3

我一直在尝试为 Window 的 apply() 方法编写自定义逻辑。基本上我想减少窗口中的所有元素，然后将时间戳附加到该值，所以我从 DataStream 创建了一个 WindowedStream，但是当我尝试为 apply() 定义函数时，它在编译时失败。

这是代码：

class WindowReduceFunction extends ReduceFunction[(Int, String, Int)] {
  override def reduce(a: (Int, String, Int), b: (Int, String, Int)) : (Int, String, Int) = {
    (a._1, a._2, a._3 + b._3)
  }
}

class WindowTimestampAddFunction extends WindowFunction[(Int, String, Int), (Int, String, Int, Long), (Int, String), TimeWindow] {
  override def apply(key : (Int, String), window : Window, in: Iterable[(Int, String, Int)], out: Collector[(Int, String, Int, Long)]) {
    for(row <- in) {
      out.collect((row._1, row._2, row._3, window.maxTimestamp()))
    }
  }
}

DataStream 的类型是 [Int, String, Int]，键是 [Int, String]。没有 apply() 的代码运行和编译没有错误，但是当我输入：

myWindowedStream.apply(new WindowReduceFunction(), new WindowTimestampAddFunction())

当它失败并且无法编译时，给出错误：

[ERROR]   [R](preAggregator: ((Int, String, Int), (Int, String, Int)) => (Int, String, Int), windowFunction: (org.apache.flink.api.java.tuple.Tuple, org.apache.flink.streaming.api.windowing.windows.TimeWindow, Iterable[(Int, String, Int)], org.apache.flink.util.Collector[R]) => Unit)(implicit evidence$6: org.apache.flink.api.common.typeinfo.TypeInformation[R])org.apache.flink.streaming.api.scala.DataStream[R] <and>
[ERROR]   [R](preAggregator: org.apache.flink.api.common.functions.ReduceFunction[(Int, String, Int)], function: org.apache.flink.streaming.api.scala.function.WindowFunction[(Int, String, Int),R,org.apache.flink.api.java.tuple.Tuple,org.apache.flink.streaming.api.windowing.windows.TimeWindow])(implicit evidence$5: org.apache.flink.api.common.typeinfo.TypeInformation[R])org.apache.flink.streaming.api.scala.DataStream[R]
[ERROR]  cannot be applied to (WindowReduceFunction, WindowTimestampAddFunction)
[ERROR]       .apply(new WindowReduceFunction(), new WindowTimestampAddFunction())
[ERROR]        ^
[ERROR] one error found
4

1 回答 1

3

您正在使用索引位置键，如 inkeyBy(1)或字段表达式键，如 in keyBy("field")。这意味着 key 的类型WindowedStream是类型Tuple（org.apache.flink.api.java.tuple.Tuple具体来说）。

如果您将您的第三个通用参数更改WindowFunction为Tuplefrom(Int, String)它应该可以工作。您还可以更改keyBy调用以使用 lambda 函数，然后您可以在WindowedStream. 例如：keyBy( in => (in._1, in._2)。

于 2016-04-28T14:56:29.507 回答