executor: Improve the performance of WindowExec by using sliding window

[mmyj]

What problem does this PR solve?

Improve the performance of WindowExec by using sliding window or segment tree #12967

What is changed and how it works?

1 add a interface named SlidingWindowAggFunc for evaluating the window aggregate functions using sliding window
2 modify the rowFrameWindowProcessor.appendResult2Chunk to use sliding window

Check List

Tests

• No code

Code changes

• Has exported function/method change
• Has interface methods change

Benchmark

windowFuncs := []string{
	ast.AggFuncCount,
}
rows := []int{1000, 100000}
ndvs := []int{10, 1000}
frames := []*core.WindowFrame{
	{Type: ast.Rows, Start: &core.FrameBound{Type: ast.Preceding, Num: 10}, End: &core.FrameBound{Type: ast.Following, Num: 10}},
	{Type: ast.Rows, Start: &core.FrameBound{Type: ast.Preceding, Num: 100}, End: &core.FrameBound{Type: ast.Following, Num: 100}},
}

before

BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)-8         	    4304	    249372 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)#01-8      	    1572	    760644 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)-8       	    4851	    230780 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)#01-8    	    5337	    232417 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)-8       	      51	  22412645 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)#01-8    	       8	 139406682 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)-8     	      52	  22229838 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)#01-8  	      15	  74082167 ns/op

now

BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)-8         	   13956	     85242 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)#01-8      	   16231	     76437 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)-8       	    7408	    165960 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)#01-8    	    7059	    163368 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)-8       	     181	   6624173 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)#01-8    	     177	   6605687 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)-8     	     183	   6427620 ns/op
BenchmarkWindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)#01-8  	     206	   5707492 ns/op

benchstat

name                                                                       old time/op  new time/op  delta
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)-8          447µs ± 1%   127µs ± 3%  -71.68%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)#01-8      1.82ms ± 1%  0.11ms ± 3%  -93.99%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)-8        265µs ± 1%   237µs ± 1%  -10.58%  (p=0.000 n=10+9)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)#01-8     266µs ± 1%   238µs ± 1%  -10.48%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)-8       44.8ms ± 1%  10.6ms ± 2%  -76.24%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)#01-8     350ms ± 3%    11ms ± 1%  -96.88%  (p=0.000 n=9+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)-8     43.0ms ± 1%  10.6ms ± 1%  -75.35%  (p=0.000 n=9+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)#01-8   181ms ± 1%     9ms ± 1%  -94.81%  (p=0.000 n=10+9)

[SunRunAway]

It has some benchmarks for window functions in executor/benchmark_test.go, could you modify it and provide a benchmark result to show how performance we gain.

[SunRunAway]

/run-all-tests

[SunRunAway]

The rest LGTM.

[mmyj]

benchstat result

name                                                                       old time/op  new time/op  delta
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)-8          447µs ± 1%   127µs ± 3%  -71.68%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:1000)#01-8      1.82ms ± 1%  0.11ms ± 3%  -93.99%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)-8        265µs ± 1%   237µs ± 1%  -10.58%  (p=0.000 n=10+9)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:1000)#01-8     266µs ± 1%   238µs ± 1%  -10.48%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)-8       44.8ms ± 1%  10.6ms ± 2%  -76.24%  (p=0.000 n=10+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:10,_rows:100000)#01-8     350ms ± 3%    11ms ± 1%  -96.88%  (p=0.000 n=9+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)-8     43.0ms ± 1%  10.6ms ± 1%  -75.35%  (p=0.000 n=9+10)
WindowFunctionsWithSlidingWindow/(func:count,_ndv:1000,_rows:100000)#01-8   181ms ± 1%     9ms ± 1%  -94.81%  (p=0.000 n=10+9)

[SunRunAway]

LGTM

[SunRunAway]

@qw4990 @XuHuaiyu PTAL