Python大数据处理库 PySpark实战 总结三_pyspark实验总结

共享变量

• 广播变量 broadcast

  • 广播变量允许程序缓存一个只读变量在集群的每台机器上，而不是每个任务保存一个拷贝。借助广播变量，可以用一种更高效的方法来共享一些数据，比如一个全局配置文件。

  • from pyspark.sql import SparkSession
    spark = SparkSession.builder.master("local[*]").appName("RDD Demo").getOrCreate();
    sc = spark.sparkContext
    conf = {"ip":"192.168.1.1","key":"cumt"}
    #广播变量
    brVar =  sc.broadcast(conf)
    a = brVar.value #获取广播变量值
    print(a) #{'ip': '192.168.1.1', 'key': 'cumt'}
    print(a["key"]) #cumt
    brVar.unpersist() #更新广播变量
    conf["key"] = "jackwang"
    brVar =  sc.broadcast(conf) #再次广播
    a = brVar.value #获取广播新变量值
    print(a) #{'ip': '192.168.1.1', 'key': 'jackwang'}
    #destroy()可将广播变量的数据和元数据一同销毁，销毁后不能使用
    brVar.destroy()
    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16

• 累加器 accumulator

  • 只能利用关联操作做加操作的变数，能够快速执行操作，在调试时对作业的执行过程中的相关事件进行计数，不同节点上的计算任务都可以利用add方法 给累加器加值 为了保证准确性，只能使用一次动作操作，如果需要多细，则在RDD对象上执行cache或persist操作来切断依赖

  • rdd = sc.range(1,101)
    #创建累加器，初始值0
    acc = sc.accumulator(0)
    def fcounter(x):
    	global acc
      if x % 2 == 0 :
      acc += 1
      #unsupported operand type(s) for -=
      #acc -= 1
    rdd_counter =  rdd.map(fcounter)
    print(acc.value) #0 fcounter函数的逻辑还未执行
    #保证多次正确获取累加器值
    rdd_counter.persist()
    print(rdd_counter.count()) #100
    print(acc.value) #50
    print(rdd_counter.count()) #100
    print(acc.value) #50
    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17

DataFrames 与 Spark SQL

• 包括数据自身和结构信息
• 支持嵌套数据类型，array和map。
• 支持多种数据来源上构建

创建DataFrames

from pyspark.sql import SparkSession
spark = SparkSession.builder.master('local[1]').appName('DataFrames').getOrCreate()

#用list创建DataFrames
a = [('Jack', 32),('Smith', 33)]
df = spark.createDataFrame(a)
print(df.collect()) #[Row(_1='Jack', _2=32), Row(_1='Smith', _2=33)]
df.show()
+-----+---+
|   _1| _2|
+-----+---+
| Jack| 32|
|Smith| 33|
+-----+---+

# 添加结构信息
df2 = spark.createDataFrame(a, ['name', 'age'])
print(df2.collect())	#[Row(name='Jack', age=32), Row(name='Smith', age=33)]
df2.show() 
+-----+---+
| name|age|
+-----+---+
| Jack| 32|
|Smith| 33|
+-----+---+

#具有类型的DataFrame对象
from pyspark.sql.types import *
a = [('Jack', 32),('Smith', 33)]
rdd = sc.parallelize(a)
# StructField(name,类型，是否可以为空)
schema = StructType([
   StructField("name", StringType(), True),
   StructField("age", IntegerType(), True)])
df = spark.createDataFrame(rdd, schema)
# 简化版本
df2 = spark.createDataFrame(rdd,"name:string,age:int")
df.printSchema()
# root
#  |-- name: string (nullable = true)
#  |-- age: integer (nullable = true)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

Spark SQL基本用法

spark.sql 查询

spark.udf.register() 注册自定义的函数

df.select().where() 过滤

df.agg({字段名:方法}).show()

df.describe() 查看相关变量

spark.read.format(‘json’) .load() 读取文件

df.join(deptDf, peopleDf.deptId == deptDf.id,‘inner’)

df.na.fill({‘name’: ‘unknown’,‘salary’: 0, }).show() 空值填充

df.withColumn(“Timestamp”, df.Id.cast(“timestamp”)) 修改字段类型

# 查询
a = [('Jack', 32),('Smith', 33),('李四', 36)]
rdd = sc.parallelize(a)
df = spark.createDataFrame(rdd, "name: string, age: int")
df.createOrReplaceTempView("user") #创建临时表

df2 = spark.sql("select count(*) as counter from user")
df2.show()
+-------+
|counter|
+-------+
|      3|
+-------+
df2 = spark.sql("select *,age+1 as next from user where age < 36")
df2.show()
+-----+---+----+
| name|age|next|
+-----+---+----+
| Jack| 32|  33|
|Smith| 33|  34|
+-----+---+----+

