Optimization for ValueListOutput

[CodingFabian]

There are many cases where a ValueListOutput is used when the amount of returned values is known

like com.lambdaworks.redis.RedisCommandBuilder.zrange(K, long, long) will have stop-start and
com.lambdaworks.redis.RedisCommandBuilder.hmget(K, K...) will have keys.length

For that we could add a second constructor that takes an additional capacity argument for the array list. backwards compatible would be to use "10" as capacity.

what do you think?

[mp911de]

I think there's some potential to improve. Redis array responses contain the size in their header. When using streaming, outputs don't really require an eager initialization with ArrayList because the result is streamed to a Subscriber interface instead of the list.

It would make sense to distinguish and have a late initialization that either initializes ArrayList with the response size as soon as we receive the response or, for streaming, do not create ArrayList instances at all.

[CodingFabian]

Good idea, the latter is what I did for our use case already as soon as I discovered it. I really like the layered API of lettuce, so I can easily choose convenience or maximum control and performance.

[mp911de]

Together with a few other optimizations (Returning Collections.emptyList() if array response is empty) we end up with the following benchmark:

4.4

Benchmark                                                     Mode  Cnt    Score    Error  Units
ValueListOutputBenchmark.measure100Elements                   avgt   40  444,795 ± 11,781  ns/op
ValueListOutputBenchmark.measure100ElementsWithResizeElement  avgt   40  568,426 ± 11,864  ns/op
ValueListOutputBenchmark.measure16Elements                    avgt   40   66,082 ±  1,858  ns/op
ValueListOutputBenchmark.measure16ElementsWithResizeElement   avgt   40  131,118 ±  2,453  ns/op
ValueListOutputBenchmark.measureSingleElement                 avgt   40   12,097 ±  0,482  ns/op
ValueListOutputBenchmark.measureZeroElement                   avgt   40    0,278 ±  0,005  ns/op

4.3

Benchmark                                                     Mode  Cnt    Score    Error  Units
ValueListOutputBenchmark.measure100Elements                   avgt   40  641,254 ± 58,630  ns/op
ValueListOutputBenchmark.measure100ElementsWithResizeElement  avgt   40  661,805 ± 51,144  ns/op
ValueListOutputBenchmark.measure16Elements                    avgt   40  144,651 ± 24,637  ns/op
ValueListOutputBenchmark.measure16ElementsWithResizeElement   avgt   40  144,673 ± 36,781  ns/op
ValueListOutputBenchmark.measureSingleElement                 avgt   40   13,783 ±  2,029  ns/op
ValueListOutputBenchmark.measureZeroElement                   avgt   40    6,893 ±  1,297  ns/op

[mp911de]

Done.

[CodingFabian]

Quite a nice speedup :-)