大数据——Flink Broadcast State 广播状态

Posted 2021-09-16 Vicky_Tang

一、BroadcastState 的介绍 

广播状态(Broadcast State)是 Operator State 的一种特殊类型。如果我们需要将配置 、规则等低吞吐事件流广播到下游所有 Task 时，就可以使用 BroadcastState。下游的 Task 接收这些配置、规则并保存为 BroadcastState，所有Task 中的状态保持一致，作用于另一个数据流的计算中。
简单理解：一个低吞吐量流包含一组规则，我们想对来自另一个流的所有元素基于此规则进行评估。
场景：动态更新计算规则。
广播状态与其他操作符状态的区别在于:

• 它有一个 map 格式，用于定义存储结构
• 它仅对具有广播流和非广播流输入的特定操作符可用
• 这样的操作符可以具有不同名称的多个广播状态

 

二、BroadcastState 操作流程

三、案例实现

• 从端口读取Json数据作为事件流
• 从mysql读取数据作为广播流
• 关联广播流和事件流
• 匹配对应的用户信息

package cn.kgc.broadcast

import java.sql.{Connection, DriverManager, PreparedStatement}

import com.alibaba.fastjson.JSON
import org.apache.flink.api.common.state.{BroadcastState, MapStateDescriptor}
import org.apache.flink.configuration.Configuration
import org.apache.flink.streaming.api.datastream.BroadcastStream
import org.apache.flink.streaming.api.functions.co.BroadcastProcessFunction
import org.apache.flink.streaming.api.functions.source.{RichParallelSourceFunction, SourceFunction}
import org.apache.flink.streaming.api.scala._
import org.apache.flink.util.Collector

// (001,'tom',18,'北京',15830010002)
// 定义样例类 接受 MySQL的用户数据
case class BaseUserInfo(id:Long,name:String,age:Int,city:String,phone:Long)

// user_id、user_name、user_addrss、behaviour、url
// 输出数据类型
case class UserVisitInfo(id:Long,name:String,city:String,behaviour:String,url:String)

// 实现广播ProcessFunction
class MyBroadcastFunc extends BroadcastProcessFunction[String,(Long, BaseUserInfo),UserVisitInfo]{

  lazy val mapStateDes = new MapStateDescriptor[Long, BaseUserInfo]("mapState",classOf[Long],classOf[BaseUserInfo])

  // 处理的是日志流中的每条数据
  override def processElement(value: String, ctx: BroadcastProcessFunction[String, (Long, BaseUserInfo), UserVisitInfo]#ReadOnlyContext, out: Collector[UserVisitInfo]): Unit = {
    // {"user_id":"001","ts":"2021-07-10 11:10:05","behaviour":"browse","url":"https://www.tb1.com/1.html"}
    val user_id = JSON.parseObject(value).getLong("user_id")
    val behaviour = JSON.parseObject(value).getString("behaviour")
    val url = JSON.parseObject(value).getString("url")

    val mapState = ctx.getBroadcastState(mapStateDes)
    val userInfo = mapState.get(user_id)

    out.collect(UserVisitInfo(user_id,userInfo.name,userInfo.city,behaviour,url))

  }

  // 处理的是广播流的每个值
  override def processBroadcastElement(value: (Long, BaseUserInfo), ctx: BroadcastProcessFunction[String, (Long, BaseUserInfo), UserVisitInfo]#Context, out: Collector[UserVisitInfo]): Unit = {
    val mapState: BroadcastState[Long, BaseUserInfo] = ctx.getBroadcastState(mapStateDes)
    mapState.put(value._1,value._2)
  }
}


class UserSourceFunc extends RichParallelSourceFunction[BaseUserInfo]{

  var conn:Connection = _
  var statement: PreparedStatement = _
  var flag:Boolean = true

  override def open(parameters: Configuration): Unit = {
    conn = DriverManager.getConnection("jdbc:mysql://localhost:3306/test?characterEncoding=utf-8&serverTimezone=UTC","root","liu911223")
    statement = conn.prepareStatement("select * from base_user")
  }

  override def run(ctx: SourceFunction.SourceContext[BaseUserInfo]): Unit = {
    while (flag){
      Thread.sleep(5000)
      val resultSet = statement.executeQuery()
      while (resultSet.next()){
        val id = resultSet.getLong(1)
        val name = resultSet.getString(2)
        val age = resultSet.getInt(3)
        val city = resultSet.getString(4)
        val phone = resultSet.getLong(5)
        ctx.collect(BaseUserInfo(id,name,age,city,phone))
      }
    }
  }

  override def cancel(): Unit = {
    flag = false
  }

  override def close(): Unit = {
    if (statement != null) statement.close()
    if (conn != null) conn.close()
  }
}
object BroadcastDemo01 {
  def main(args: Array[String]): Unit = {
    val env = StreamExecutionEnvironment.getExecutionEnvironment
    env.setParallelism(1)

    // 定义为KV，一方面是为了广播的时候定义为map，另一方面是为了做关联操作
    val userBaseDS: DataStream[(Long, BaseUserInfo)] = env.addSource(new UserSourceFunc)
      .map(user => (user.id, user))
    val mapStateDes = new MapStateDescriptor[Long, BaseUserInfo]("mapState",classOf[Long],classOf[BaseUserInfo])
    val broadCastStream: BroadcastStream[(Long, BaseUserInfo)] = userBaseDS.broadcast(mapStateDes)

    // 日志JSON数据
    val dataInfoDS: DataStream[String] = env.socketTextStream("master",1314)

    dataInfoDS.connect(broadCastStream)
      .process(new MyBroadcastFunc)
      .print()

    env.execute()
  }
}