Flink CEP 不在事件时间工作但在处理时间工作

Flink CEP not Working in event time but working in Processing Time

当我使用 Flink CEP 代码处理时间(默认配置)时,我能够获得所需的模式匹配,但是在将环境配置为事件时间时,我无法获得任何模式匹配。

 def main(args: Array[String]): Unit = {
    val env = StreamExecutionEnvironment.getExecutionEnvironment
    env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime)
    env.enableCheckpointing(3000) // checkpoint every 3000 msec
     val lines = env.addSource(consumerKafkaSource.consume("bank_transaction_2", "192.168.2.201:9092", "192.168.2.201:2181", "http://192.168.2.201:8081"))

  val eventdate = ExtractAndAssignEventTime.assign(lines, "unix", "datetime", 3) //Extracting date time here

    val event = eventdate.keyBy(v => v.get("customer_id").toString.toInt)
   val pattern1 = Pattern.begin[GenericRecord]("start").where(v=>v.get("state").toString=="FAILED").next("d").where(v=>v.get("state").toString=="FAILED")
      val patternStream = CEP.pattern(event, pattern1)
    val warnID = patternStream.sideOutputLateData(latedata).select(value =>  {
      val v = value.mapValues(c => c.toList.toString)
      Json(DefaultFormats).write(v).replace("\\"", "\"")
        //.replace("List(","{").replace(")","}")
    })
    val latedatastream = warnID.getSideOutput(latedata)
    latedatastream.print("late_data")


    warnID.print("warning")
    event.print("event")

时间戳提取码

object ExtractAndAssignEventTime {
  def assign(stream:DataStream[GenericRecord],timeFormat:String,timeColumn:String,OutofOrderTime:Int ):DataStream[GenericRecord] ={
    if(!(timeFormat.equalsIgnoreCase("Unix"))){
      val EventTimeStream=stream.assignTimestampsAndWatermarks(new BoundedOutOfOrdernessTimestampExtractor[GenericRecord](Time.seconds(3)) {
        override def extractTimestamp(t: GenericRecord): Long = {
          new java.text.SimpleDateFormat(timeFormat).parse(t.get(timeColumn).toString).getTime
        }
      })
      EventTimeStream
    }
    else{
      val EventTimeStream=stream.assignTimestampsAndWatermarks(new BoundedOutOfOrdernessTimestampExtractor[GenericRecord](Time.seconds(OutofOrderTime)) {
        override def extractTimestamp(t: GenericRecord): Long = {
          (t.get(timeColumn).toString.toLong)
        }
      })
      EventTimeStream
    }
  }

请帮我解决这个问题。提前致谢!

既然你使用的是AssingerWithPeriodicWatermark,你还需要设置setAutowatermarkInterval,这样Flink才会使用这个区间来生成水印。

您可以通过调用 env.getConfig.setAutoWatermarkInterval([interval]) 来完成此操作。

Event Time CEP是基于Watermarks的,所以如果不生成那么基本上就没有输出。

我遇到了同样的问题,我刚刚“解决”了它,但答案没有多大意义(至少对我而言),如您所见。

解释:

在我的原始代码中,我有这个:

var env = StreamExecutionEnvironment.getExecutionEnvironment
env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime)
env.setParallelism(1)
env.getConfig.setAutoWatermarkInterval(1)

...

var stream : DataStream[String] = env.readTextFile("/home/luca/Desktop/input")
    
    
var tupleStream = stream.map(new S2TMapFunction())
tupleStream.assignTimestampsAndWatermarks(new PlacasPunctualTimestampAssigner())

val pattern = Pattern.begin[(String,Double,Double,String,Int,Int)]("follow").where(new SameRegionFunction())

val patternStream = CEP.pattern(newTupleStream,pattern)

val result = patternStream.process(new MyPatternProcessFunction())

根据我的日志记录,我看到 SameRegionFunctionMyPatternProcessFunction 都没有被执行,至少可以说这是非常出乎意料的。

答案:

因为我一无所知,所以我决定测试让我的流再通过一个转换函数,只是为了检查我的事件是否真的被插入到流中。所以,我提交 tupleStream 到一个映射操作,生成 newTupleStream,像这样:

var env = StreamExecutionEnvironment.getExecutionEnvironment
env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime)
env.setParallelism(1)
env.getConfig.setAutoWatermarkInterval(1)

...

var stream : DataStream[String] = env.readTextFile("/home/luca/Desktop/input")


/* I created 'DoNothingMapFunction', where the output event = input event*/
var tupleStream = stream.map(new S2TMapFunction())
var newTupleStream = tupleStream.assignTimestampsAndWatermarks(new PlacasPunctualTimestampAssigner()).map(new DoNothingMapFunction())


val pattern = Pattern.begin[(String,Double,Double,String,Int,Int)]("follow").where(new SameRegionFunction())

val patternStream = CEP.pattern(newTupleStream,pattern)

val result = patternStream.process(new MyPatternProcessFunction())

然后SameRegionFunctionMyPatternProcessFunction决定运行。

观测:

我改了行:

var newTupleStream = tupleStream.assignTimestampsAndWatermarks(new PlacasPunctualTimestampAssigner()).map(new DoNothingMapFunction())

对此:

var newTupleStream = tupleStream.assignTimestampsAndWatermarks(new PlacasPunctualTimestampAssigner())

而且它也有效。显然只是另一个级别的间接就足以让它工作,虽然我不清楚为什么会这样。