Flink 中的Window

1 Flink 中的Window 概述

   streaming 流式计算是一种被设计用于处理无限数据集的数据处理引擎，而无限数据集是指一种不断增长的本质上无限的数据集，而 window 是一种切割无限数据为有限块进行处理的手段。

 

2 Window 可以分成两类：

Ø CountWindow：根据窗口中相同 key 元素的数量来触发执行，执行时只计算元素数量达到窗口大小的 key 对应的结果。

Ø TimeWindow：将指定时间范围内的所有数据组成一个 window，一次对一个window 里面的所有数据进行计算

3 TimeWindow 分为三类：

滚动窗口（Tumbing):依据固定的窗口长度对数据进行切片，一个数据可以被统计一次，所有的数据只能落在一个窗口里面。适用场景：适合做 BI 统计等（做每个时间段的聚合计算）

滑动窗口（Sliding）：固定的窗口长度和滑动间隔，一个数据可以被统计多次，元素会被分配到多个窗口中。 适用场景：对最近一个时间段内的统计（求某接口最近 5min 的失败率来决定是否要报警）。

会话窗口（Session)：不会有重叠和固定的开始时间和结束时间的情况，当它在一个固定的时间周期内不再收到元素，即非活动间隔产生，那这个窗口就会关闭

  CountWindow

滚动窗口:默认的 CountWindow 是一个滚动窗口，只需要指定窗口大小即可，当相同key元素数量达到窗口大小时，就会触发窗口的执行

滑动窗口:和滚动窗口的函数名是完全一致的，只是在传参数时需要传入两个参数，一个是 window_size，一个是 sliding_size。

下面代码中的 sliding_size 设置为了 2，也就是说，每收到两个相同 key 的数据就计算一次，每一次计算的 window 范围是 5 个元素。

4 window function

定义对窗口收集数据的计算操作

增量聚合函数（incremental aggregation functions）

每条数据到来就进行计算，保持一个简单的状态。典型的增量聚合函数有ReduceFunction, AggregateFunction。

全窗口函数（full window functions）

先把窗口所有数据收集起来，等到计算的时候会遍历所有数据。ProcessWindowFunction 就是一个全窗口函数。

 

5 代码

滚动窗口

package chapter3

import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.streaming.api.scala.{DataStream, KeyedStream, StreamExecutionEnvironment, WindowedStream}
import org.apache.flink.api.scala._
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.windows.TimeWindow
object Window_Tumbling_Demo {
  def main(args: Array[String]): Unit = {
    //创建程序入口
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    //接收数据
    val data: DataStream[String] = env.socketTextStream("hadoop001",9999)
    //切分数据
    val spliFile: DataStream[String] = data.flatMap(_.split(" "))
    //每个单词记为1次
    val wordAndOne: DataStream[(String, Int)] = spliFile.map((_,1))
    //分流
    val keyed: KeyedStream[(String, Int), Tuple] = wordAndOne.keyBy(0)
    //指定时间窗口
    val window: WindowedStream[(String, Int), Tuple, TimeWindow] = keyed.timeWindow(Time.seconds(5))
    //对window里的数据进行聚合
    val wordAndCount: DataStream[(String, Int)] = window.sum(1)
    //打印输出
    wordAndCount.print()
    //调用execute方法
    env.execute()
  }

}

 

滑动窗口(重复消费)

package chapter3
import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.api.scala._
import org.apache.flink.streaming.api.scala.{DataStream, KeyedStream, StreamExecutionEnvironment, WindowedStream}
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.windows.TimeWindow
object Window_Sliding_Demo {
  def main(args: Array[String]): Unit = {
    //创建程序入口
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    //接收数据
    val data: DataStream[String] = env.socketTextStream("hadoop001",9999)
    //切分
    val spliFile: DataStream[String] = data.flatMap(_.split(" "))
    //记为1次
    val wordAndOne: DataStream[(String, Int)] = spliFile.map((_,1))
    //分流
    val keyed: KeyedStream[(String, Int), Tuple] = wordAndOne.keyBy(0)
    //指定窗口
    val window: WindowedStream[(String, Int), Tuple, TimeWindow] = keyed.timeWindow(Time.seconds(15),Time.seconds(5))
    //聚合
    val wordAndCount: DataStream[(String, Int)] = window.sum(1)
    //打印输出
    wordAndCount.print()
    //调用execute方法
    env.execute()
  }

