Flink实时计算pv、uv的几种方法

实时统计pv、uv是再常见不过的大数据统计需求了，前面出过一篇SparkStreaming实时统计pv,uv的案例，这里用Flink实时计算pv,uv。

我们需要统计不同数据类型每天的pv，uv情况,并且有如下要求.

• 每秒钟要输出最新的统计结果；
• 程序永远跑着不会停,所以要定期清理内存里的过时数据；
• 收到的消息里的时间字段并不是按照顺序严格递增的,所以要有一定的容错机制；
• 访问uv并不一定每秒钟都会变化,重复输出对IO是巨大的浪费,所以要在uv变更时在一秒内输出结果,未变更时不输出；

Flink数据流上的类型和操作

DataStream是flink流处理最核心的数据结构，其它的各种流都可以直接或者间接通过DataStream来完成相互转换，一些常用的流直接的转换关系如图：

可以看出，DataStream可以与KeyedStream相互转换，KeyedStream可以转换为WindowedStream，DataStream不能直接转换为WindowedStream，WindowedStream可以直接转换为DataStream。各种流之间虽然不能相互直接转换，但是都可以通过先转换为DataStream，再转换为其它流的方法来实现。

在这个计算pv,uv的需求中就主要用到DataStream、KeyedStream以及WindowedStream这些数据结构。

这里需要用到window和watermark，使用窗口把数据按天分割，使用watermark可以通过“水位”来定期清理窗口外的迟到数据，起到清理内存的作用。

业务代码

我们的数据是json类型的，含有date,helperversion,guid这3个字段，在实时统计pv,uv这个功能中，其它字段可以直接丢掉，当然了在离线数据仓库中，所有有含义的业务字段都是要保留到hive当中的。
其它相关概念就不说了，会专门介绍，这里直接上代码吧。

<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
<modelVersion>4.0
56c
.0</modelVersion>

<groupId>com.ddxygq</groupId>
<artifactId>bigdata</artifactId>
<version>1.0-SNAPSHOT</version>

<properties>
<scala.version>2.11.8</scala.version>
<flink.version>1.7.0</flink.version>
<pkg.name>bigdata</pkg.name>
</properties>

<dependencies>
<dependency>
<groupId>org.apache.flink</groupId>
<artifactId>flink-scala_2.11</artifactId>
<version>{flink.version}</version>
</dependency>
<dependency>
<groupId>org.apache.flink</groupId>
<artifactId>flink-streaming-scala_2.11</artifactId>
<version>flink.version</version>
</dependency>

<dependency>
<groupId>org.apache.flink</groupId>
<artifactId>flink-streaming-java_2.11</artifactId>
<version>{flink.version}</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.apache.flink/flink-connector-kafka-0.8 -->
<dependency>
<groupId>org.apache.flink</groupId>
<artifactId>flink-connector-kafka-0.10_2.11</
56c
artifactId>
<version>flink.version</version>
</dependency>

<build>
<!--测试代码和文件-->
<!--<testSourceDirectory>{basedir}/src/test</testSourceDirectory>-->
<finalName>basedir/src/test</testSourceDirectory>−−><finalName>{pkg.name}</finalName>
<sourceDirectory>src/main/java</sourceDirectory>
<resources>
<resource>
<directory>src/main/resources</directory>
<includes>
<include>*.properties</include>
<include>*.xml</include>
</includes>
<filtering>false</filtering>
</resource>
</resources>
<plugins>
<!-- 跳过测试插件-->
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-surefire-plugin</artifactId>
<configuration>
<skip>true</skip>
</configuration>
</plugin>
<!--编译scala插件-->
<plugin>
<groupId>org.scala-tools&l
56c
t;/groupId>
<artifactId>maven-scala-plugin</artifactId>
<version>2.15.2</version>
<executions>
<execution>
<goals>
<goal>compile</goal>
<goal>testCompile</goal>
</goals>
</execution>
</executions>
</plugin>
</plugins>
</build>
</project>

主要代码，主要使用scala开发：

package com.ddxygq.bigdata.flink.streaming.pvuv

import java.util.Properties

import com.alibaba.fastjson.JSON
import org.apache.flink.runtime.state.filesystem.FsStateBackend
import org.apache.flink.streaming.api.CheckpointingMode
import org.apache.flink.streaming.api.functions.timestamps.BoundedOutOfOrdernessTimestampExtractor
import org.apache.flink.streaming.api.scala.{DataStream, StreamExecutionEnvironment}
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.triggers.ContinuousProcessingTimeTrigger
import org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer010
import org.apache.flink.streaming.util.serialization.SimpleStringSchema
imp
ad0
ort org.apache.flink.streaming.api.scala.extensions._
import org.apache.flink.api.scala._

