adding GetChild benchmark

[Aaronontheweb]

Working on cleaning up the read-side resolve pipeline inside Akka.Cluster.Sharding and Akka.Remote. This seemed like a good place to start.

[to11mtm]

How many layers deep do the children get here? Is there a way we can somehow measure difference in perf based on how deep the path is?

[Aaronontheweb]

Data so far:

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19041.1165 (2004/May2020Update/20H1)
AMD Ryzen 7 1700, 1 CPU, 16 logical and 8 physical cores
.NET SDK=5.0.302
  [Host]     : .NET Core 3.1.17 (CoreCLR 4.700.21.31506, CoreFX 4.700.21.31502), X64 RyuJIT
  DefaultJob : .NET Core 3.1.17 (CoreCLR 4.700.21.31506, CoreFX 4.700.21.31502), X64 RyuJIT

Method	Mean	Error	StdDev	Allocated
ResolveChild	62.07 ns	0.971 ns	0.811 ns	-

[Aaronontheweb]

How many layers deep do the children get here? Is there a way we can somehow measure difference in perf based on how deep the path is?

I can add a method to traverse the graph a few layers deeper, yes

[Aaronontheweb]

Ideally, this is the path we want to optimize read-side in Akka.Remote:

akka.net/src/core/Akka/Actor/ActorRefProvider.cs

Lines 465 to 479 in 426111f

	internal IInternalActorRef ResolveActorRef(IInternalActorRef actorRef, IReadOnlyCollection<string> pathElements)
	{
	if (pathElements.Count == 0)
	{
	_log.Debug("Resolve of empty path sequence fails (per definition)");
	return _deadLetters;
	}
	var child = actorRef.GetChild(pathElements);
	if (child.IsNobody())
	{
	_log.Debug("Resolve of path sequence [/{0}] failed", ActorPath.FormatPathElements(pathElements));
	return new EmptyLocalActorRef(_system.Provider, actorRef.Path / pathElements, _eventStream);
	}
	return (IInternalActorRef)child;
	}

It gets called twice per-read and doesn't appear to be cached at all.

[Aaronontheweb]

Ah, I take that back - looks like we do cache inside RARP here:

/// <summary>
        /// Resolves a deserialized path into an <see cref="IActorRef"/>
        /// </summary>
        /// <param name="path">The path of the actor we are attempting to resolve.</param>
        /// <returns>A local <see cref="IActorRef"/> if it exists, <see cref="ActorRefs.Nobody"/> otherwise.</returns>
        public IActorRef ResolveActorRef(string path)
        {
            if (IgnoreActorRef.IsIgnoreRefPath(path))
                return IgnoreRef;

            // using thread local LRU cache, which will call InternalResolveActorRef
            // if the value is not cached
            if (_actorRefResolveThreadLocalCache == null)
            {
                return InternalResolveActorRef(path); // cache not initialized yet
            }
            return _actorRefResolveThreadLocalCache.Cache.GetOrCompute(path);
        }

[Aaronontheweb]

Actually, funnily enough no - that doesn't get used at all inside the AkkaPduCodec

[Aaronontheweb]

Done, expanded the benchmark to include some more scenarios.

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19041.1165 (2004/May2020Update/20H1)
AMD Ryzen 7 1700, 1 CPU, 16 logical and 8 physical cores
.NET SDK=5.0.302
  [Host]     : .NET Core 3.1.17 (CoreCLR 4.700.21.31506, CoreFX 4.700.21.31502), X64 RyuJIT
  DefaultJob : .NET Core 3.1.17 (CoreCLR 4.700.21.31506, CoreFX 4.700.21.31502), X64 RyuJIT

Method	Mean	Error	StdDev	Gen 0	Allocated
ResolveChild	64.69 ns	0.145 ns	0.129 ns	-	-
Resolve3DeepChildRepointableActorRef	652.15 ns	9.237 ns	8.641 ns	0.0839	352 B
Resolve3DeepChildLocalActorRef	311.74 ns	4.924 ns	4.605 ns	0.0629	264 B