flink使用sum是怎么计算的

代码

object ReduceAgg {
  def main(args: Array[String]) {
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    val tup: DataStream[(Int, Int)] = env.fromElements((1,2),(2,5),(2,3),(10,3),(2,8),(3,6))

    val gpT: KeyedStream[(Int, Int), Tuple] = tup.keyBy(1)
//    gpT.print()
    //gpT.print()
//    val reduce: DataStream[(Int, Int)] = gpT.reduce((before, after) => {
//      (before._1, before._2 + after._2)
//    })
    val sum: DataStream[(Int, Int)] = gpT.sum(1)
    sum.print()
    env.execute()
//    gpT.reduce((left,right)=>{
//      (left._1,right._1+right._2)
//    })
  }
}

结果

4> (1,2)
4> (2,5)
4> (2,3)
4> (2,6)
1> (2,8)
1> (3,6)

（2,3）与（2,5）不会做滚动求和吗，最不能理解的是(2,6)是怎么来的

应该要keyBy(0)