Performance Improvements Suggestion: Avoid `locks` for Singletons and for RootScope

[rafaelsc]

Currently, Jab uses locks to guarantee Singleton is only created one time.

But alternative Lazy<T> is a faster alternative to lock.

Instead of

 private Jab.Performance.Basic.Singleton.Singleton1? _ISingleton1;
 
 Jab.Performance.Basic.Singleton.ISingleton1 IServiceProvider<Jab.Performance.Basic.Singleton.ISingleton1>.GetService()
 {
     if (_ISingleton1 == null)
     lock (this)
     if (_ISingleton1 == null){
         _ISingleton1 = new Jab.Performance.Basic.Singleton.Singleton1();
     }
     return _ISingleton1;
 }

The JAB uses this code

 public ImprovedContainerSingleton()
 {
     _ISingleton1 = new(() => new Jab.Performance.Basic.Singleton.Singleton1());
 }

 private Lazy<Jab.Performance.Basic.Singleton.Singleton1> _ISingleton1;

 Jab.Performance.Basic.Singleton.ISingleton1 IServiceProvider<Jab.Performance.Basic.Singleton.ISingleton1>.GetService() => _ISingleton1.Value;

We can see up to 35% of performance improvements.

Method	NumbersOfCalls	NumbersOfClasses	Mean	Error	StdDev	Ratio	RatioSD
Jab	1	1	3.190 ns	0.3329 ns	0.0182 ns	1.00	0.00
Improved_Jab	1	1	1.968 ns	0.1505 ns	0.0082 ns	0.62	0.00
MEDI	1	1	9.357 ns	0.2965 ns	0.0163 ns	2.93	0.02
Jab	100	3	693.577 ns	30.0415 ns	1.6467 ns	1.00	0.00
Improved_Jab	100	3	548.498 ns	7.3230 ns	0.4014 ns	0.79	0.00
MEDI	100	3	2,909.288 ns	33.8740 ns	1.8567 ns	4.19	0.01

See Details: Jab.Performance.Basic.Singleton.BasicSingletonBenchmark-report-github.md

[pakrym]

The problem with Lazy is that all instances need to be eagerly initialized on container creation, leading to a very high startup cost especially for large container.

Maybe LazyInitializer.EnsureInitialized is the way to go here.

[rafaelsc]

This is the Startup benchmark for the code above. (From PR #168)

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated	Alloc Ratio
Jab_Singleton	9.048 ns	4.151 ns	0.2275 ns	0.0067	-	56 B	1.00
Improved_Jab_Singleton	47.049 ns	6.874 ns	0.3768 ns	0.0488	0.0001	408 B	?
MEDI_Singleton	1,927.726 ns	2,530.384 ns	138.6989 ns	0.8640	0.2155	7232 B	?

You are right, this change will cause a bigger startup timer and more allocation.

[rafaelsc]

Benchmark when using LazyInitializer.EnsureInitialized

Singleton benchmark

Method	NumbersOfCalls	NumbersOfClasses	Mean	Error	StdDev	Ratio	RatioSD
Jab	1	1	3.437 ns	3.4240 ns	0.1877 ns	1.00	0.00
Improved_Jab	1	1	1.833 ns	0.4405 ns	0.0241 ns	0.53	0.04
Improved2_Jab	1	1	2.131 ns	3.3616 ns	0.1843 ns	0.62	0.09
MEDI	1	1	9.669 ns	6.6185 ns	0.3628 ns	2.82	0.18
Jab	100	3	689.601 ns	2.0599 ns	0.1129 ns	1.00	0.00
Improved_Jab	100	3	628.522 ns	1,178.7369 ns	64.6105 ns	0.91	0.09
Improved2_Jab	100	3	590.958 ns	1,019.3358 ns	55.8732 ns	0.86	0.08
MEDI	100	3	2,935.810 ns	1,653.3193 ns	90.6240 ns	4.26	0.13

Startup benchmark

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated	Alloc Ratio
Jab_Singleton	9.702 ns	7.7862 ns	0.4268 ns	0.0067	-	56 B	1.00
Improved_Jab_Singleton	55.588 ns	76.3415 ns	4.1845 ns	0.0488	0.0001	408 B	?
Improved_2_Jab_Singleton	8.722 ns	0.1648 ns	0.0090 ns	0.0067	-	56 B	?
MEDI_Singleton	2,197.545 ns	1,339.7199 ns	73.4346 ns	0.8640	0.2155	7232 B	?

Code

private Jab.Performance.Basic.Singleton.Singleton1? _ISingleton1;

Jab.Performance.Basic.Singleton.ISingleton1 IServiceProvider<Jab.Performance.Basic.Singleton.ISingleton1>.GetService() => LazyInitializer.EnsureInitialized<Jab.Performance.Basic.Singleton.Singleton1>(ref this._ISingleton1);

[rafaelsc]

See: https://github.com/pakrym/jab/pull/168#discussion_r1465940230