# 自定义函数
strlen = spark.udf.register("strLen", lambda x: len(x)) #注册了一个自定义的函数
a = [('Jack', 32),('Smith', 33),('李四', 36)]
rdd = sc.parallelize(a)
df = spark.createDataFrame(rdd, "name: string, age: int")
df.createOrReplaceTempView("user")
df2 = spark.sql("select *,strLen(name) as len from user")
df2.show()
+-----+---+---+
| name|age|len|
+-----+---+---+
| Jack| 32|  4|
|Smith| 33|  5|
|  李四| 36|  2|
+-----+---+---+

# 选择查看特定列
df.select("name").show()

# 查找并过滤 
df.select("name").where(strlen("name")>2).show()
df.filter(df.age > 32).show()
 +-----+
 | name|
 +-----+
 | Jack|
 |Smith|
 +-----+

# 聚合找最大值 agg({字段名：集合函数})
df.agg({"age": "max"}).show()
 +--------+
 |max(age)|
 +--------+
 |      36|
 +--------+
  
# 查看字段个数、均值、方差、最大、最小值 
df.describe(['age']).show()
+-------+------------------+
|summary|               age|
+-------+------------------+
|  count|                 3|
|   mean|33.666666666666664|
| stddev| 2.081665999466133|
|    min|                32|
|    max|                36|
+-------+------------------+

# 读写取parquet文件 
df.write.parquet("myuser.parquet")
spark.read.parquet("myuser.parquet").show()
df.write.csv('user.csv','append')
spark.read.csv

# 读取json文件
df = spark.read.format('json') .load('hdfs://localhost:9000/user.json')
df.show()
+---+------+------+-----+------+
|age|deptId|gender| name|salary|
+---+------+------+-----+------+
| 32|    01|     男|   张三|  5000|
| 33|    01|     男|   李四|  6000|
| 38|    01|     女|   王五|  5500|
| 42|    02|     男| Jack|  7000|
| 27|    02|     女|Smith|  6500|
| 45|    02|     女| Lily|  9500|
+---+------+------+-----+------+
#打印字段的类型
print(df.dtypes)
[('age', 'bigint'), ('deptId', 'string'), ('gender', 'string'), ('name', 'string'), ('salary', 'bigint')]

# 透视 pivot
df2 = df.groupBy("deptId").pivot("gender") .sum("salary")
df2.show()
+------+-----+-----+
|deptId|    女|    男|
+------+-----+-----+
|    01| 5500|11000|
|    02|16000| 7000|
+------+-----+-----+

#条件选择
df.select("name",df.salary.between(6000,9500)).show()
df.select("name","age").where(df.name.like("Smi%")).show()

# 关联查询  求各个部门男女的平均工资 和最大年龄
# 用户表
a = [   
  ('01','张三', '男',32,5000),
  ('01','李四', '男',33,6000),
  ('01','王五', '女',38,5500),
  ('02','Jack', '男',42,7000),
  ('02','Smith', '女',27,6500),
  ('02','Lily', '女',45,9500)
]
rdd = sc.parallelize(a)
peopleDf = spark.createDataFrame(rdd, \
  "deptId:string,name:string,gender:string,age:int,salary:int")
# 部门表
b = [  
  ('01','销售部'),
  ('02','研发部')
]
rdd2 = sc.parallelize(b)
deptDf = spark.createDataFrame(rdd2, "id:string,name:string")

#join函数第三个参数默认为inner,其他选项为:
# inner, cross, outer, full, full_outer, left, left_outer, 
# right, right_outer, left_semi, and left_anti.
peopleDf.join(deptDf, peopleDf.deptId == deptDf.id,'inner') \
  .groupBy(deptDf.name, peopleDf.gender) \
  .agg({"salary": "avg", "age": "max"}) \
  .sort(deptDf.name, peopleDf.gender) \
  .show()
+----+------+-----------+--------+
|name|gender|avg(salary)|max(age)|
+----+------+-----------+--------+
| 研发部|     男|     7000.0|      42|
| 销售部|     男|     5500.0|      33|
| 销售部|     女|     5500.0|      38|
| 研发部|     女|     8000.0|      45|
+----+------+-----------+--------+


