当前位置:网站首页>Flink: from introduction to Zhenxiang (6. Flink implements UDF function - realizes more fine-grained control flow)
Flink: from introduction to Zhenxiang (6. Flink implements UDF function - realizes more fine-grained control flow)
2020-11-08 12:06:00 【osc_15vyay19】
Flink Provides a variety of data conversion operations , But in the actual business process, there are many data structures that need to be processed in business 、 Rules and so on , You need to write your own business code , It's used at this time flink Provided function class (Function Class)
Flink Exposed everything udf Function interface ( The implementation mode is interface or abstract class ), for example MapFunction,FilterFunction,ProcessFunction etc. .
A small chestnut , To filter the data to sensor3 Start with data
Still com.mafei.apitest Create a new one scala Object UDFTest1
The rest of the code is the same as before , Read the file and do some simple processing , A custom function class is added here MyFilterFunction, When use , Just add... To the logic .filter The method can ,
package com.mafei.apitest
import org.apache.flink.api.common.functions.{FilterFunction, ReduceFunction, RichFilterFunction}
import org.apache.flink.streaming.api.scala.{StreamExecutionEnvironment, createTypeInformation}
// Get sensor data
case class SensorReadingTest1(id: String,timestamp: Long, temperature: Double)
object UdfTest1 {
def main(args: Array[String]): Unit = {
// Create an execution environment
val env = StreamExecutionEnvironment.getExecutionEnvironment
case class Person(name: String, age: Int)
val inputStream= env.readTextFile("/opt/java2020_study/maven/flink1/src/main/resources/sensor.txt")
env.setParallelism(1)
// inputStream.print()
// First convert to sample class type
val dataStream = inputStream
.map(data => {
val arr = data.split(",") // according to , Split data , To get the results
SensorReadingTest1(arr(0), arr(1).toLong, arr(2).toDouble) // Generate data for a sensor class , Parameters are passed in the middle toLong and toDouble Because the default split is string category
// }).filter(new MyFilterFunction)
// }).filter(_.id.startsWith("sensor1")) // If it's very simple logic , You can also write anonymous classes like this , It's the same effect as writing a function
// }).filter(new RichFilterFunction[SensorReadingTest1] {
// override def filter(t: SensorReadingTest1): Boolean =
// t.id.startsWith("sensor3")
// }) // Anonymous class implementation effect , And above 2 The effects are the same
}).filter(new KeywordFilterFunction("sensor3")) // You can also pass in the parameters to be filtered
dataStream.print()
env.execute("udf test")
}
}
// Customize a function class , Filter it , Implement... In the interface filter The method can
class MyFilterFunction extends FilterFunction[SensorReadingTest1] {
override def filter(t: SensorReadingTest1): Boolean = t.id.startsWith("sensor3")
}
// Custom function class , Same as above , Added the transmission reference ,
class KeywordFilterFunction(keyword: String) extends FilterFunction[SensorReadingTest1]{
override def filter(t: SensorReadingTest1): Boolean =
t.id.startsWith(keyword)
}
Code structure and running effect diagram
RichMap
Mainly do some data processing and other operations , The code demonstrates MapperDemo and RichMapDemo The difference and operation effect of
package com.mafei.apitest
import org.apache.flink.api.common.functions.{FilterFunction, MapFunction, RichMapFunction}
import org.apache.flink.configuration.Configuration
import org.apache.flink.streaming.api.scala.{StreamExecutionEnvironment, createTypeInformation}
// Get sensor data
case class SensorReadingTest2(id: String,timestamp: Long, temperature: Double)
object UdfTest2 {
def main(args: Array[String]): Unit = {
// Create an execution environment
val env = StreamExecutionEnvironment.getExecutionEnvironment
case class Person(name: String, age: Int)
val inputStream= env.readTextFile("/opt/java2020_study/maven/flink1/src/main/resources/sensor.txt")
env.setParallelism(1)
// inputStream.print()
// First convert to sample class type
val dataStream = inputStream
.map(data => {
val arr = data.split(",") // according to , Split data , To get the results
SensorReadingTest2(arr(0), arr(1).toLong, arr(2).toDouble) // Generate data for a sensor class , Parameters are passed in the middle toLong and toDouble Because the default split is string category
}).map(new RichMapDemo())
dataStream.print()
env.execute("udf test")
}
}
class MapperDemo extends MapFunction[SensorReadingTest2, String]{
override def map(t: SensorReadingTest2): String = t.id+" Test to add some strings "
}
// Rich function , There are more classes than above open and close Other methods , Can do some database connection and other operations
class RichMapDemo extends RichMapFunction[SensorReadingTest2, String]{
// The main operations here are initialization , When starting the call , The whole process will only be called once , It is similar to the variables loaded by class initialization , Like database connection and so on
override def open(parameters: Configuration): Unit = {
println(" A database connection was made ..........")
// Get runtime context
getRuntimeContext()
}
// Every data goes through this method
override def map(in: SensorReadingTest2): String = in.id+" Test the rich function and add some strings "
override def close(): Unit = {
// Follow open similar , When the task stops , You can do something like release database connection and so on
print(" Closed database connection ......")
}
}
Running effect : You can see , The whole process , Only one database connection operation
A database connection was made ..........
sensor1 Test the rich function and add some strings
sensor2 Test the rich function and add some strings
sensor3 Test the rich function and add some strings
sensor4 Test the rich function and add some strings
sensor4 Test the rich function and add some strings
sensor4 Test the rich function and add some strings
Closed database connection ......
版权声明
本文为[osc_15vyay19]所创,转载请带上原文链接,感谢
边栏推荐
- PMP考试通过心得分享
- Flink从入门到真香(7、Sink数据输出-文件)
- Adobe Lightroom / LR 2021 software installation package (with installation tutorial)
- 维图PDMS切图软件
- 笔试面试题目:判断单链表是否有环
- Service architecture and transformation optimization process of e-commerce trading platform in mogujie (including ppt)
- It's 20% faster than python. Are you excited?
- On monotonous stack
- This time Kwai tiktok is faster than shaking.
- Flink's sink: a preliminary study
猜你喜欢
2018中国云厂商TOP5:阿里云、腾讯云、AWS、电信、联通 ...
Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
Is software testing training class easy to find a job
Implementation of verification code recognition in Python opencv pytesseract
Ali! Visual computing developer's series of manuals (with internet disk link)
Flink从入门到真香(6、Flink实现UDF函数-实现更细粒度的控制流)
Flink's sink: a preliminary study
Flink从入门到真香(7、Sink数据输出-文件)
Istio traffic management -- progress gateway
2天,利用下班后的4小时开发一个测试工具
随机推荐
Windows10关机问题----只有“睡眠”、“更新并重启”、“更新并关机”,但是又不想更新,解决办法
2 days, using 4 hours after work to develop a test tool
个人目前技术栈
Written interview topic: looking for the lost pig
华为云重大变革:Cloud&AI 升至华为第四大 BG ,火力全开
Adobe Lightroom /Lr 2021软件安装包(附安装教程)
Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
Ubuntu20.04下访问FTP服务器乱码问题+上传文件
Personal current technology stack
笔试面试题目:求丢失的猪
浅谈单调栈
Close to the double 11, he made up for two months and successfully took the offer from a large factory and transferred to Alibaba
虚拟机中安装 macOS 11 big sur
Understanding design patterns
来自朋友最近阿里、腾讯、美团等P7级Python开发岗位面试题
C language I blog assignment 03
Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
阿里教你深入浅出玩转物联网平台!(附网盘链接)
分布式文档存储数据库之MongoDB基础入门