rdd

0.分类

算子就是分布式集合对象的api

rdd算子分为两类:1.transformation 2.action

https://blog.csdn.net/weixin_45271668/article/details/106441457

1 共享变量

https://blog.csdn.net/Android_xue/article/details/79780463

https://chowdera.com/2022/02/202202091419262471.html

两种共享变量:广播变量(broadcast variable)与累加器(accumulator)

广播变量解决了什么问题?
分布式集合RDD和本地集合进行关联使用的时候, 降低内存占用以及减少网络IO传输, 提高性能.

累加器解决了什么问题?
分布式代码执行中, 进行全局累加

1.广播变量

2.累加器

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
sc = spark.sparkContext
rdd1=sc.parallelize([1,2,3,4,5,6,7,8,9,10],2)
count=sc.accumulator(0)

def map_func(data):
global count
count+=data

# count = sc.accumulator(0)
# rdd3=rdd2.map(lambda x:x)
# rdd3.collect()
# print(count)
start_time = time.time()
result=rdd1.reduce(lambda a, b: a + b)
end_time = time.time()
print(result)
print(end_time - start_time)
# print(count)
start_time = time.time()
rdd2 = rdd1.map(map_func)
rdd2.collect()
end_time = time.time()
print(count)
print(end_time - start_time)
1
2
3
4
55
1.092909574508667
55
0.09459614753723145

累加器和reduce都可以得到聚合结果,效率???谁先执行 谁短,怎么衡量

2 *ByKey 操作

https://blog.csdn.net/weixin_43810802/article/details/120772452

https://blog.csdn.net/zhuzuwei/article/details/104446388

https://blog.csdn.net/weixin_40161254/article/details/103472056

Use reduceByKey instead of groupByKey

groupByKey creates a lot of shuffling which hampers the performance, while reduceByKey does not shuffle the data as much

3 reduce

1
2
3
4
5
6
7
in:
table=pd.DataFrame({"num":[1,1,2]})
table = spark.createDataFrame(table)
from pyspark.sql import Row
table.rdd.reduce(lambda a,b:Row(num=a[0]+b[0]))
out:
Row(num=4)

聚合的作用

注意点就是 a,b 操作后的数据类型和a,b保持一致,举个例子 a+b 和a ,b类型一致,否则(a+b)+c会报错

4 collect、 take、top、first,foreach

1 collect

collect()

返回包含此RDD中的所有元素的列表

注意:因为所有数据都会加载到driver,所有只适用于数据量不大的情况

2 first

first()

返回RDD中的第一个元素

3 take

take(num)

取RDD的前num个元素

4 top

top(num, key=None)

排序

Get the top N elements from an RDD

5 foreach

foreach(f)

Applies a function to all elements of this RDD 分区输出

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
def f(x):
print(x)

sc.parallelize([1,2,3,4,5]).foreach(f)

output:
4
3
2
1
5
会变
3
5
2
4
1

:D 一言句子获取中...