centos 7.2 spark 2.3.3 scala 2.11.11 java 1.8.0_202-ea
spark-shell中为scala语法格式
1.distinct 去重
val c = sc.parallerlize(List("Gnu","Cat","Rat","Dog","Gnu","Rat"),2) 初始化rdd,将数据均匀加载到2个partition中
c.distinct.collect
>>res1: Array[String]=Array(Dog,Gnu,Cat,Rat)
2. c.fisrt
first取RDD第一个Partition中的第一个记录
>>res2:String = Gnu
3.filter 过滤
val a = sc.parallelize(1 to 10,3)
val b = a.filter(_ % 2 ==0)
b.collect
>>res3:Array[Int] = Array(2,4,6,8,10)
4.filterByRange 返回指定范围内RDD记录,只能作用于排序RDD
val randRDD = sc.parallelize(List((2,"cat"),(6,"mouse"),(7,"cup),(3,"book"),(4,"tv"),(1,"screen"),(5,"heater")),3)
val sortedRDD = randRDD.sortByKey()
sortRDD.filterByRange(1,3).collect
>>res4:Array[(Int,String)] = Array((1,screen),(2,cat),(3,book))
5.foreach 遍历RDD内每个记录
val c = sc.parallelize(List("cat","dog","tiger","lion","gnu"),3)
c.foreach(x => println(x + "is ym"))
>>lion is ym
gnu is ym
cat is ym
tiger is ym
dog is ym
6.foreachPartition 遍历RDD内每一个Partition(每个Partition对应一个值)
val b = sc.parallelize(List(1,2,3,4,5,6,7,8),3)
b.foreachPartition(x => println(x.reduce(_ + _ )))
>> 6
15
15
7.fullOuterJoin
rdd1.fullOuterJoin[rdd2] 对两个PairRDD进行外连接 ,相同的key值的全部value组合,没有相同key的也保留,值用None填充
val pairRDD1 = sc.parallelize(List(("cat",2),("cat",5),("book",40)))
val pairRDD2 = sc.parallelize(List(("cat",2),("cup",5),("book",40)))
pairRDD1.fullOuterJoin(pairRDD2).collect
>>res5: Array[(String,(Option[Int],Option[Int]))] = Array((book,(Some(40),Some(40))), (cup,(None,Some(5))), (cat,(Some(2),Some(2))), (cat,(Some(5),Some(2)))
8.groupBy 根据给定的规则 来分组
val a = sc.parallelize(1 to 9,3)
a.groupBy(x => {if (x % 2 == 0) "even" else "odd" }).collect
>> res6:Array[(String,Seq[Int])] = Array((even,ArrayBuffer(2,4,6,8)),(odd,ArrayBuffer(1,3,5,7,9)))
groupBy中使用的方法函数写法还可写作:
def myfunc(a:Int):Int =
{
a % 2
}
a.groupBy(myfunc).collect
或
def myfunc(a:Int):Int=
{
a % 2
}
a.groupBy(x => myfunc(x),3).collect
a.groupBy(myfunc(_),1).collect
例 将groupBy的条件设置为 partition ,同时自定义数据分区的规则
package sometest
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
object SparkApplication{
def main(args:Array[String]){
val conf = new SparkConf()
val sc = new SparkContext(conf).setAppName("GroupPartition").setMaster("spark://master:7077")
val a = sc.parallelize(1 to 9 , 3)
val p = new MyPartitioner()
val b = a.groupBy((x:Int) => {x},p) //这里按照自定义分区规则P重新分区,然后groupBy
// b的形式为RDD[(Int,Iterable[Int])] 比如说 (1,CompactBuffer(1))
def myfunc(index:Int,iter:Iterator[(Int,Iterable[Int])]): Iterator[(Int,(Iterable[Int],Int))] = {
iter.map(a => (index,(a._2,a._1))) //a._2这种写法表示a中的第2个元素
}
val c = b.mapPartitionsWithIndex(myfunc)
println("This is Result for My :")
c.collect().foreach(println)
}
自定义分区规则
package sometest
import org.apache.spark.Partitioner
/**
*自定义数据分区规则
**/
class MyPartitioner extends Partitioner{
def numPartitions:Int = 2 //设置分区数
def getPartition(key:Any):Int =
{
val code = key match
{
case null => 0
case key:Int => key % numPartitions //取余
case _ => key.hashCode % numPartitions
}
if(code < 0 ){ // 对 hashCode为负数的结果进行处理
code + numPartitions
}
else{
code
}
}
override def equals(other:Any):Boolean = // java标准的判断相等的函数, Spark内部比较两个RDD的分区是否一样时 会用到这个这个函数
{
other match
{
case h:MyPartitioner => h.numPartitions == numPartitions
case _ => false
}
}
}
打包成sparkAction.jar后 使用命令执行 spark-submit --class sometest.SparkApplication ~/sparkAction.jar
输出结果为:
This is Result for My :
(0,(CompactBuffer(4),4))
( 0,( CompactBuffer(6),6))
