RDD常用算子
This commit is contained in:
罗祥
parent
2bf214d18b
commit
ceaf2b87b3
@ -1,27 +0,0 @@
|
||||
val list = List(3,6,9,10,12,21)
|
||||
val listRDD = sc.parallelize(list)
|
||||
val intsRDD = listRDD.map(_*10)
|
||||
intsRDD.foreach(println)
|
||||
|
||||
sc.parallelize(list).map(_*10).foreach(println)
|
||||
|
||||
|
||||
sc.parallelize(list).filter(_>=10).foreach(println)
|
||||
|
||||
val list = List(List(1, 2), List(3), List(), List(4, 5))
|
||||
sc.parallelize(list).flatMap(_.toList).map(_*10).foreach(println)
|
||||
|
||||
|
||||
val list = List(1,2,3,4,5)
|
||||
sc.parallelize(list).reduce((x,y) => x+y)
|
||||
sc.parallelize(list).reduce(_+_)
|
||||
|
||||
|
||||
val list = List(("hadoop", 2), ("spark", 3), ("spark", 5), ("storm", 6),("hadoop", 2))
|
||||
sc.parallelize(list).reduceByKey(_+_).foreach(println)
|
||||
|
||||
|
||||
|
||||
val list = List(("hadoop", 2), ("spark", 3), ("spark", 5), ("storm", 6),("hadoop", 2))
|
||||
sc.parallelize(list).groupByKey().map(x=>(x._1,x._2.toList)).foreach(println)
|
||||
|
||||
@ -9,7 +9,7 @@
|
||||
<version>1.0</version>
|
||||
|
||||
<properties>
|
||||
<scala.version>2.12.8</scala.version>
|
||||
<scala.version>2.12</scala.version>
|
||||
</properties>
|
||||
|
||||
<build>
|
||||
@ -27,23 +27,20 @@
|
||||
|
||||
|
||||
<dependencies>
|
||||
<!--spark核心依赖-->
|
||||
<dependency>
|
||||
<groupId>org.apache.spark</groupId>
|
||||
<artifactId>spark-core_2.12</artifactId>
|
||||
<artifactId>spark-core_${scala.version}</artifactId>
|
||||
<version>2.4.0</version>
|
||||
</dependency>
|
||||
<dependency>
|
||||
<groupId>org.scalatest</groupId>
|
||||
<artifactId>scalatest_2.12</artifactId>
|
||||
<version>3.0.1</version>
|
||||
<scope>test</scope>
|
||||
</dependency>
|
||||
<!--单元测试依赖包-->
|
||||
<dependency>
|
||||
<groupId>junit</groupId>
|
||||
<artifactId>junit</artifactId>
|
||||
<version>4.12</version>
|
||||
</dependency>
|
||||
<!--Exception in thread "main" java.lang.ArrayIndexOutOfBoundsException: 10582-->
|
||||
<!--如果出现异常:Exception in thread "main" java.lang.ArrayIndexOutOfBoundsException: 10582
|
||||
则是因为paranamer版本问题，添加下面的依赖包-->
|
||||
<dependency>
|
||||
<groupId>com.thoughtworks.paranamer</groupId>
|
||||
<artifactId>paranamer</artifactId>
|
||||
|
||||
@ -3,6 +3,8 @@ package rdd.scala
|
||||
import org.apache.spark.{SparkConf, SparkContext}
|
||||
import org.junit.{After, Test}
|
||||
|
||||
import scala.collection.mutable.ListBuffer
|
||||
|
||||
class TransformationTest {
|
||||
|
||||
val conf: SparkConf = new SparkConf().setAppName("TransformationTest").setMaster("local[2]")
|
||||
@ -11,10 +13,185 @@ class TransformationTest {
|
||||
|
||||
@Test
|
||||
def map(): Unit = {
|
||||
val list = List(3, 6, 9, 10, 12, 21)
|
||||
val list = List(1, 2, 3)
|
||||
sc.parallelize(list).map(_ * 10).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def filter(): Unit = {
|
||||
val list = List(3, 6, 9, 10, 12, 21)
|
||||
sc.parallelize(list).filter(_ >= 10).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def flatMap(): Unit = {
|
||||
val list = List(List(1, 2), List(3), List(), List(4, 5))
|
||||
sc.parallelize(list).flatMap(_.toList).map(_ * 10).foreach(println)
|
||||
|
||||
val lines = List("spark flume spark",
|
||||
"hadoop flume hive")
|
||||
sc.parallelize(lines).flatMap(line => line.split(" ")).
