------+---------+------+
|customerId|paymentId|amount|
+----------+---------+------+
|       101|        1|  2500|
|       103|        3|   500|
|       102|        2|  1110|
|       102|        4|   400|
+----------+---------+------+

从上面结果可以看出，LEFT SEMI JOIN 其实可以用 IN/EXISTS 来改写：

scala> order.registerTempTable("order")
warning: there was one deprecation warning (since 2.0.0); for details, enable `:setting -deprecation' or `:replay -deprecation'

scala> customer.registerTempTable("customer")
warning: there was one deprecation warning (since 2.0.0); for details, enable `:setting -deprecation' or `:replay -deprecation'

scala> val r = spark.sql("select * from order where customerId in (select customerId from customer)")
r: org.apache.spark.sql.DataFrame = [paymentId: int, customerId: int ... 1 more field]

scala> r.show
+---------+----------+------+
|paymentId|customerId|amount|
+---------+----------+------+
|        1|       101|  2500|
|        3|       103|   500|
|        2|       102|  1110|
|        4|       102|   400|
+---------+----------+------+

LEFT SEMI JOIN 可以用下图表示：

如果想及时了解Spark、Hadoop或者HBase相关的文章，欢迎关注微信公众号：iteblog_hadoop

LEFT ANTI JOIN

与 LEFT SEMI JOIN 相反，LEFT ANTI JOIN 只会返回没有匹配到右表的左表数据。而且下面三种写法也是等效的：

val leftAntiJoinDf = customer.join(order,Seq("customerId"), "leftanti")
val leftAntiJoinDf = customer.join(order,Seq("customerId"), "left_anti")
val leftAntiJoinDf = customer.join(order,Seq("customerId"), "anti")

scala> leftAntiJoinDf.show
+----------+----------+
|customerId|      name|
+----------+----------+
|       105|iteblog003|
|       106|iteblog004|
|       104|iteblog002|
+----------+----------+

同理，LEFT ANTI JOIN 也可以用 NOT IN 来改写：

scala> val r = spark.sql("select * from customer where customerId not in (select customerId from order)")
r: org.apache.spark.sql.DataFrame = [customerId: int, name: string]

scala> r.show
+----------+----------+
|customerId|      name|
+----------+----------+
|       104|iteblog002|
|       105|iteblog003|
|       106|iteblog004|
+----------+----------+

LEFT SEMI ANTI 可以用下图表示：