( 0,( CompactBuffer(8),8))
( 0,( CompactBuffer(2),2))
( 0,( CompactBuffer(1),1))
( 0,( CompactBuffer(3),3))
( 0,( CompactBuffer(7),7))
( 0,( CompactBuffer(9),9))
( 0,( CompactBuffer(5),5))
9.groupByKey [Pair]
类似于groupBy ,不过函数作用于key,而groupBy的函数是作用于每个数据的
val a = sc.parallelize(List("dog","tiger","lion","cat","spider","eagle"),2)
val b = a.keyBy(_.length)
b.groupByKey.collect
输出res11:Array[(Int,Iterable[String])] = Array((4,CompactBuffer(lion)),(6,CompactBuffer(spider)),(3,CompactBuffer(dog,cat)),(5,CompactBuffer(tiger,eagle)))
10 .histogram[Double] 计算数据直方图 (数值数据分布的精确图形表示)
计算给定数据中的最大值和最小值 ,然后将这个范围段平均分成n组,统计给定数据中每组的频数
一般来说,范围段为横轴 ,各组的统计个数为纵坐标
val a = sc.parallelize(List(1.1,1.2,1.3,2.0,2.1,7.4,7.5,7.6,8.8,9.0),3)
a.histogram(5) //将样本数据分成 5 组
res11: (Array[Double],Array[Long]) = (Array(1.1,2.68,4.26,5.84,7.42,9.0),Array(5,0,0,1,4))
11 .intersection 返回两个RDD的交集(内连接)
val x=sc.parallelize(1 to 20)
val y =sc.parallelize(10 to 30)
val z = x.intersection(y)
z.collect
res74: Array[Int] = Array(16,17,18,10,19,11,20,12,13,14,15)
内连接
val a = sc.parallelize(List("dog","salmon","salmon","rat","elephant"),3)
val b = a.keyBy(_.length) //Array[(Int,String)]=Array((3,dog),(3,rat),(6,salmon),(6(salmon),(8,elephant))
val c = sc.parallelize(List("dog","cat","gnu","salmon","rabbit","turkey","wolf",bear","bee"),3)
val d = c.keyBy(_.length)
b.join(d).collect
输出 res0: Array[(Int,(String,String))] = Array((6,(salmon,salmon)), (6,(salmon,rabbit)),(6,(salmon.turkey)), (6,(salmon,salmon)),
(6,(salmon,rabbit)), (6,(salmon,turkey)), (3,(dog,dog)), (3,(dog,cat)), (3,(dog,gnu)) ,(3,(dog,bee)), (3,(rat,dog)),(3,(rat,cat)), (3,(rat,gnu)), (,(rat,bee)))
12 .keys[Pair] 返回 key,value列表中的所有key
val a = sc.parallelize(List((3,"dog"),(5,"tiger"),(4,"lion"),(3,"cat"),(7,"panther"),(5,"eagle")),2)
a.keys.collect
res2: Array[Int] = Array(3,5,4,3,7,5)
13 . lookup 查找指定记录
val a = sc.parallelize(List((3,"dog"),(5,"tiger"),(4,"lion"),(3,"cat"),,(7,"panther"),(5,"eagle")),2)
a.lookup(5)
res8: Seq[String] = WrappedArray(tiger,eagle)
14 .max 返回最大值
借用上述的a
a.max
res9: (Int,String) = (7,panther)
val y =sc.parallelize(10 to 30)
y.max
res10: Int = 30
15 . mean 平均值
y.mean
res13: Double = 20.0
16 . persist,cache 设置RDD的存储级别
val c = sc.parallelize(List("Gnu","Cat","Rat","Dog","Gnu","Rat"),2)
c.getStorageLevel
res14: org.apache.spark.storage.StorageLevel = StorageLevel(1 replicas)
c.cache
res15: c.type = ParallelCollectionRDD[41] at parallelize at <console>:24
c.getStorageLevel
res16:org.apache.spark.storage.StorageLevel = StorageLevel(memory, deserialized, 1 replicas)
17 . sample 根据给定比例对数据进行采样
sample(withReplacement, fraction, seed)
withReplacement : 是否使用随机数替换
fraction : 对数据进行采样的比例
seed : 随机数生成器种子
val a = sc.parallelize(1 to 10000,3)
a.sample(false,0.1,0).count
res17:Long = 1032
a.sample(true,0.3,0).count
res18: Long = 3110
a.sample(true,0.3,13).count
res20 : Long = 2952
18 .saveAsTextFile保存到文本数据 (默认 文件系统是hdfs)
textFile读取文本数据
val a = sc.parallelize(11 to 19,3)
a.saveAsTextFile("test/tf") //实际上是保存到文件夹 test/tf ,由于并行化因子为3,一个Partition对应一个par-000x
val b = sc.textFile("test/tf")
b.collect
res4: Array[String] = Array(11,12,13,14,15,16,17,18,19)
19 .take 返回数据集中的前N个数据
val b = sc.parallelize(List("dog","cat","ape","salmon","gnu"),2)
b.take(2)
res5: Array[String] = Array(dog,cat)
20 .union,++ 对两个RDD数据进行并集 ,合并两个RDD
val a = sc.parallelize( 1 to 5,1)
val b = sc.parallelize(5 to 7,1)
(a++b).collect
Array[Int] = Array(1,2,3,4,5,5,6,7)