|
||||
map(word => (word, 1)).reduceByKey(_ + _).foreach(println)
|
||||
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def mapPartitions(): Unit = {
|
||||
val list = List(1, 2, 3, 4, 5, 6)
|
||||
sc.parallelize(list, 3).mapPartitions(iterator => {
|
||||
val buffer = new ListBuffer[Int]
|
||||
while (iterator.hasNext) {
|
||||
buffer.append(iterator.next() * 100)
|
||||
}
|
||||
buffer.toIterator
|
||||
}).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def mapPartitionsWithIndex(): Unit = {
|
||||
val list = List(1, 2, 3, 4, 5, 6)
|
||||
sc.parallelize(list, 3).mapPartitionsWithIndex((index, iterator) => {
|
||||
val buffer = new ListBuffer[String]
|
||||
while (iterator.hasNext) {
|
||||
buffer.append(index + "分区:" + iterator.next() * 100)
|
||||
}
|
||||
buffer.toIterator
|
||||
}).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def sample(): Unit = {
|
||||
val list = List(1, 2, 3, 4, 5, 6)
|
||||
sc.parallelize(list).sample(withReplacement = false, 0.5).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def union(): Unit = {
|
||||
val list1 = List(1, 2, 3)
|
||||
val list2 = List(4, 5, 6)
|
||||
sc.parallelize(list1).union(sc.parallelize(list2)).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def intersection(): Unit = {
|
||||
val list1 = List(1, 2, 3, 4, 5)
|
||||
val list2 = List(4, 5, 6)
|
||||
sc.parallelize(list1).intersection(sc.parallelize(list2)).foreach(println)
|
||||
}
|
||||
|
||||
@Test
|
||||
def distinct(): Unit = {
|
||||
val list = List(1, 2, 2, 4, 4)
|
||||
sc.parallelize(list).distinct().foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def groupByKey(): Unit = {
|
||||
val list = List(("hadoop", 2), ("spark", 3), ("spark", 5), ("storm", 6), ("hadoop", 2))
|
||||
sc.parallelize(list).groupByKey().map(x => (x._1, x._2.toList)).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def reduceByKey(): Unit = {
|
||||
val list = List(("hadoop", 2), ("spark", 3), ("spark", 5), ("storm", 6), ("hadoop", 2))
|
||||
sc.parallelize(list).reduceByKey(_ + _).foreach(println)
|
||||
}
|
||||
|
||||
@Test
|
||||
def aggregateByKey(): Unit = {
|
||||
val list = List(("hadoop", 3), ("hadoop", 2), ("spark", 4), ("spark", 3), ("storm", 6), ("storm", 8))
|
||||
sc.parallelize(list, numSlices = 6).aggregateByKey(zeroValue = 0, numPartitions = 5)(
|
||||
seqOp = math.max(_, _),
|
||||
combOp = _ + _
|
||||
).getNumPartitions
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def sortBy(): Unit = {
|
||||
val list01 = List((100, "hadoop"), (90, "spark"), (120, "storm"))
|
||||
sc.parallelize(list01).sortByKey(ascending = false).foreach(println)
|
||||
|
||||
val list02 = List(("hadoop", 100), ("spark", 90), ("storm", 120))
|
||||
sc.parallelize(list02).sortBy(x => x._2, ascending = false).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def join(): Unit = {
|
||||
val list01 = List((1, "student01"), (2, "student02"), (3, "student03"))
|
||||
val list02 = List((1, "teacher01"), (2, "teacher02"), (3, "teacher03"))
|
||||
sc.parallelize(list01).join(sc.parallelize(list02)).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def cogroup(): Unit = {
|
||||
val list01 = List((1, "a"), (1, "a"), (2, "b"), (3, "e"))
|
||||
val list02 = List((1, "A"), (2, "B"), (3, "E"))
|
||||
val list03 = List((1, "[ab]"), (2, "[bB]"), (3, "eE"), (3, "eE"))
