spark一个重要的特性就是共享变量:
共享变量原理:
默认情况下,如果在一个算子的函数中使用到了某个外部的变量,那么这个变量的值会被拷贝到每个task中。此时每个task只能操作自己的那份变量副本。如果多个task想要共享某个变量,那么这种方式是做不到的。
Spark为此提供了两种共享变量,一种是Broadcast Variable(广播变量),另一种是Accumulator(累加变量)。Broadcast Variable会将使用到的变量,仅仅为每个节点拷贝一份,更大的用处是优化性能,减少网络传输以及内存消耗。Accumulator则可以让多个task共同操作一份变量,主要可以进行累加操作。
broadcast:通过调用SparkContext的broadcast()方法,来针对某个变量创建广播变量。然后在算子的函数内,使用到广播变量时,每个节点只会拷贝一份副本了。每个节点可以使用广播变量的value()方法获取值。记住,广播变量,是只读的。
Accumulator,主要用于多个节点对一个变量进行共享性的操作。Accumulator只提供了累加的功能。但是确给我们提供了多个task对一个变量并行操作的功能。但是task只能对Accumulator进行累加操作,不能读取它的值。只有Driver程序可以读取Accumulator的值。
.png
案例实战:
1、java版本:
broadcast:
package cn.spark.study.core;
import java.util.Arrays;
import java.util.List;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.Function;
import org.apache.spark.api.java.function.VoidFunction;
import org.apache.spark.broadcast.Broadcast;
public class broadcastdemo {
public static void main(String[] args) {
SparkConf conf = new SparkConf()
.setAppName("collectionparallelize")
.setMaster("local");
JavaSparkContext sc = new JavaSparkContext(conf);
List numbers = Arrays.asList(1,2,3,4,5,6,7,8,9,10);
JavaRDD numbersrdd = sc.parallelize(numbers);
int factor = 3;
final Broadcast broadcast = sc.broadcast(factor);
JavaRDD mutinumbersrdd = numbersrdd.map(new Function(){
private static final long serialVersionUID = 1L;
@Override
public Integer call(Integer v1) throws Exception {
return v1 * broadcast.value();
}
});
mutinumbersrdd.foreach(new VoidFunction(){
private static final long serialVersionUID = 1L;
@Override
public void call(Integer t) throws Exception {
System.out.println(t);
}
});
sc.close();
}
}
accmulator:
package cn.spark.study.core;
import java.util.Arrays;
import java.util.List;
import org.apache.spark.Accumulator;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.VoidFunction;
public class accumulatordemo {
public static void main(String[] args) {
SparkConf conf = new SparkConf()
.setAppName("collectionparallelize")
.setMaster("local");
JavaSparkContext sc = new JavaSparkContext(conf);
List numbers = Arrays.asList(1,2,3,4,5,6,7,8,9,10);
JavaRDD numbersrdd = sc.parallelize(numbers);
final Accumulator sum = sc.accumulator(0);
numbersrdd.foreach(new VoidFunction(){
private static final long serialVersionUID = 1L;
@Override
public void call(Integer t) throws Exception {
sum.add(t);
}
});
System.out.println(sum.value());
}
}
2、scala版本:
broadcast:
package com.spark.study.core
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
object accumulatordemo {
def main(args:Array[String]){
val conf = new SparkConf()
.setAppName("collectionparallelize")
.setMaster("local");
val sc = new SparkContext(conf);
val numbers = Array(1,2,3,4,5,6,7,8,9,10)
val numbersrdd = sc.parallelize(numbers, 1)
val sum = sc.accumulator(0)
val total = numbersrdd.foreach(num => sum+=num)
println(sum)
}
}
accumulator:
package com.spark.study.core
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
object broadcastdemo {
def main(args:Array[String]){
val conf = new SparkConf()
.setAppName("collectionparallelize")
.setMaster("local");
val sc = new SparkContext(conf);
val numbers = Array(1,2,3,4,5,6,7,8,9,10)
val numbersrdd = sc.parallelize(numbers, 1)
val factor =3
val sum = sc.broadcast(factor)
val total = numbersrdd.map(num => num * sum.value)
total.foreach(f=>println(f))
}
}
来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/30541278/viewspace-2153549/,如需转载,请注明出处,否则将追究法律责任。