Flink Sliding窗口无法正常工作

Question

I have a datastream 1,2,3,4,5,6.....

Am applying sliding countWindow as below

inputStream.keyBy("id").countWindow( 2,1);

Expected Output

1,2

2,3

3,4 ..

Actual Output

1

1,2

2,3

3,4

Why does it slide first before accumulating the window size

Answer 1

First of all the expected output you provided is wrong. You specified a window size to be 2 minutes. So the output(assuming it is start and end of a window) should be:

1:00:00, 1:01:00
1:01:00, 1:02:00

The first event with timestamp 1:00:00 should be assigned to windows (0:59:00, 1:01:00) and (1:00:00, 1:02:00) . I believe that answers your question.

After edit:

For the countWindow the same rule is applied. The first element belongs to two windows. It is easier to reason with a countWindow(4,2) . Have a look at a basic example:

val sEnv = StreamExecutionEnvironment.getExecutionEnvironment
sEnv.setParallelism(1)

sEnv.fromCollection((1 to 10)).countWindowAll(4, 2).apply(
  (window, numbers, collector: Collector[Seq[Int]]) =>
    collector.collect(numbers.toSeq)
).print()

sEnv.execute()

The output is:

List(1, 2)
List(1, 2, 3, 4)
List(3, 4, 5, 6)
List(5, 6, 7, 8)
List(7, 8, 9, 10)

See that the first window that first element belongs starts in the past.

Answer 2

I understood thanks to Dawid Wysakowicz's answer. I just wanted to add a figure hoping it could help understanding.

Indeed, in sliding windows, each element has to be entailed into 2 windows. That is, the first element has to be in 2 windows as well.