|
||||
sc.parallelize(list01).cogroup(sc.parallelize(list02), sc.parallelize(list03)).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def cartesian(): Unit = {
|
||||
val list1 = List("A", "B", "C")
|
||||
val list2 = List(1, 2, 3)
|
||||
sc.parallelize(list1).cartesian(sc.parallelize(list2)).foreach(println)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def reduce(): Unit = {
|
||||
val list = List(1, 2, 3, 4, 5)
|
||||
sc.parallelize(list).reduce((x, y) => x + y)
|
||||
sc.parallelize(list).reduce(_ + _)
|
||||
}
|
||||
|
||||
// 继承Ordering[T],实现自定义比较器
|
||||
class CustomOrdering extends Ordering[(Int, String)] {
|
||||
override def compare(x: (Int, String), y: (Int, String)): Int
|
||||
= if (x._2.length > y._2.length) 1 else -1
|
||||
}
|
||||
|
||||
@Test
|
||||
def takeOrdered(): Unit = {
|
||||
val list = List((1, "hadoop"), (1, "storm"), (1, "azkaban"), (1, "hive"))
|
||||
// 定义隐式默认值
|
||||
implicit val implicitOrdering = new CustomOrdering
|
||||
sc.parallelize(list).takeOrdered(5)
|
||||
}
|
||||
|
||||
|
||||
@Test
|
||||
def countByKey(): Unit = {
|
||||
val list = List(("hadoop", 10), ("hadoop", 10), ("storm", 3), ("storm", 3), ("azkaban", 1))
|
||||
sc.parallelize(list).countByKey()
|
||||
}
|
||||
|
||||
@Test
|
||||
def saveAsTextFile(): Unit = {
|
||||
val list = List(("hadoop", 10), ("hadoop", 10), ("storm", 3), ("storm", 3), ("azkaban", 1))
|
||||
sc.parallelize(list).saveAsTextFile("/usr/file/temp")
|
||||
}
|
||||
|
||||
@Test
|
||||
def saveAsSequenceFile(): Unit = {
|
||||
val list = List(("hadoop", 10), ("hadoop", 10), ("storm", 3), ("storm", 3), ("azkaban", 1))
|
||||
sc.parallelize(list).saveAsSequenceFile("/usr/file/sequence")
|
||||
}
|
||||
|
||||
|
||||
@After
|
||||
def destroy(): Unit = {
|
||||
sc.stop()
|
||||
|
||||
@ -0,0 +1,420 @@
|
||||
# Transformation 和 Action 常用算子
|
||||
|
||||
<nav>
|
||||
<a href="#一Transformation">一、Transformation</a><br/>
|
||||
<a href="#11-map">1.1 map</a><br/>
|
||||
<a href="#12-filter">1.2 filter</a><br/>
|
||||
<a href="#13-flatMap">1.3 flatMap</a><br/>
|
||||
<a href="#14-mapPartitions">1.4 mapPartitions</a><br/>
|
||||
<a href="#15-mapPartitionsWithIndex">1.5 mapPartitionsWithIndex</a><br/>
|
||||
<a href="#16-sample">1.6 sample</a><br/>
|
||||
<a href="#17-union">1.7 union</a><br/>
|
||||
<a href="#18-intersection">1.8 intersection</a><br/>
|
||||
<a href="#19-distinct">1.9 distinct</a><br/>
|
||||
<a href="#110-groupByKey">1.10 groupByKey</a><br/>
|
||||
<a href="#111-reduceByKey">1.11 reduceByKey</a><br/>
|
||||
<a href="#112-sortBy--sortByKey">1.12 sortBy & sortByKey </a><br/>
|
||||
<a href="#113-join">1.13 join</a><br/>
|
||||
<a href="#114-cogroup">1.14 cogroup</a><br/>
|
||||
<a href="#115-cartesian">1.15 cartesian</a><br/>
|
||||
<a href="#116-aggregateByKey">1.16 aggregateByKey</a><br/>
|
||||
<a href="#二Action">二、Action</a><br/>
|
||||
<a href="#21-reduce">2.1 reduce</a><br/>
|
||||
<a href="#22-takeOrdered">2.2 takeOrdered</a><br/>
|
||||
<a href="#23-countByKey">2.3 countByKey</a><br/>
|
||||
<a href="#24-saveAsTextFile">2.4 saveAsTextFile</a><br/>
|
||||
</nav>
|
||||
|
||||
## 一、Transformation
|
||||
|
||||
下表为spark官网给出的常用的Transformation算子：
|
||||
|
||||
| Transformation算子 | Meaning（含义） |
|
||||
