Flink：窗口评估

Question

I am using window(SlidingEventTimeWindows.of(Time.seconds(120), Time.seconds(1))) sliding window for real_time streaming of data. I am also using the interface AssignerWithPeriodicWatermarks to extract the timestamps and assign watermarks. I am applying the custom method apply() on windowed stream.

In this scenario, what will trigger the window evaluation( apply() execution), is it happened after every 1 sec. which is equal to slide of window or is it after encountering the watermark in the stream?

Answer 1

A window is evaluated when its end timestamp is smaller than the current watermark. When a window operator receives a new watermarks, it checks all active windows and finalizes all windows whose end time is before the new watermark.

If you have a window that slides by 1 second and watermarks that arrive approximately every five seconds, each watermark will trigger the evaluation of five windows. You can think of the watermark as a mechanism to set the clock of the application. If the operator does not know that the clock advanced, it will not perform a computation.