pyspark之Structured Streaming文件file案例

# generate_file.py 
# 生成数据 生成500个文件,每个文件1000条数据
# 生成数据格式：eventtime name province action ()时间 用户名 省份 动作)
import os 
import time
import shutil
import time

FIRST_NAME = ['Zhao', 'Qian', 'Sun', 'Li', 'Zhou', 'Wu', 'Zheng', 'Wang']
SECOND_NAME = ['San', 'Si', 'Wu', 'Chen', 'Yang', 'Min', 'Jie', 'Qi']
PROVINCE = ['BeiJing', 'ShanDong', 'ShangHai', 'HeNan', 'HaErBin']
ACTION = ['login', 'logout', 'purchase']

PATH = "/opt/software/tmp/"
DATA_PATH = "/opt/software/tmp/data/"
    # 初始化环境
    def test_Setup():
        if os.path.exists(DATA_PATH):
            shutil.rmtree(DATA_PATH)
        os.mkdir(DATA_PATH)

    # 清理数据，恢复测试环境
    def test_TearDown():
        shutile.rmtree(DATA_PATH)

    # 数据保存文件
    def writeAndMove(filename,content):
        with open(PATH+filename,'wt',encoding='utf-8') as f:
            f.write(content)
        shutil.move(PATH+filename,DATA_PATH+filename)

if __name__ == '__main__':

    test_Setup()
    
    for i in range(500):
        filename = "user_action_{}.log".format(i)
        """
        验证spark输出模式，complete和update,增加代码，第一个文件i=0时，设置PROVINCE = "TAIWAN"
        """
        if i == 0:
            province= ['TaiWan']
        else:
            province = PROVINCE
        content = ""
        for _ in range(1000):
            content += "{} {} {} {}\n".format(str(int(time.time())),random.choice(FIRST_NAME)+random.choice(SECOND_NAME),random.choice(province),random.choice(ACTION))
        writeAndMove(filename,content)    
        time.sleep(10)



# spark_file_test.py  
# 读取DATA文件夹下面文件，按照省份统计数据，主要考虑window情况，按照window情况测试，同时针对    outputMode和输出console和mysql进行考虑，其中保存到mysql时添加batch字段

from pyspark.sql import SparkSession,DataFrame
from pyspark.sql.functions import split,lit,from_unixtime

DATA_PATH = "/opt/software/tmp/data/"

if __name__ == '__main__':
    spark = SparkSession.builder.getOrCreate()
    lines = spark.readStream.format("text").option("seq","\n").load(DATA_PATH)
    # 分隔符为空格
    userinfo = lines.select(split(lines.value," ").alias("info"))
    # 第一个为eventtime  第二个为name   第三个为province  第四个为action 
    # userinfo['info'][0]等同于userinfo['info'].getIterm(0)
    user = userinfo.select(from_unixtime(userinfo['info'][0]).alias('eventtime'),
        userinfo['info'][1].alias('name'),userinfo['info'][2].alias('province'),
        userinfo['info'][3].alias('action'))
    """
    测试1：数据直接输出到控制台，由于没有采用聚合，输出模式选择update
    user.writeStream.outputMode("update").format("console").trigger(processingTime="8 seconds").start().awaitTermination()
    """
    """
    测试2：数据存储到数据库，新建数据库表，可以通过printSchema()查看数据类型情况
    def insert_into_mysql_batch(df:DataFrame,batch):
        if df.count()>0:
        # 此处将batch添加到df中，采用lit函数
            data = df.withColumn("batch",lit(batch))
            data.write.format("jdbc"). \
        option("driver","com.mysql.jdbc.Driver"). \
        option("url","jdbc:mysql://localhost:3306/spark").option("user","root").\
        option("password","root").option("dbtable","user_log").\
        option("batchsize",1000).mode("append").save()    
        else:
            pass
user.writeStream.outputMode("update").foreachBatch((insert_into_mysql_batch)).trigger(processingTime="20 seconds").start().awaitTermination()
    """
    """
    测试3：数据按照省份统计后，输出到控制台，分析complete和update输出模式区别,针对该问题，调整输入，province="TaiWan"只会输入1次，即如果输出方式complete,则每batch都会输出，update的话，只会出现在一个batch
    userProvinceCounts = user.groupBy("province").count()
    userProvinceCounts = userProvinceCounts.select(userProvinceCounts['province'],userProvinceCounts["count"].alias('sl'))
# 测试输出模式complete:complete将总计算结果都进行输出
"""
batch 0
TaiWan 1000
batch 1
TaiWan 1000
其他省份  sl
batch 2
TaiWan 1000
其他省份  sl
"""    userProvinceCounts.writeStream.outputMode("complete").format("console").trigger(processingTime="20 seconds").start().awaitTermination() 
# 测试输出模式update:update只输出相比上个批次变动的内容(新增或修改)
batch 0
TaiWan 1000  
batch 1 中没有TaiWan输出
userProvinceCounts.writeStream.outputMode("complete").format("console").trigger(processingTime="20 seconds").start().awaitTermination() 
    """