我能和Flink CEP做一场懒惰的比赛吗？

Question

我想使用FlinkCEP只在模式上进行“懒惰”匹配。我该怎么做？例如，我有一个输入流ACABCABCB，我希望在一个followedBy C上匹配，只得到3个匹配，而不是6个匹配。

我创建了下面的示例来说明我的问题。

val env = StreamExecutionEnvironment.createLocalEnvironment(1)
env.setStreamTimeCharacteristic(TimeCharacteristic.ProcessingTime)

case class MyEvent(id: Int, kind: String, value: String)
case class MyAggregatedEvent(id: Int, concatenatedValue: String)

val eventStream = env.fromElements(
  MyEvent(1, "A", "1"), MyEvent(1, "C", "1"),
  MyEvent(1, "A", "2"), MyEvent(1, "B", "1"), MyEvent(1, "C", "2"),
  MyEvent(1, "A", "3"), MyEvent(1, "D", "2"), MyEvent(1, "C", "3"),
  MyEvent(1, "B", "3")
)

val pattern: Pattern[MyEvent, _] = Pattern
  .begin[MyEvent]("pA").where(e => e.kind == "A")
  .next("pC").where(e => e.kind == "C")
  .within(Time.seconds(5))

val patternNextStream: PatternStream[MyEvent] = CEP.pattern(eventStream.keyBy(_.id), pattern)

val outNextStream: DataStream[MyAggregatedEvent] = patternNextStream.flatSelect {
  (pattern: scala.collection.mutable.Map[String, MyEvent], collector: Collector[MyAggregatedEvent]) =>
    val partA = pattern.get("pA").get
    val partC = pattern.get("pC").get

    collector.collect(MyAggregatedEvent(partA.id, partA.value + "=>" + partC.value))
}
outNextStream.print()

env.execute("Experiment")

这给了我以下输出：

MyAggregatedEvent(1,1=>1)

当我将模式更改为：

val pattern: Pattern[MyEvent, _] = Pattern
  .begin[MyEvent]("pA").where(e => e.kind == "A")
  .followedBy("pC").where(e => e.kind == "C")
  .within(Time.seconds(5))

然后打印以下内容：

MyAggregatedEvent(1,1=>1) MyAggregatedEvent(1,1=>2) MyAggregatedEvent(1,2=>2) MyAggregatedEvent(1,1=>3) MyAggregatedEvent(1,2=>3) MyAggregatedEvent(1,3=>3)

如何创建一个只匹配每个事件一次的模式，以便我的输出是：

MyAggregatedEvent(1,1=>1) MyAggregatedEvent(1,2=>2) MyAggregatedEvent(1,3=>3)

Answer 1

目前，Flink的CEP库不支持这一点。匹配的语义还不能控制。我认为最好首先添加一个MATCH_ALL和一个匹配的MATCH_FIRST模式。一旦看到完全匹配的序列，MATCH_FIRST就会丢弃所有的中间状态。这应该涵盖您的用例。