A project aimed to compare different .NET API client libraries.

Use case :

• A user trying to save a payment in a different currency
• An external call to an API giving back rates between two currency
  • For benchmark test purpose, it has been hardcoded with the payload of Free forex API

Simple API endpoint taking payment data and using different library to call a external API.

• No persistence as it's aimed to measure library performance

Minimal API aimed to avoid rate limiting and network latency for Benchmark performances

docker build -t restlibraries-api . docker run -d -p 80:80 --name restlibraries-api restlibraries-api

Project respecting Clean Architecture & using one endpoint / handler doing the same process for each library.

• Take a payment payload

• If currency is different, makes an external API call to get the rates between currency

• Return payment result

• An post payment endpoint per API client library :

  • HttpClient
  • Refit
  • Restsharp

• cd .\RestLibraries\test\BenchMarkTests
• dotnet test --filter "Category=HttpClient" -c Release
• dotnet test --filter "Category=Refit" -c Release
• dotnet test --filter "Category=Restsharp" -c Release

Benchmark aren't running smoothly due to .NetCore HttpClient connection pool management (build with HttpClientFactory under the hood). It's still a bit blurry and I may have made a rookie mistake somewhere. Dotnet runtime github issue

   at System.Net.Sockets.Socket.AwaitableSocketAsyncEventArgs.System.Threading.Tasks.Sources.IValueTaskSource<System.Int32>.GetResult(Int16 token)
   at System.Net.Http.HttpConnection.InitialFillAsync(Boolean async)
   at System.Net.Http.HttpConnection.SendAsyncCore(HttpRequestMessage request, Boolean async, CancellationToken cancellationToken)

I've tried a few things without luck :

• Make sure HttpClient was implemented as a singleton to make sure that it's not disposed
• Playing with MaxConnectionsPerServer settings for HttpClient

    .ConfigurePrimaryHttpMessageHandler(() =>
    {
        return new SocketsHttpHandler()
        {
            MaxConnectionsPerServer = 10
        };
    })

1. HttpClient

BenchmarkDotNet=v0.13.5, OS=Windows 10 (10.0.19045.2728/22H2/2022Update)
11th Gen Intel Core i7-11800H 2.30GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK=6.0.405
  [Host]   : .NET 6.0.13 (6.0.1322.58009), X64 RyuJIT AVX2
  ShortRun : .NET 6.0.13 (6.0.1322.58009), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3  

Updated (With System.Net.Http.Json deserialisation)

2. Refit

BenchmarkDotNet=v0.13.5, OS=Windows 10 (10.0.19045.2728/22H2/2022Update)
11th Gen Intel Core i7-11800H 2.30GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK=6.0.405
  [Host]   : .NET 6.0.13 (6.0.1322.58009), X64 RyuJIT AVX2
  ShortRun : .NET 6.0.13 (6.0.1322.58009), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3  

3. Restsharp

Only library to throw socket exception with Job.ShortRun

BenchmarkDotNet=v0.13.5, OS=Windows 10 (10.0.19045.2728/22H2/2022Update)
11th Gen Intel Core i7-11800H 2.30GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK=6.0.405
  [Host]   : .NET 6.0.13 (6.0.1322.58009), X64 RyuJIT AVX2
  ShortRun : .NET 6.0.13 (6.0.1322.58009), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3  

With singleton

|              Method |     Mean |    Error |   StdDev |   Gen0 |   Gen1 | Allocated |
|-------------------- |---------:|---------:|---------:|-------:|-------:|----------:|
| BenchMark_Restsharp | 116.4 μs | 352.7 μs | 19.33 μs | 2.4414 | 0.9766 |  29.62 KB |

As all software engineering question, best answer would be it depends.

1. HttpClient
   • It will induce a bit more boiler plate due to HttpRequest exception handling
   • Best performance usage for API endpoint needing an high throughput
     • I'd consider this if DB access isn't a bottleneck & endpoint are called a lot
   • Easy to make mistakes with disposable objects (stream, ...)
2. Refit
   • Library with the least amount of boilerplate
   • Performance isn't that far for HttpClient (need to confirm with huge json payload)
   • Library hasn't been updated since 08/02/2022
3. Restsharp
   • Less boilerplate than HttpClient but more than Refit
   • Benchmark looks to have too much error to be reliable except memory allocation
   • Active development

Further reading :

• https://github.com/joseftw/jos.httpclient

• https://www.stevejgordon.co.uk/httpclient-connection-pooling-in-dotnet-core

• https://www.aspnetmonsters.com/2016/08/2016-08-27-httpclientwrong/

• https://www.stevejgordon.co.uk/sending-and-receiving-json-using-httpclient-with-system-net-http-json