spark2.x由浅入深深到底系列六之RDD java api详解三

学习任何spark知识点之前请先正确理解spark，可以参考：正确理解spark

本文详细介绍了spark key-value类型的rdd java api

一、key-value类型的RDD的创建方式
1、sparkContext.parallelizePairs

JavaPairRDD<String, Integer> javaPairRDD =
sc.parallelizePairs(Arrays.asList(new Tuple2("test", 3), new Tuple2("kkk", 3)));
//结果：[(test,3), (kkk,3)]
System.out.println("javaPairRDD = " + javaPairRDD.collect());

2、keyBy的方式

public class User implements Serializable {
private String userId;

private Integer amount;

public User(String userId, Integer amount) {
this.userId = userId;
this.amount = amount;
}

@Override
public String toString() {
return "User{" +
"userId='" + userId + '\'' +
", amount=" + amount +
'}';
}
}

JavaRDD<User> userJavaRDD = sc.parallelize(Arrays.asList(new User("u1", 20)));
JavaPairRDD<String, User> userJavaPairRDD = userJavaRDD.keyBy(new Function<User, String>() {
@Override
public String call(User user) throws Exception {
return user.getUserId();
}
});
//结果：[(u1,User{userId='u1', amount=20})]
System.out.println("userJavaPairRDD = " + userJavaPairRDD.collect());

3、zip的方式

JavaRDD<Integer> rdd = sc.parallelize(Arrays.asList(1, 1, 2, 3, 5, 8, 13));
//两个rdd zip也是创建key-value类型RDD的一种方式
JavaPairRDD<Integer, Integer> zipPairRDD = rdd.zip(rdd);
//结果：[(1,1), (1,1), (2,2), (3,3), (5,5), (8,8), (13,13)]
System.out.println("zipPairRDD = " + zipPairRDD.collect());

4、groupBy的方式

JavaRDD<Integer> rdd = sc.parallelize(Arrays.asList(1, 1, 2, 3, 5, 8, 13));
Function<Integer, Boolean> isEven = new Function<Integer, Boolean>() {
@Override
public Boolean call(Integer x) throws Exception {
return x % 2 == 0;
}
};
//将偶数和奇数分组，生成key-value类型的RDD
JavaPairRDD<Boolean, Iterable<Integer>> oddsAndEvens = rdd.groupBy(isEven);
//结果：[(false,[1, 1, 3, 5, 13]), (true,[2, 8])]
System.out.println("oddsAndEvens = " + oddsAndEvens.collect());
//结果：1
System.out.println("oddsAndEvens.partitions.size = " + oddsAndEvens.partitions().size());

oddsAndEvens = rdd.groupBy(isEven, 2);
//结果：[(false,[1, 1, 3, 5, 13]), (true,[2, 8])]
System.out.println("oddsAndEvens = " + oddsAndEvens.collect());
//结果：2
System.out.println("oddsAndEvens.partitions.size = " + oddsAndEvens.partitions().size());

二、combineByKey

JavaPairRDD<String, Integer> javaPairRDD =
sc.parallelizePairs(Arrays.asList(new Tuple2("coffee", 1), new Tuple2("coffee", 2),
new Tuple2("panda", 3), new Tuple2("coffee", 9)), 2);

//当在一个分区中遇到新的key的时候，对这个key对应的value应用这个函数
Function<Integer, Tuple2<Integer, Integer>> createCombiner = new Function<Integer, Tuple2<Integer, Integer>>() {
@Override
public Tuple2<Integer, Integer> call(Integer value) throws Exception {
return new Tuple2<>(value, 1);
}
};
//当在一个分区中遇到已经应用过上面createCombiner函数的key的时候，对这个key对应的value应用这个函数
Function2<Tuple2<Integer, Integer>, Integer, Tuple2<Integer, Integer>> mergeValue =
new Function2<Tuple2<Integer, Integer>, Integer, Tuple2<Integer, Integer>>() {
@Override
public Tuple2<Integer, Integer> call(Tuple2<Integer, Integer> acc, Integer value) throws Exception {
return new Tuple2<>(acc._1() + value, acc._2() + 1);
}
};
//当需要对不同分区的数据进行聚合的时候应用这个函数
Function2<Tuple2<Integer, Integer>, Tuple2<Integer, Integer>, Tuple2<Integer, Integer>> mergeCombiners =
new Function2<Tuple2<Integer, Integer>, Tuple2<Integer, Integer>, Tuple2<Integer, Integer>>() {
@Override
public Tuple2<Integer, Integer> call(Tuple2<Integer, Integer> acc1, Tuple2<Integer, Integer> acc2) throws Exception {
return new Tuple2<>(acc1._1() + acc2._1(), acc1._2() + acc2._2());
}
};