| ------------------------------------------------------------ | ------------------------------------------------------------ |
|
||||
| **map**(*func*) | 对原RDD中每个元素运用 *func* 函数，并生成新的RDD |
|
||||
| **filter**(*func*) | 对原RDD中每个元素使用*func* 函数进行过滤，并生成新的RDD |
|
||||
| **flatMap**(*func*) | 与 map 类似，但是每一个输入的 item 被映射成 0 个或多个输出的 items（ *func* 返回类型需要为 Seq ）。 |
|
||||
| **mapPartitions**(*func*) | 与 map 类似，但函数单独在RDD的每个分区上运行， *func*函数的类型为 Iterator\<T> => Iterator\<U> ，其中T是RDD的类型，即RDD[T] |
|
||||
| **mapPartitionsWithIndex**(*func*) | 与 mapPartitions 类似，但 *func* 类型为 (Int, Iterator\<T>) => Iterator\<U> ，其中第一个参数为分区索引 |
|
||||
| **sample**(*withReplacement*, *fraction*, *seed*) | 数据采样，有三个可选参数：设置是否放回（withReplacement）、采样的百分比（*fraction*）、随机数生成器的种子（seed）； |
|
||||
| **union**(*otherDataset*) | 合并两个RDD |
|
||||
| **intersection**(*otherDataset*) | 求两个RDD的交集 |
|
||||
| **distinct**([*numTasks*])) | 去重 |
|
||||
| **groupByKey**([*numTasks*]) | 按照key值进行分区，即在一个 (K, V) 对的 dataset 上调用时，返回一个 (K, Iterable\<V>) <br/>**Note:** 如果分组是为了在每一个 key 上执行聚合操作（例如，sum 或 average)，此时使用 `reduceByKey` 或 `aggregateByKey` 性能会更好<br>**Note:** 默认情况下，并行度取决于父 RDD 的分区数。可以传入 `numTasks` 参数进行修改。 |
|
||||
| **reduceByKey**(*func*, [*numTasks*]) | 按照key值进行分组，并对分组后的数据执行归约操作。 |
|
||||
| **aggregateByKey**(*zeroValue*,*numPartitions*)(*seqOp*, *combOp*, [*numTasks*]) | 当调用（K，V）对的数据集时，返回（K，U）对的数据集，其中使用给定的组合函数和zeroValue聚合每个键的值。与groupByKey类似，reduce任务的数量可通过第二个参数进行配置。 |
|
||||
| **sortByKey**([*ascending*], [*numTasks*]) | 按照key进行排序，其中的key需要实现Ordered特质，即可比较 |
|
||||
| **join**(*otherDataset*, [*numTasks*]) | 在一个 (K, V) 和 (K, W) 类型的 dataset 上调用时，返回一个 (K, (V, W)) pairs 的 dataset，等价于内连接操作。如果想要执行外连接，可以使用`leftOuterJoin`, `rightOuterJoin` 和 `fullOuterJoin` 等算子。 |
|
||||
| **cogroup**(*otherDataset*, [*numTasks*]) | 在一个 (K, V) 对的 dataset 上调用时，返回一个 (K, (Iterable\<V>, Iterable\<W>)) tuples 的 dataset。 |
|
||||
| **cartesian**(*otherDataset*) | 在一个 T 和 U 类型的 dataset 上调用时，返回一个 (T, U) 类型的 dataset（即笛卡尔积）。 |
|
||||
| **coalesce**(*numPartitions*) | 将RDD中的分区数减少为numPartitions。 |
|
||||
| **repartition**(*numPartitions*) | 随机重新调整RDD中的数据以创建更多或更少的分区，并在它们之间进行平衡。 |
|
||||
| **repartitionAndSortWithinPartitions**(*partitioner*) | 根据给定的 partitioner（分区器）对 RDD 进行重新分区，并对分区中的数据按照 key 值进行排序。这比调用 `repartition` 然后再 sorting（排序）效率更高，因为它可以将排序过程推送到 shuffle 操作所在的机器。 |
|
||||
|
||||
下面分别给出这些算子的基本使用实例：
|
||||
|
||||
### 1.1 map
|
||||
|
||||
```scala
|
||||
val list = List(1,2,3)
|
||||
sc.parallelize(list).map(_ * 10).foreach(println)
|
||||
|
||||
// 输出结果： 10 20 30 （这里为了节省篇幅去掉了换行,后文亦同）
|
||||
```
|
||||
|
||||
### 1.2 filter
|
||||
|
||||
```scala
|
||||
val list = List(3, 6, 9, 10, 12, 21)
|
||||
sc.parallelize(list).filter(_ >= 10).foreach(println)
|
||||
|
||||
// 输出： 10 12 21
|
||||
```
|
||||
|
||||
### 1.3 flatMap
|
||||
|
||||
与 map 类似，但是每一个输入的 item 被映射成 0 个或多个输出的 items（ *func* 返回类型需要为 Seq ）。
|
||||
|
||||
```scala
|
||||
val list = List(List(1, 2), List(3), List(), List(4, 5))
|
||||
sc.parallelize(list).flatMap(_.toList).map(_ * 10).foreach(println)
|
||||
|
||||
// 输出结果 ： 10 20 30 40 50
|
||||
```
|
||||
|
||||
flatMap 这个算子在日志分析中使用概率非常高，这里进行一下演示：
|
||||
|
||||
拆分输入的每行数据为单个单词，并赋值为1，代表出现一次，之后按照单词分组并统计其出现总次数，代码如下：
|
||||
|
||||
```scala
|
||||
val lines = List("spark flume spark",
|
||||
"hadoop flume hive")
|
||||
sc.parallelize(lines).flatMap(line => line.split(" ")).