/**
* @ Author: keguang
* @ Date: 2019/3/18 17:34
* @ version: v1.0.0
* @ description:
*/
object PvUvCount {
def main(args: Array[String]): Unit = {
val env = StreamExecutionEnvironment.getExecutionEnvironment

// 容错
env.enableCheckpointing(5000)
env.getCheckpointConfig.setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE)
env.setStateBackend(new FsStateBackend("file:///D:/space/IJ/bigdata/src/main/scala/com/ddxygq/bigdata/flink/checkpoint/flink/tagApp"))

// kafka 配置
val ZOOKEEPER_HOST = "hadoop01:2181,hadoop02:2181,hadoop03:2181"
val KAFKA_BROKERS = "hadoop01:9092,hadoop02:9092,hadoop03:9092"
val TRANSACTION_GROUP = "flink-count"
val TOPIC_NAME = "flink"
val kafkaProps = new Properties()
kafkaProps.setProperty("zookeeper.connect", ZOOKEEPER_HOST)
kafkaProps.setProperty("bootstrap.servers", KAFKA_BROKERS)
kafkaProps.setProperty("group.id", TRANSACTION_GROUP)

// watrmark 允许数据延迟时间
val MaxOutOfOrderness = 86400 * 1000L

// 消费kafka数据
val streamData: DataStream[(String, String, String)] = env.addSource(
new FlinkKafkaConsumer010[String](TOPIC_NAME, new SimpleStringSchema(), kafkaProps)
).assignTimestampsAndWatermarks(new BoundedOutOfOrdernessTimestampExtractor[String](Time.milliseconds(MaxOutOfOrderness)) {
override def extractTimestamp(element: String): Long = {
val t = JSON.parseObject(element)
val time = JSON.parseObject(JSON.parseObject(t.getString("message")).getString("decrypted_data")).getString("time")
time.toLong
}
}).map(x => {
var date = "error"
var guid = "error"
var helperversion = "error"
try {
val messageJsonObject = JSON.parseObject(JSON.parseObject(x).getString("message"))
val datetime = messageJsonObject.getString("time")
date = datetime.split(" ")(0)
// hour = datetime.split(" ")(1).substring(0, 2)
val decrypted_data_string = messageJsonObject.getString("decrypted_data")
if (!"".equals(decrypted_data_string)) {
val decrypted_data = JSON.parseObject(decrypted_data_string)
guid = decrypted_data.getString("guid").trim
helperversion = decrypted_data.getString("helperversion")
}
} catch {
case e: Exception => {
println(e)
}
}
(date, helperversion, guid)
})
// 这上面是设置watermark并解析json部分
// 聚合窗口中的数据，可以研究下applyWith这个方法和OnWindowedStream这个类
val resultStream = streamData.keyBy(x => {
x._1 + x._2
}).timeWindow(
1044
Time.days(1))
.trigger(ContinuousProcessingTimeTrigger.of(Time.seconds(1)))
.applyWith(("", List.empty[Int], Set.empty[Int], 0L, 0L))(
foldFunction = {
case ((_, list, set, _, 0), item) => {
val date = item._1
val helperversion = item._2
val guid = item._3
(date + "_" + helperversion, guid.hashCode +: list, set + guid.hashCode, 0L, 0L)
}
}
, windowFunction = {
case (key, window, result) => {
result.map {
case (leixing, list, set, _, _) => {
(leixing, list.size, set.size, window.getStart, window.getEnd)
}
}
}
}
).keyBy(0)
.flatMapWithState[(String, Int, Int, Long, Long),(Int, Int)]{
case ((key, numpv, numuv, begin, end), curr) =>

curr match {
case Some(numCurr) if numCurr == (numuv, numpv) =>
(Seq.empty, Some((numuv, numpv))) //如果之前已经有相同的数据,则返回空结果
case _ =>
(Seq((key, numpv, numuv, begin, end)), Some((numuv, numpv)))
}
}

// 最终结果
val resultedStream = resultStream.map(x => {
val keys = x._1.split("_")
val date = keys(0)
val helperversion = keys(1)
(date, helperversion, x._2, x._3)
})

resultedStream.print()
env.execute("PvUvCount")

}
}

使用List集合的size保存pv,使用Set集合的size保存uv，从而达到实时统计pv,uv的目的。
这里用了几个关键的函数：
applyWith：里面需要的参数，初始状态变量，和foldFunction ，windowFunction ；

存在的问题

显然，当数据量很大的时候，这个List集合和Set集合会很大，并且这里的pv是否可以不用List来存储，而是通过一个状态变量，不断做累加，对应操作就是更新状态来完成。

