Improve the Decorate method performance for cases where we have a huge…

[Mariachi1231]

Improve the Decorate method performance for cases where we have a huge amount of services to decorate. In TryDecorateDescriptors, we have N collection travers, where N - is amount of descriptors found for serviceType. For a huge amount of descriptors and a huge amount of types to decorate, we can have performance degradation at the moment of container initialization. These changes are meant to avoid it by memorization of descriptor position.

[khellang]

Thanks for this @Mariachi1231! 🙏🏻 Just curious, how huge are we talking? There's definitely more performance to be gained for decoration 😅

[Mariachi1231]

Thanks for this @Mariachi1231! 🙏🏻 Just curious, how huge are we talking? There's definitely more performance to be gained for decoration 😅

Hi, that's thanks to your convenient set of extensions of IServiceCollection. Regarding performance, I added an optimistic benchmark, you can find it in my fork. Results on my workstation are next:

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19042.1165 (20H2/October2020Update)
Intel Core i7-7700 CPU 3.60GHz (Kaby Lake), 1 CPU, 8 logical and 4 physical cores
.NET SDK=6.0.100-preview.7.21379.14
  [Host]     : .NET Core 3.1.18 (CoreCLR 4.700.21.35901, CoreFX 4.700.21.36305), X64 RyuJIT
  Job-UFPBGU : .NET Core 3.1.18 (CoreCLR 4.700.21.35901, CoreFX 4.700.21.36305), X64 RyuJIT

InvocationCount=10  LaunchCount=1  UnrollFactor=1  

Method	TotalDescriptors	TotalDecorable	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Allocated
TryDecorate_worst_case	1000	3	52.79 μs	3.912 μs	11.350 μs	49.26 μs	40.15 μs	85.38 μs	1.00	0.00	2 KB
TryDecorateOld_worst_case	1000	3	66.63 μs	2.215 μs	6.101 μs	67.17 μs	56.90 μs	84.36 μs	1.29	0.22	2 KB
TryDecorate_worst_case	1000	30	52.89 μs	0.816 μs	0.940 μs	52.77 μs	51.36 μs	55.24 μs	1.00	0.00	7 KB
TryDecorateOld_worst_case	1000	30	163.36 μs	7.059 μs	19.324 μs	159.19 μs	122.00 μs	219.89 μs	3.29	0.30	7 KB
TryDecorate_worst_case	1000	300	93.65 μs	4.974 μs	14.271 μs	95.18 μs	63.31 μs	133.24 μs	1.00	0.00	59 KB
TryDecorateOld_worst_case	1000	300	971.55 μs	28.577 μs	83.811 μs	971.50 μs	756.50 μs	1,180.46 μs	10.65	1.99	53 KB
TryDecorate_worst_case	1000	800	156.42 μs	12.315 μs	35.924 μs	166.26 μs	111.67 μs	259.39 μs	1.00	0.00	149 KB
TryDecorateOld_worst_case	1000	800	1,696.13 μs	33.876 μs	82.460 μs	1,706.78 μs	1,467.23 μs	1,860.15 μs	10.58	2.36	135 KB
TryDecorate_worst_case	10000	3	321.00 μs	32.240 μs	95.060 μs	269.83 μs	235.15 μs	593.21 μs	1.00	0.00	2 KB
TryDecorateOld_worst_case	10000	3	506.09 μs	29.729 μs	85.775 μs	518.54 μs	388.89 μs	724.00 μs	1.64	0.36	2 KB
TryDecorate_worst_case	10000	30	349.41 μs	31.908 μs	94.082 μs	343.81 μs	241.42 μs	606.89 μs	1.00	0.00	7 KB
TryDecorateOld_worst_case	10000	30	1,458.38 μs	44.367 μs	130.816 μs	1,412.01 μs	1,269.57 μs	1,793.13 μs	4.41	0.98	7 KB
TryDecorate_worst_case	10000	300	361.36 μs	35.543 μs	104.242 μs	296.89 μs	262.54 μs	623.25 μs	1.00	0.00	59 KB
TryDecorateOld_worst_case	10000	300	10,949.34 μs	214.899 μs	201.016 μs	11,012.30 μs	10,642.51 μs	11,400.92 μs	21.28	2.24	53 KB
TryDecorate_worst_case	10000	800	438.69 μs	40.516 μs	119.461 μs	400.35 μs	312.90 μs	749.00 μs	1.00	0.00	149 KB
TryDecorateOld_worst_case	10000	800	27,022.49 μs	448.536 μs	397.615 μs	26,946.15 μs	26,535.08 μs	28,029.69 μs	44.42	3.26	135 KB

In general, with such changes, performance will be better with more decorating services. Let's consider regular CQRS based applications with usages of wrappers, for an application with, for instance, 100 queries. Usually, queries are wrapped in a couple of decorators. For example, we can have a logger decorator, params validation decorator, metrics decorator and this means that we already have 300 executions of IndexOf through the total amount of descriptors in the IServiceCollection. With changes, instead, we will have just a single iteration instead of 300 executions of IndexOf through all descriptors for a single wrapped type.