|
||||
map(word=>(word,1)).reduceByKey(_+_).foreach(println)
|
||||
|
||||
// 输出：
|
||||
(spark,2)
|
||||
(hive,1)
|
||||
(hadoop,1)
|
||||
(flume,2)
|
||||
```
|
||||
|
||||
### 1.4 mapPartitions
|
||||
|
||||
与 map 类似，但函数单独在RDD的每个分区上运行， *func*函数的类型为Iterator\<T> => Iterator\<U> （其中T是RDD的类型），即输入和输出都必须是可迭代类型。
|
||||
|
||||
```scala
|
||||
val list = List(1, 2, 3, 4, 5, 6)
|
||||
sc.parallelize(list, 3).mapPartitions(iterator => {
|
||||
val buffer = new ListBuffer[Int]
|
||||
while (iterator.hasNext) {
|
||||
buffer.append(iterator.next() * 100)
|
||||
}
|
||||
buffer.toIterator
|
||||
}).foreach(println)
|
||||
//输出结果
|
||||
100 200 300 400 500 600
|
||||
```
|
||||
|
||||
### 1.5 mapPartitionsWithIndex
|
||||
|
||||
与 mapPartitions 类似，但 *func* 类型为 (Int, Iterator\<T>) => Iterator\<U> ，其中第一个参数为分区索引。
|
||||
|
||||
```scala
|
||||
val list = List(1, 2, 3, 4, 5, 6)
|
||||
sc.parallelize(list, 3).mapPartitionsWithIndex((index, iterator) => {
|
||||
val buffer = new ListBuffer[String]
|
||||
while (iterator.hasNext) {
|
||||
buffer.append(index + "分区:" + iterator.next() * 100)
|
||||
}
|
||||
buffer.toIterator
|
||||
}).foreach(println)
|
||||
//输出
|
||||
0分区:100
|
||||
0分区:200
|
||||
1分区:300
|
||||
1分区:400
|
||||
2分区:500
|
||||
2分区:600
|
||||
```
|
||||
|
||||
### 1.6 sample
|
||||
|
||||
数据采样，有三个可选参数：设置是否放回（withReplacement）、采样的百分比（*fraction*）、随机数生成器的种子（seed）:
|
||||
|
||||
```scala
|
||||
val list = List(1, 2, 3, 4, 5, 6)
|
||||
sc.parallelize(list).sample(withReplacement = false, 0.5).foreach(println)
|
||||
```
|
||||
|
||||
### 1.7 union
|
||||
|
||||
合并两个RDD：
|
||||
|
||||
```scala
|
||||
val list1 = List(1, 2, 3)
|
||||
val list2 = List(4, 5, 6)
|
||||
sc.parallelize(list1).union(sc.parallelize(list2)).foreach(println)
|
||||
// 输出: 1 2 3 4 5 6
|
||||
```
|
||||
|
||||
### 1.8 intersection
|
||||
|
||||
求两个RDD的交集：
|
||||
|
||||
```scala
|
||||
val list1 = List(1, 2, 3, 4, 5)
|
||||
val list2 = List(4, 5, 6)
|
||||
sc.parallelize(list1).intersection(sc.parallelize(list2)).foreach(println)
|
||||
// 输出: 4 5
|
||||
```
|
||||
|
||||
### 1.9 distinct
|
||||
|
||||
去重：
|
||||
|
||||
```scala
|
||||
val list = List(1, 2, 2, 4, 4)
|
||||
sc.parallelize(list).distinct().foreach(println)
|
||||
// 输出: 4 1 2
|
||||
```
|
||||
|
||||
### 1.10 groupByKey
|
||||
|
||||
按照键进行分组：
|
||||
|
||||
```scala
|
||||
val list = List(("hadoop", 2), ("spark", 3), ("spark", 5), ("storm", 6), ("hadoop", 2))
|
||||
sc.parallelize(list).groupByKey().map(x => (x._1, x._2.toList)).foreach(println)
|
||||
|
||||
//输出：
|
||||
(spark,List(3, 5))
|
||||
(hadoop,List(2, 2))
|
||||
(storm,List(6))
|
||||
```
|
||||
|
||||