JavaPairRDD<String, Tuple2<Integer, Integer>> combineByKeyRDD =
javaPairRDD.combineByKey(createCombiner, mergeValue, mergeCombiners);
//结果：[(coffee,(12,3)), (panda,(3,1))]
System.out.println("combineByKeyRDD = " + combineByKeyRDD.collect());

combineByKey的数据流如下：



对于combineByKey的原理讲解详细见: spark core RDD api原理详解
三、aggregateByKey

JavaPairRDD<String, Tuple2<Integer, Integer>> aggregateByKeyRDD =
javaPairRDD.aggregateByKey(new Tuple2<>(0, 0), mergeValue, mergeCombiners);
//结果：[(coffee,(12,3)), (panda,(3,1))]
System.out.println("aggregateByKeyRDD = " + aggregateByKeyRDD.collect());
//aggregateByKey是由combineByKey实现的，上面的aggregateByKey就是等于下面的combineByKeyRDD
Function<Integer, Tuple2<Integer, Integer>> createCombinerAggregateByKey =
new Function<Integer, Tuple2<Integer, Integer>>() {
@Override
public Tuple2<Integer, Integer> call(Integer value) throws Exception {
return mergeValue.call(new Tuple2<>(0, 0), value);
}
};
//结果是： [(coffee,(12,3)), (panda,(3,1))]
System.out.println(javaPairRDD.combineByKey(createCombinerAggregateByKey, mergeValue, mergeCombiners).collect());

四、reduceByKey

JavaPairRDD<String, Integer> reduceByKeyRDD = javaPairRDD.reduceByKey(new Function2<Integer, Integer, Integer>() {
@Override
public Integer call(Integer value1, Integer value2) throws Exception {
return value1 + value2;
}
});
//结果：[(coffee,12), (panda,3)]
System.out.println("reduceByKeyRDD = " + reduceByKeyRDD.collect());
//reduceByKey底层也是combineByKey实现的，上面的reduceByKey等于下面的combineByKey
Function<Integer, Integer> createCombinerReduce = new Function<Integer, Integer>() {
@Override
public Integer call(Integer integer) throws Exception {
return integer;
}
};
Function2<Integer, Integer, Integer> mergeValueReduce =
new Function2<Integer, Integer, Integer>() {
@Override
public Integer call(Integer integer, Integer integer2) throws Exception {
return integer + integer2;
}
};
//结果：[(coffee,12), (panda,3)]
System.out.println(javaPairRDD.combineByKey(createCombinerReduce, mergeValueReduce, mergeValueReduce).collect());

五、foldByKey

JavaPairRDD<String, Integer> foldByKeyRDD = javaPairRDD.foldByKey(0, new Function2<Integer, Integer, Integer>() {
@Override
public Integer call(Integer integer, Integer integer2) throws Exception {
return integer + integer2;
}
});
//结果：[(coffee,12), (panda,3)]
System.out.println("foldByKeyRDD = " + foldByKeyRDD.collect());
//foldByKey底层也是combineByKey实现的，上面的foldByKey等于下面的combineByKey
Function2<Integer, Integer, Integer> mergeValueFold =
new Function2<Integer, Integer, Integer>() {
@Override
public Integer call(Integer integer, Integer integer2) throws Exception {
return integer + integer2;
}
};

Function<Integer, Integer> createCombinerFold = new Function<Integer, Integer>() {
@Override
public Integer call(Integer integer) throws Exception {
return mergeValueFold.call(0, integer);
}
};
//结果：[(coffee,12), (panda,3)]
System.out.println(javaPairRDD.combineByKey(createCombinerFold, mergeValueFold, mergeValueFold).collect());

六、groupByKey

JavaPairRDD<String, Iterable<Integer>> groupByKeyRDD = javaPairRDD.groupByKey();
//结果：[(coffee,[1, 2, 9]), (panda,[3])]
System.out.println("groupByKeyRDD = " + groupByKeyRDD.collect());
//groupByKey底层也是combineByKey实现的，上面的groupByKey等于下面的combineByKey
Function<Integer, List<Integer>> createCombinerGroup = new Function<Integer, List<Integer>>() {
@Override
public List<Integer> call(Integer integer) throws Exception {
List<Integer> list = new ArrayList<>();
list.add(integer);
return list;
}
};
Function2<List<Integer>, Integer, List<Integer>> mergeValueGroup = new Function2<List<Integer>, Integer, List<Integer>>() {
@Override
public List<Integer> call(List<Integer> integers, Integer integer) throws Exception {
integers.add(integer);
return integers;
}
};
Function2<List<Integer>, List<Integer>, List<Integer>> mergeCombinersGroup =
new Function2<List<Integer>, List<Integer>, List<Integer>>() {
@Override
public List<Integer> call(List<Integer> integers, List<Integer> integers2) throws Exception {
integers.addAll(integers2);
return integers;
}
};
//结果：[(coffee,[1, 2, 9]), (panda,[3])]
System.out.println(javaPairRDD.combineByKey(createCombinerGroup, mergeValueGroup, mergeCombinersGroup).collect());

对于api原理性的东西很难用文档说明清楚，如果想更深入，更透彻的理解api的原理，可以参考： spark core RDD api原理详解