# 获取所有的列名
peopleDf.columns

# 去重
peopleDf.distinct().show()

# 删除数据列
peopleDf.drop("gender").show()

# 从一个DataFrame上移除另一个DataFrame
f1 = spark.createDataFrame(
        [("a", 1), ("a", 1), ("a", 1), ("a", 2), ("b",  3), ("c", 4)], ["C1", "C2"])
df2 = spark.createDataFrame([("a", 1), ("b", 3)], ["C1", "C2"])
df1.exceptAll(df2).show()
 +---+---+
 | C1| C2|
 +---+---+
 |  a|  1|
 |  a|  1|
 |  a|  2|
 |  c|  4|
 +---+---+

# 求交集
df1.intersectAll(df2).show()

# 空值替换
a = [   
  ('01','张三', '男',32,5000),
  ('01', None, '男',33,6000),
  ('01','王五', '女',36,None),
  ('02','Jack', '男',42,7000),
  ('02','Smith', '女',27,6500),
  ('02','Lily', '女',45,None),
]
rdd = sc.parallelize(a)
peopleDf = spark.createDataFrame(rdd,\
   "deptId:string,name:string,gender:string,age:int,salary:int")
# 将空值进行替换
peopleDf.na.fill({'name': 'unknown','salary': 0, }).show()
+------+-------+------+---+------+
|deptId|   name|gender|age|salary|
+------+-------+------+---+------+
|    01|     张三|     男| 32|  5000|
|    01|unknown|     男| 33|  6000|
|    01|     王五|     女| 36|     0|
|    02|   Jack|     男| 42|  7000|
|    02|  Smith|     女| 27|  6500|
|    02|   Lily|     女| 45|     0|
+------+-------+------+---+------+

# 转成JSON格式
peopleDf.toJSON().collect()
['{"deptId":"01","name":"张三","gender":"男","age":32,"salary":5000}',
 '{"deptId":"01","gender":"男","age":33,"salary":6000}',
 '{"deptId":"01","name":"王五","gender":"女","age":36}',
 '{"deptId":"02","name":"Jack","gender":"男","age":42,"salary":7000}',
 '{"deptId":"02","name":"Smith","gender":"女","age":27,"salary":6500}',
 '{"deptId":"02","name":"Lily","gender":"女","age":45}']

# 增加计算列 和 改名
peopleDf.withColumn("age2",peopleDf.age+1) \
        .withColumnRenamed("name","姓名") \
        .show()
    
# 日期处理
df = spark.createDataFrame(sc.parallelize([("2016-08-26",)]),"Id:string")
df.show()
+----------+
|        Id|
+----------+
|2016-08-26|
+----------+
df2 = df.withColumn("Timestamp", df.Id.cast("timestamp"))
df3 = df2.withColumn("Date", df.Id.cast("date"))
df3.show()
+----------+--------------------+----------+
|        Id|           Timestamp|      Date|
+----------+--------------------+----------+
|2016-08-26|2016-08-26 00:00:...|2016-08-26|
+----------+--------------------+----------+

df = spark.createDataFrame([('2020-05-10',),('2020-05-09',)], ['date'])
from pyspark.sql.functions import add_months
df.select(add_months(df.date, 1).alias('next_month')).show()
+----------+
|next_month|
+----------+
|2020-06-10|
|2020-06-09|
+----------+

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239

编写Spark程序并提交

这是依据蒙特卡洛模拟算法来计算圆周率，在一个变长为2的正方形内画圆，正方形的面积是4，圆的半径是1，面积是Pi*R^2=Pi 需要计算圆的面积，这里取圆心为坐标轴原点，在其中随机取点，共选n个点

from pyspark.sql import SparkSession
from random import random
from operator import add
spark = SparkSession.builder \
        .master("local[*]") \
        .appName("Pi Demo") \
        .getOrCreate();
sc = spark.sparkContext
#############################################
if __name__ == "__main__":
    n = 100000 * 20
    def f(_):
        x = random() * 2 - 1
        y = random() * 2 - 1
        return 1 if x ** 2 + y ** 2 <= 1 else 0
    count = sc.parallelize(range(1, n + 1), 20) .map(f).reduce(add)
    print("Pi is roughly %f" % (4.0 * count / n))
    spark.stop()

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

spark-submit重要参数

• –master local或者yarn或者spark://HOST:PORT
• –deploy-mode client(默认，在客户端上启动Driver，逻辑运行在Client上)或者cluster cluster不支持Python
• –class CLASS_NAME 针对Java和Scala程序
• –name NAME 区分不同程序
• –jars JARS 逗号隔开jar包，将程序代码即依赖资源打包成jar包
• –py-files PY_FILES 逗号隔开的.zip .egg .py文件，放置在PYTHONPATH下
• –driver-memory MEM 配置Driver内存