### 1.11 reduceByKey
|
||||
|
||||
按照键进行归约操作：
|
||||
|
||||
```scala
|
||||
val list = List(("hadoop", 2), ("spark", 3), ("spark", 5), ("storm", 6), ("hadoop", 2))
|
||||
sc.parallelize(list).reduceByKey(_ + _).foreach(println)
|
||||
|
||||
//输出
|
||||
(spark,8)
|
||||
(hadoop,4)
|
||||
(storm,6)
|
||||
```
|
||||
|
||||
### 1.12 sortBy & sortByKey
|
||||
|
||||
按照键进行排序：
|
||||
|
||||
```scala
|
||||
val list01 = List((100, "hadoop"), (90, "spark"), (120, "storm"))
|
||||
sc.parallelize(list01).sortByKey(ascending = false).foreach(println)
|
||||
// 输出
|
||||
(120,storm)
|
||||
(90,spark)
|
||||
(100,hadoop)
|
||||
```
|
||||
|
||||
按照指定元素进行排序：
|
||||
|
||||
```scala
|
||||
val list02 = List(("hadoop",100), ("spark",90), ("storm",120))
|
||||
sc.parallelize(list02).sortBy(x=>x._2,ascending=false).foreach(println)
|
||||
// 输出
|
||||
(storm,120)
|
||||
(hadoop,100)
|
||||
(spark,90)
|
||||
```
|
||||
|
||||
### 1.13 join
|
||||
|
||||
在一个 (K, V) 和 (K, W) 类型的 dataset 上调用时，返回一个 (K, (V, W)) pairs 的 dataset，等价于内连接操作。如果想要执行外连接，可以使用`leftOuterJoin`, `rightOuterJoin` 和 `fullOuterJoin` 等算子。
|
||||
|
||||
```scala
|
||||
val list01 = List((1, "student01"), (2, "student02"), (3, "student03"))
|
||||
val list02 = List((1, "teacher01"), (2, "teacher02"), (3, "teacher03"))
|
||||
sc.parallelize(list01).join(sc.parallelize(list02)).foreach(println)
|
||||
|
||||
// 输出
|
||||
(1,(student01,teacher01))
|
||||
(3,(student03,teacher03))
|
||||
(2,(student02,teacher02))
|
||||
```
|
||||
|
||||
### 1.14 cogroup
|
||||
|
||||
在一个 (K, V) 对的 dataset 上调用时，返回多个类型为 (K, (Iterable\<V>, Iterable\<W>))的元组所组成的dataset。
|
||||
|
||||
```scala
|
||||
val list01 = List((1, "a"),(1, "a"), (2, "b"), (3, "e"))
|
||||
val list02 = List((1, "A"), (2, "B"), (3, "E"))
|
||||
val list03 = List((1, "[ab]"), (2, "[bB]"), (3, "eE"),(3, "eE"))
|
||||
sc.parallelize(list01).cogroup(sc.parallelize(list02),sc.parallelize(list03)).foreach(println)
|
||||
|
||||
// 输出： 同一个RDD中的元素先按照key进行分组，然后再对不同RDD中的元素按照key进行分组
|
||||
(1,(CompactBuffer(a, a),CompactBuffer(A),CompactBuffer([ab])))
|
||||
(3,(CompactBuffer(e),CompactBuffer(E),CompactBuffer(eE, eE)))
|
||||
(2,(CompactBuffer(b),CompactBuffer(B),CompactBuffer([bB])))
|
||||
|
||||
```
|
||||
|
||||
### 1.15 cartesian
|
||||
|
||||
计算笛卡尔积：
|
||||
|
||||
```scala
|
||||
val list1 = List("A", "B", "C")
|
||||
val list2 = List(1, 2, 3)
|
||||
sc.parallelize(list1).cartesian(sc.parallelize(list2)).foreach(println)
|
||||
|
||||
//输出笛卡尔积
|
||||
(A,1)
|
||||
(A,2)
|
||||
(A,3)
|
||||
(B,1)
|
||||
(B,2)
|
||||
(B,3)
|
||||
(C,1)
|
||||
(C,2)
|
||||
(C,3)
|
||||
```
|
||||
|
||||
### 1.16 aggregateByKey
|
||||
