flink广播状态详解

接下来来讲讲flink中的广播状态
/**
 * 搜索用户匹配行为 使用广播状态
 * 去匹配比如该用户是否是先login再buy 或者先login再pay这种组合行为
 */
case class Action(userid:String,action:String)
case class Pattern1(act1:String,act2:String)
object BroadcastStateExample {
  def main(args: Array[String]): Unit = {
    val env = StreamExecutionEnvironment.getExecutionEnvironment
    env.setParallelism(1)
    //定义一个实时的用户行为流
    val actionDS = env.fromElements(
      Action("zhangsan","login"),
      Action("zhangsan","buy"),
      Action("lisi","login"),
      Action("lisi","pay")
    )

    //定义一个行为模式流，代表了要检查的标准
    val patternDS = env.fromElements(
      Pattern1("login","buy"),
      Pattern1("login","pay")
    )

    //建造广播流的描述器 广播流底层是kv形式 所以也必须用mapState的解释器
    // 这里的两个类型分别指的是map的key和value的类型 因为这里无所谓key 所以也可以写成unit类和Pattern1类 写String也没事
    val msd = new MapStateDescriptor[String, Pattern1]("pattern", classOf[String], classOf[Pattern1])
    //按照描述器制造1个广播流
    val broadcastDS = patternDS.broadcast(msd)

    //将事件流和广播流链接在一起
    actionDS.keyBy(_.userid).connect(broadcastDS).process(new PatternFilter).print()

    env.execute()
  }
}
class PatternFilter extends KeyedBroadcastProcessFunction[String,Action,Pattern1,(String,Pattern1)]{
  //定义值状态 保存上一次的用户行为
  private lazy val preAct: ValueState[String] = getRuntimeContext.getState(new ValueStateDescriptor[String]("pre_action", classOf[String]))


  override def processElement(in1: Action, readOnlyContext: KeyedBroadcastProcessFunction[String, Action, Pattern1, (String, Pattern1)]#ReadOnlyContext, collector: Collector[(String, Pattern1)]): Unit = {
    //获得广播流
    val bcDS = readOnlyContext.getBroadcastState(new MapStateDescriptor[String,Pattern1]("pattern",classOf[String],classOf[Pattern1]))
    //查看用户属于哪种行为
    if(bcDS.contains(s"${preAct.value()}_${in1.action}")){
      collector.collect((in1.userid,bcDS.get(s"${preAct.value()}_${in1.action}")))
      //清除状态
      preAct.clear()
    }
    //记录本次用户行为
    preAct.update(in1.action)
  }

  //优先执行 广播流方法？因为要先加载完广播流
  override def processBroadcastElement(in2: Pattern1, context: KeyedBroadcastProcessFunction[String, Action, Pattern1, (String, Pattern1)]#Context, collector: Collector[(String, Pattern1)]): Unit = {
    //获取广播流
    val bcState = context.getBroadcastState(new MapStateDescriptor[String,Pattern1]("pattern",classOf[String],classOf[Pattern1]))
    //将你获得的规则数据存放到广播流中
    bcState.put(s"${in2.act1}_${in2.act2}",in2)
  }
}
flink广播状态详解

猜你喜欢
