操作依赖什么规则来生成新的事件时间戳

Question

我在操作后加入两个流以创建一个新流。代码如下：

DataStream<NewTableA> join1 =
    oldTableADataStream
        .keyBy(t -> t.getFa3())
        .join(tableBDataStream)
        .where(new oldTableAKeySelector())
        .equalTo(new TableBKeySelector())
        .window(EventTimeSessionWindows.withGap(Time.milliseconds(WIN_GAP_TIME)))
        .allowedLateness(Time.milliseconds(allowedLateness))
        .apply(new oldTableAJoinTableBFunc());
        //.assignTimestampsAndWatermarks(new assignTSAndWMLastMax<>(maxOutOfOrderness));


   join1.process(
    new ProcessFunction<NewTableA, NewTableA>() {
      @Override
      public void processElement(NewTableA value, Context ctx, Collector<NewTableA> out)
          throws Exception {
        System.out.println(" NewTableA wmts:" + ctx.timerService().currentWatermark());
        System.out.println(" NewTableA ts:" + ctx.timestamp() + " " + value);
      }
    });

oldTableAJoinTableBFunc 的代码如下

public class oldTableAJoinTableBFunc implements JoinFunction<OldTableA, TableB, NewTableA> {



  @Override
  public NewTableA join(OldTableA oldTableA, TableB tableB) throws Exception {

    //System.out.println("join1 on");

    NewTableA newTableA = new NewTableA();

    newTableA.setPA1(oldTableA.getPa1());
    newTableA.setA2(oldTableA.getA2());
    newTableA.setFA3(oldTableA.getFa3());
    newTableA.setFA4(oldTableA.getFa4());
    newTableA.setB2(tableB.getB2());
    newTableA.setB3(tableB.getB3());
    // importance!!!
    newTableA.setTs(oldTableA.getTs());

    return newTableA;
  }
}

以上示例，oldTableADataStream 在活动时间加入 tableBDataStream 到 join1 流。

我发现了一个有趣的现象。 join1中事件的时间戳是flink自动创建的。

当我创建oldTableADataStream和tableBDataStream的测试数据时，我故意设置了所有的1000000010和1000000044。但是在join和apply函数之后，新流join1 中事件的时间戳被flink改变了，打印如下：

NewTableA wmts:**1000000042**
 NewTableA ts:**1000000143** NewTableA{PA1=10, A2='a20', FA3=21, B2='b21', B3='b31', FA4=39, C2='null', FC3=null, D2='null', D3='null', ts=**1000000011**}
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 NewTableA ts:**1000000143** NewTableA{PA1=1, A2='a20', FA3=20, B2='b20', B3='b30', FA4=30, C2='null', FC3=null, D2='null', D3='null', ts=**1000000010**}
 NewTableA wmts:1000000042
 NewTableA ts:1000000143 NewTableA{PA1=11, A2='a20', FA3=21, B2='b21', B3='b31', FA4=40, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA ts:1000000135 NewTableA{PA1=21, A2='a20', FA3=22, B2='b22', B3='b32', FA4=30, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA ts:1000000138 NewTableA{PA1=38, A2='a20', FA3=23, B2='b23', B3='b33', FA4=34, C2='null', FC3=null, D2='null', D3='null', ts=1000000015}
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 NewTableA ts:1000000138 NewTableA{PA1=57, A2='a20', FA3=24, B2='b24', B3='b34', FA4=36, C2='null', FC3=null, D2='null', D3='null', ts=1000000018}
 NewTableA ts:1000000143 NewTableA{PA1=2, A2='a20', FA3=20, B2='b20', B3='b30', FA4=31, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA wmts:1000000042
 NewTableA ts:1000000143 NewTableA{PA1=12, A2='a20', FA3=21, B2='b21', B3='b31', FA4=41, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 NewTableA ts:1000000135 NewTableA{PA1=22, A2='a20', FA3=22, B2='b22', B3='b32', FA4=31, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA wmts:1000000042
 NewTableA ts:1000000138 NewTableA{PA1=58, A2='a20', FA3=24, B2='b24', B3='b34', FA4=36, C2='null', FC3=null, D2='null', D3='null', ts=1000000018}
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 NewTableA ts:1000000143 NewTableA{PA1=13, A2='a20', FA3=21, B2='b21', B3='b31', FA4=42, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA ts:1000000138 NewTableA{PA1=39, A2='a20', FA3=23, B2='b23', B3='b33', FA4=34, C2='null', FC3=null, D2='null', D3='null', ts=1000000015}
 NewTableA wmts:1000000042
 NewTableA ts:1000000138 NewTableA{PA1=59, A2='a20', FA3=24, B2='b24', B3='b34', FA4=36, C2='null', FC3=null, D2='null', D3='null', ts=1000000018}
 NewTableA ts:1000000135 NewTableA{PA1=23, A2='a20', FA3=22, B2='b22', B3='b32', FA4=32, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA ts:1000000143 NewTableA{PA1=3, A2='a20', FA3=20, B2='b20', B3='b30', FA4=32, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 NewTableA ts:1000000143 NewTableA{PA1=14, A2='a20', FA3=21, B2='b21', B3='b31', FA4=43, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA ts:1000000138 NewTableA{PA1=40, A2='a20', FA3=23, B2='b23', B3='b33', FA4=34, C2='null', FC3=null, D2='null', D3='null', ts=1000000016}
 NewTableA ts:1000000135 NewTableA{PA1=24, A2='a20', FA3=22, B2='b22', B3='b32', FA4=33, C2='null', FC3=null, D2='null', D3='null', ts=1000000011}
 NewTableA wmts:1000000042
 NewTableA wmts:1000000042
 ......

似乎没有规则创建新事件的时间戳，1000000143和1000000138和1000000135等是如何计算的？好像和水印没什么关系，因为时间戳水印是1000000042，和同时事件的时间戳不同。

这个操作依赖什么规则来生成新的事件时间戳，我没找到官方的说明，谁能给个链接？

Answer 1

时间窗口创建的输出事件的时间戳设置为窗口内的最大时间戳。所以它基本上是包含事件的会话的会话边界。

由于您拥有基于 FA3 的键控会话窗口，因此您将获得取决于 FA3 的值：20 和 21 为 1000000143，22 为 1000000135，23 和 24 为 1000000138。