|
||||
当调用（K，V）对的数据集时，返回（K，U）对的数据集，其中使用给定的组合函数和zeroValue聚合每个键的值。与groupByKey类似，reduce任务的数量可通过第二个参数`numPartitions`进行配置。示例如下：
|
||||
|
||||
```scala
|
||||
// 为了清晰，以下所有参数均使用具名传参
|
||||
val list = List(("hadoop", 3), ("hadoop", 2), ("spark", 4), ("spark", 3), ("storm", 6), ("storm", 8))
|
||||
sc.parallelize(list,numSlices = 2).aggregateByKey(zeroValue = 0,numPartitions = 3)(
|
||||
seqOp = math.max(_, _),
|
||||
combOp = _ + _
|
||||
).collect.foreach(println)
|
||||
//输出结果：
|
||||
(hadoop,3)
|
||||
(storm,8)
|
||||
(spark,7)
|
||||
```
|
||||
|
||||
这里使用了`numSlices = 2`指定aggregateByKey父操作parallelize的分区数量为2，其执行流程如下：
|
||||
|
||||
<div align="center"> <img src="https://github.com/heibaiying/BigData-Notes/blob/master/pictures/spark-aggregateByKey.png"/> </div>
|
||||
|
||||
基于同样的执行流程，如果`numSlices = 1`，则意味着只有输入一个分区，则其最后一步combOp相当于是无效的，执行结果为：
|
||||
|
||||
```properties
|
||||
(hadoop,3)
|
||||
(storm,8)
|
||||
(spark,4)
|
||||
```
|
||||
|
||||
同样的，如果每个单词对一个分区，即`numSlices = 6`，此时相当于求和操作，执行结果为：
|
||||
|
||||
```properties
|
||||
(hadoop,5)
|
||||
(storm,14)
|
||||
(spark,7)
|
||||
```
|
||||
|
||||
最后一个问题是`aggregateByKey(zeroValue = 0,numPartitions = 3)`的第二个参数`numPartitions `决定的是什么？实际上这个参数决定的是输出RDD的分区数量，想要验证这个问题，可以对上面代码进行改写，使用`getNumPartitions`方法获取分区数量：
|
||||
|
||||
```scala
|
||||
sc.parallelize(list,numSlices = 6).aggregateByKey(zeroValue = 0,numPartitions = 3)(
|
||||
seqOp = math.max(_, _),
|
||||
combOp = _ + _
|
||||
).getNumPartitions
|
||||
```
|
||||
|
||||
<div align="center"> <img src="https://github.com/heibaiying/BigData-Notes/blob/master/pictures/spark-getpartnum.png"/> </div>
|
||||
|
||||
## 二、Action
|
||||
|
||||
下表为spark官网给出的常用的Action算子：
|
||||
|
||||
| Action（动作） | Meaning（含义） |
|
||||
| -------------------------------------------------- | ------------------------------------------------------------ |
|
||||
| **reduce**(*func*) | 使用函数*func*执行归约操作 |
|
||||
| **collect**() | 以一个 array 数组的形式返回 dataset 的所有元素，适用于小结果集。 |
|
||||
| **count**() | 返回 dataset 中元素的个数。 |
|
||||
| **first**() | 返回 dataset 中的第一个元素，等价于 take(1)。 |
|
||||
| **take**(*n*) | 将数据集中的前 *n* 个元素作为一个 array 数组返回。 |
|
||||
| **takeSample**(*withReplacement*, *num*, [*seed*]) | 对一个 dataset 进行随机抽样 |
|
||||
| **takeOrdered**(*n*, *[ordering]*) | 按自然顺序（natural order）或自定义比较器（custom comparator）排序后返回前 *n* 个元素。只适用于小结果集，因为所有数据都会被加载到驱动程序的内存中进行排序。 |
|
||||
| **saveAsTextFile**(*path*) | 将 dataset 中的元素以文本文件的形式写入本地文件系统、HDFS 或其它 Hadoop 支持的文件系统中。Spark 将对每个元素调用 toString 方法，将元素转换为文本文件中的一行记录。 |
|
||||
| **saveAsSequenceFile**(*path*) | 将 dataset 中的元素以Hadoop SequenceFile 的形式写入到本地文件系统、HDFS 或其它 Hadoop 支持的文件系统中。该操作要求RDD中的元素需要实现 Hadoop 的 Writable 接口。对于Scala语言而言，它可以将Spark中的基本数据类型自动隐式转换为对应Writable类型。(目前仅支持Java and Scala) |
