Flink实例（106）：自定义时间和窗口的操作符（十二）自定义窗口分配器 周、月

自定义 WindowAssigner

如果我们定义按天、小时、分钟的滚动窗口都很容易实现。

但是如果我们要定义一周(周日开始或周一)，一个月(1号开始)的滚动窗口，那么现有API基本没法实现或很难实现。

对此就需要我们实现一个自定义的窗口分配器。

package com.atguigu.exercise.ETLHIVE

import java.text.SimpleDateFormat
import java.util
import java.util.{Calendar, Collections, Date}

import com.meda.utils.DateHelper
import org.apache.flink.api.common.ExecutionConfig
import org.apache.flink.api.common.typeutils.TypeSerializer
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment
import org.apache.flink.streaming.api.windowing.assigners.WindowAssigner
import org.apache.flink.streaming.api.windowing.triggers.{EventTimeTrigger, Trigger}
import org.apache.flink.streaming.api.windowing.windows.TimeWindow

class CustomWindowAssigner [T](tag: String) extends WindowAssigner[T, TimeWindow]{
  //窗口分配的主要方法，需要为每一个元素指定所属的分区
  override def assignWindows(t: T, timestamp: Long, windowAssignerContext: WindowAssigner.WindowAssignerContext): util.Collection[TimeWindow] = {
    var offset: (Long, Long) = null
    tag match {
      case "month" => offset = getTimestampFromMon(timestamp)
      case "week" => offset = getTimestampFromWeek(timestamp)
    }
    //分配窗口
    Collections.singletonList(new TimeWindow(offset._1, offset._2))
  }

  //注意此处需要进行类型的转换，否则或编译出错，java版本好像没问题，但是java对于上面的offset处理有点难搞，所以放弃了
  override def getDefaultTrigger(streamExecutionEnvironment: StreamExecutionEnvironment): Trigger[T, TimeWindow] = EventTimeTrigger.create().asInstanceOf[Trigger[T, TimeWindow]]

  override def getWindowSerializer(executionConfig: ExecutionConfig): TypeSerializer[TimeWindow] = new TimeWindow.Serializer

  //是否使用事件时间
  override def isEventTime: Boolean = true

  /**
   * 获取指定时间戳当月时间戳范围
   * eg:2020-03-12 11:35:13 (timestamp=1583984113960l)
   * 结果为：(1582992000000,1585670399999)=>(2020-03-01 00:00:00,2020-03-31 23:59:59)
   *
   * @param timestamp 时间戳
   * @return
   */

  def getTimestampFromMon(timestamp: Long): (Long, Long) = {
    val calendar = Calendar.getInstance()
    calendar.setTime(DateHelper.getInstance().getDateFromStr(new SimpleDateFormat("yyyyMM01000000").format(new Date(timestamp)), "yyyyMMddHHmmss"))
    val numsOfMon: Long = calendar.getActualMaximum(Calendar.DAY_OF_MONTH)
    calendar.set(Calendar.DAY_OF_MONTH, 1)
    val start: Long = calendar.getTimeInMillis
    val end: Long = start + numsOfMon * 24 * 60 * 60 * 1000 - 1
    (start, end)
  }

  /**
   * 获取指定时间戳本周时间范围（从周日开始）
   * eg:2020-03-14 23:59:59 (timestamp=1583895064000l)
   * 结果为：(1583596800000,1584201599999)=>(2020-03-08 00:00:00,2020-03-14 23:59:59)
   *
   * @param timestamp 时间戳
   * @return
   */
  def getTimestampFromWeek(timestamp: Long): (Long, Long) = {
    val calendar = Calendar.getInstance()
    calendar.setTime(DateHelper.getInstance().getDateFromStr(new SimpleDateFormat("yyyyMMdd000000").format(new Date(timestamp)), "yyyyMMddHHmmss"))
    //    calendar.setFirstDayOfWeek(Calendar.SUNDAY)//设置周日为首日  默认值，一般不用设置
    calendar.set(Calendar.DAY_OF_WEEK, Calendar.SUNDAY)
    val start: Long = calendar.getTimeInMillis
    (start, start + 7 * 24 * 60 * 60 * 1000l - 1)
  }
}

---

//输入数据
case class Top100Input(event_id: String, date_d: String, timeStamp: Long, uid: Long, weekTag: String, monthTag: String)

//调用
val dStream: DataStream[Top100Input] = ...

dStream
      .keyBy(_.weekTag)
      .window(new CustomWindowAssigner[Top100Input]("week"))

dStream
      .keyBy(_.monthTag)
      .window(new CustomWindowAssigner[Top100Input]("month"))