Improve HttpClient resilience benchmarks

[martintmk]

Addiing a new benchmark that compares no-resilience, standard resilience and standard hedging resilience.

BenchmarkDotNet=v0.13.2, OS=Windows 11 (10.0.22621.1848), VM=Hyper-V
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK=8.0.100-preview.6.23309.5
  [Host] : .NET 8.0.0 (8.0.23.30704), X64 RyuJIT AVX2

Job=MediumRun  Toolchain=InProcessEmitToolchain  IterationCount=15
LaunchCount=2  WarmupCount=10

|                    Method |       Mean |    Error |   StdDev | Ratio | RatioSD |   Gen0 | Allocated | Alloc Ratio |
|-------------------------- |-----------:|---------:|---------:|------:|--------:|-------:|----------:|------------:|
|             DefaultClient |   236.4 ns |  2.24 ns |  3.28 ns |  1.00 |    0.00 | 0.0191 |     488 B |        1.00 |
| StandardResilienceHandler | 4,101.4 ns | 21.96 ns | 32.19 ns | 17.35 |    0.22 | 0.1602 |    4152 B |        8.51 |
|    StandardHedgingHandler | 8,060.2 ns | 58.09 ns | 85.15 ns | 34.10 |    0.35 | 0.2289 |    5880 B |       12.05 |

Microsoft Reviewers: Open in CodeFlow

[RussKie]

compares no-resilience, standard resilience and standard hedging resilience.

Could you give me a high level overview of these? I'm new to this space, and I'm trying to understand why there's such a significant difference:

Method	Mean	Allocated
DefaultClient	236.4 ns	488 B
StandardResilienceHandler	4,101.4 ns	4152 B
StandardHedgingHandler	8,060.2 ns	5880 B

Thanks

[martintmk]

compares no-resilience, standard resilience and standard hedging resilience.

Could you give me a high level overview of these? I'm new to this space, and I'm trying to understand why there's such a significant difference:

Method Mean Allocated
DefaultClient 236.4 ns 488 B
StandardResilienceHandler 4,101.4 ns 4152 B
StandardHedgingHandler 8,060.2 ns 5880 B
Thanks

Sure!

• DefaultClient: plain HttpClient without any handlers
• StandardResilienceHandler: this handler combines 5 Polly V7 policies to improve the resilience. It's the handler we recommend to all production services that required HTTP communication. It consists of these policies:
  • Overal Timeout
  • Rate Limiting
  • Retry
  • Circuit Breaker
  • Attempt Timeout
• StandardHedgingHandler: hedging handler replaces the retry with hedging policy that is able to improve latency by sending multiple concurrent requests if the pending ones are too slow. Improves p95 latency. In general, used for more advanced scenarios as it requires multiple remote endpoints to work correctly. It consists of these policies:
  • Overal Timeout
  • Rate Limiting
  • Hedging
  • Circuit Breaker
  • Attempt Timeout