改进版

使用了一个计数器来存储pv的值。

packagecom.ddxygq.bigdata.flink.streaming.pvuv

import java.util.Properties

import com.alibaba.fastjson.JSON
import org.apache.flink.api.common.accumulators.IntCounter
import org.apache.flink.runtime.state.filesystem.FsStateBackend
import org.apache.flink.streaming.api.CheckpointingMode
import org.apache.flink.streaming.api.functions.timestamps.BoundedOutOfOrdernessTimestampExtractor
import org.apache.flink.streaming.api.scala.{DataStream, StreamExecutionEnvironment}
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.triggers.ContinuousProcessingTimeTrigger
import org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer010
import org.apache.flink.streaming.util.serialization.SimpleStringSchema
import org.apache.flink.streaming.api.scala.extensions._
import org.apache.flink.api.scala._
import org.apache.flink.core.fs.FileSystem

object PvUv2 {
def main(args: Array[String]): Unit = {
val env = StreamExecutionEnvironment.getExecutionEnvironment

// 容错
env.enableCheckpointing(5000)
env.getCheckpointConfig.setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE)
env.setStateBackend(new FsStateBackend("file:///D:/space/IJ/bigdata/src/main/scala/com/ddxygq/bigdata/flink/checkpoint/streaming/counter"))

// kafka 配置
val ZOOKEEPER_HOST = "hadoop01:2181,hadoop02:2181,hadoop03:2181"
val KAFKA_BROKERS = "hadoop01:9092,hadoop02:9092,hadoop03:9092"
val TRANSACTION_GROUP = "flink-count"
val TOPIC_NAME = "flink"
val kafkaProps = new Properties()
kafkaProps.setProperty("zookeeper.connect", ZOOKEEPER_HOST)
kafkaProps.setProperty("bootstrap.servers", KAFKA_BROKERS)
kafkaProps.setProperty("group.id", TRANSACTION_GROUP)

// watrmark 允许数据延迟时间
val MaxOutOfOrderness = 86400 * 1000L

val streamData: DataStream[(String, String, String)] = env.addSource(
new FlinkKafkaConsumer010[String](TOPIC_NAME, new SimpleStringSchema(), kafkaProps)
).assignTimestampsAndWatermarks(new BoundedOutOfOr
1044
dernessTimestampExtractor[String](Time.milliseconds(MaxOutOfOrderness)) {
override def extractTimestamp(element: String): Long = {
val t = JSON.parseObject(element)
val time = JSON.parseObject(JSON.parseObject(t.getString("message")).getString("decrypted_data")).getString("time")
time.toLong
}
}).map(x => {
var date = "error"
var guid = "error"
var helperversion = "error"
try {
val messageJsonObject = JSON.parseObject(JSON.parseObject(x).getString("message"))
val datetime = messageJsonObject.getString("time")
date = datetime.split(" ")(0)
// hour = datetime.split(" ")(1).substring(0, 2)
val decrypted_data_string = messageJsonObject.getString("decrypted_data")
if (!"".equals(decrypted_data_string)) {
val decrypted_data = JSON.parseObject(decrypted_data_string)
guid = decrypted_data.getString("guid").trim
helperversion = decrypted_data.getString("helperversion")
}
} catch {
case e: Exception => {
println(e)
}
}
(date, helperversion, guid)
})

val resultStream = streamData.keyBy(x => {
x._1 + x._2
}).timeWindow(Time.days(1))
.trigger(ContinuousProcessingTimeTrigger.of(Time.seconds(1)))
.applyWith(("", new IntCounter(), Set.empty[Int], 0L, 0L))(
foldFunction = {
case ((_, cou, set, _, 0), item) => {
val date = item._1
val helperversion = item._2
val guid = item._3
cou.add(1)
(date + "_" + helperversion, cou, set + guid.hashCode, 0L, 0L)
}
}
, windowFunction = {
case (key, window, result) => {
result.map {
case (leixing, cou, set, _, _) => {
(leixing, cou.getLocalValue, set.size, window.getStart, window.getEnd)
}
}
}
}
).keyBy(0)
.flatMapWithState[(String, Int, Int, Long, Long),(Int, Int)]{
case ((key, numpv, numuv, begin, end), curr) =>

curr match {
case Some(numCurr) if numCurr == (numuv, numpv) =>
(Seq.empty, Some((numuv, numpv))) //如果之前已经有相同的数据,则返回空结果
case _ =>
(Seq((key, numpv, numuv, begin, end)), Some((numuv, numpv)))
}
}

// 最终结果
val resultedStream = resultStream.map(x => {
val keys = x._1.split("_")
val date = keys(0)
val helperversion = keys(1)
(date, helperversion, x._2, x._3)
})

val resultPath = "D:\\space\\IJ\\bigdata\\src\\main\\scala\\com\\ddxygq\\bigdata\\flink\\streaming\\pvuv\\result"
resultedStream.writeAsText(resultPath, FileSystem.WriteMode.OVERWRITE)
env.execute("PvUvCount")

}
}

参考资料

https://flink.sojb.cn/dev/event_time.html
http://wuchong.me/blog/2016/05/20/flink-internals-streams-and-operations-on-streams
https://segmentfault.com/a/1190000006235690