|
||||
| **saveAsObjectFile**(*path*) | 使用 Java 序列化后存储，可以使用 `SparkContext.objectFile()` 进行加载。(目前仅支持Java and Scala) |
|
||||
| **countByKey**() | 计算每个键出现的次数。 |
|
||||
| **foreach**(*func*) | 遍历RDD中每个元素，并对其执行*fun*函数 |
|
||||
|
||||
### 2.1 reduce
|
||||
|
||||
使用函数*func*执行归约操作：
|
||||
|
||||
```scala
|
||||
val list = List(1, 2, 3, 4, 5)
|
||||
sc.parallelize(list).reduce((x, y) => x + y)
|
||||
sc.parallelize(list).reduce(_ + _)
|
||||
|
||||
// 输出 15
|
||||
```
|
||||
|
||||
### 2.2 takeOrdered
|
||||
|
||||
按自然顺序（natural order）或自定义比较器（custom comparator）排序后返回前 *n* 个元素。需要注意的是`takeOrdered`使用隐式参数进行隐式转换，以下为其源码。所以在使用自定义排序时，需要继承Ordering[T]实现自定义比较器，然后将其作为隐式参数引入。
|
||||
|
||||
```scala
|
||||
def takeOrdered(num: Int)(implicit ord: Ordering[T]): Array[T] = withScope {
|
||||
.........
|
||||
}
|
||||
```
|
||||
|
||||
自定义规则排序：
|
||||
|
||||
```scala
|
||||
// 继承Ordering[T],实现自定义比较器，按照value值的长度进行排序
|
||||
class CustomOrdering extends Ordering[(Int, String)] {
|
||||
override def compare(x: (Int, String), y: (Int, String)): Int
|
||||
= if (x._2.length > y._2.length) 1 else -1
|
||||
}
|
||||
|
||||
val list = List((1, "hadoop"), (1, "storm"), (1, "azkaban"), (1, "hive"))
|
||||
// 引入隐式默认值
|
||||
implicit val implicitOrdering = new CustomOrdering
|
||||
sc.parallelize(list).takeOrdered(5)
|
||||
|
||||
// 输出： Array((1,hive), (1,storm), (1,hadoop), (1,azkaban)
|
||||
```
|
||||
|
||||
### 2.3 countByKey
|
||||
|
||||
计算每个键出现的次数：
|
||||
|
||||
```scala
|
||||
val list = List(("hadoop", 10), ("hadoop", 10), ("storm", 3), ("storm", 3), ("azkaban", 1))
|
||||
sc.parallelize(list).countByKey()
|
||||
|
||||
// 输出： Map(hadoop -> 2, storm -> 2, azkaban -> 1)
|
||||
```
|
||||
|
||||
### 2.4 saveAsTextFile
|
||||
|
||||
将 dataset 中的元素以文本文件的形式写入本地文件系统、HDFS 或其它 Hadoop 支持的文件系统中。Spark 将对每个元素调用 toString 方法，将元素转换为文本文件中的一行记录。
|
||||
|
||||
```scala
|
||||
val list = List(("hadoop", 10), ("hadoop", 10), ("storm", 3), ("storm", 3), ("azkaban", 1))
|
||||
sc.parallelize(list).saveAsTextFile("/usr/file/temp")
|
||||
```
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
## 参考资料
|
||||
|
||||
[RDD Programming Guide](http://spark.apache.org/docs/latest/rdd-programming-guide.html#rdd-programming-guide)
|
||||
|
||||
BIN
pictures/spark-aggregateByKey.png
Normal file
BIN
pictures/spark-aggregateByKey.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 67 KiB |
BIN
pictures/spark-getpartnum.png
Normal file
BIN
pictures/spark-getpartnum.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 13 KiB |
Loading…
x
Reference in New Issue
Block a user