}

 

 count滚动窗口 

package chapter3
import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.api.scala._
import org.apache.flink.streaming.api.scala.{DataStream, KeyedStream, StreamExecutionEnvironment, WindowedStream}
import org.apache.flink.streaming.api.windowing.windows.GlobalWindow
object CountWindow_Tumbling_Demo {
  def main(args: Array[String]): Unit = {
    //创建程序入口
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    //接收数据
    val data: DataStream[String] = env.socketTextStream("hadoop001",9999)
    //切分
    val spliFile: DataStream[String] = data.flatMap(_.split(" "))
    //每个单词记为1次
    val wordAndOne: DataStream[(String, Int)] = spliFile.map((_,1))
    //分流
    val keyed: KeyedStream[(String, Int), Tuple] = wordAndOne.keyBy(0)
    //指定窗口
    val window: WindowedStream[(String, Int), Tuple, GlobalWindow] = keyed.countWindow(3)
    //聚合
    val wordAndCount: DataStream[(String, Int)] = window.sum(1)
    //打印输出
    wordAndCount.print()
    //调用execute方法
    env.execute()
  }

count滑动窗口  size=5,代表最大输出5，slide=2 统计两个输出一次

package chapter3
import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.api.scala._
import org.apache.flink.streaming.api.scala.{DataStream, KeyedStream, StreamExecutionEnvironment, WindowedStream}
import org.apache.flink.streaming.api.windowing.windows.GlobalWindow
object CountWindow_Sliding_Demo {
  def main(args: Array[String]): Unit = {
    //创建程序入口
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    //接收数据
    val data: DataStream[String] = env.socketTextStream("node01",9999)
    //切分
    val spliFile: DataStream[String] = data.flatMap(_.split(" "))
    //每个单词记为1
    val wordAndOne: DataStream[(String, Int)] = spliFile.map((_,1))
    //分流
    val keyed: KeyedStream[(String, Int), Tuple] = wordAndOne.keyBy(0)
    //指定窗口
    val window: WindowedStream[(String, Int), Tuple, GlobalWindow] = keyed.countWindow(5,2)
    //聚合
    val wordAndCount: DataStream[(String, Int)] = window.sum(1)
    //打印输出
    wordAndCount.print()
    //调用execute方法
    env.execute()
  }

}

 增量聚合函数

来一条数据计算一条  ，value1,value2 代表接收两元组

package chapter3
import org.apache.flink.api.common.functions.ReduceFunction
import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.api.scala._
import org.apache.flink.streaming.api.scala.{DataStream, KeyedStream, StreamExecutionEnvironment, WindowedStream}
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.windows.TimeWindow
object Reduce_Demo {
  def main(args: Array[String]): Unit = {
    //创建程序入口
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    //接收数据
    val data: DataStream[String] = env.socketTextStream("hadoop001",8888)
    //切分
    val spliFile: DataStream[String] = data.flatMap(_.split(" "))
    //每个单词记为1次
    val wordAndOne: DataStream[(String, Int)] = spliFile.map((_,1))
    //分流
    val keyed: KeyedStream[(String, Int), Tuple] = wordAndOne.keyBy(0)
    //指定时间窗口
    val window: WindowedStream[(String, Int), Tuple, TimeWindow] = keyed.timeWindow(Time.seconds(5))
    //使用增量聚合函数进行聚合
    val wordAndCount: DataStream[(String, Int)] = window.reduce(new ReduceFunction[(String, Int)] {
      override def reduce(value1: (String, Int), value2: (String, Int)): (String, Int) = {
        (value1._1, value1._2 + value2._2)
      }
    })
    //打印输出
    wordAndCount.print()
    //调用execute方法
    env.execute()
  }

}

 

全窗口函数

package chapter3
import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.api.scala._
import org.apache.flink.streaming.api.TimeCharacteristic
import org.apache.flink.streaming.api.scala.function.ProcessWindowFunction
import org.apache.flink.streaming.api.scala.{DataStream, KeyedStream, StreamExecutionEnvironment, WindowedStream}
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.windows.TimeWindow
import org.apache.flink.util.Collector
object ProcessWindow_Demo {
  def main(args: Array[String]): Unit = {
    //创建程序入口
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    //接收数据
    val data: DataStream[String] = env.socketTextStream("hadoop001",8888)
    //切分
    val spliFile: DataStream[String] = data.flatMap(_.split(" "))
    //每个单词记为1次
    val wordAndOne: DataStream[(String, Int)] = spliFile.map((_,1))
    //分流，必须知道确定类型
    val keyed = wordAndOne.keyBy(_._1)
    //指定窗口
    val window = keyed.timeWindow(Time.seconds(5))
    //调用全窗口函数 in out key window 类型
    val wordAndCount: DataStream[(String, Int)] = window.process(new ProcessWindowFunction[(String, Int), (String, Int), String, TimeWindow] {
      override def process(key: String, context: Context, elements: Iterable[(String, Int)], out: Collector[(String, Int)]): Unit = {
        var sum = 0
        for (i <- elements) {
          sum += i._2
        }
        //输出之前收集结果
        out.collect(key, sum)
      }
    })
    //打印输出
    wordAndCount.print()
    //调用execute方法
    env.execute()
  }

}