Description

In profiling the .NET Podcasts application startup time on an Android Pixel 5a, I'm seeing its usage of a named HttpClient using builder.Services.AddHttpClient<ShowsService>(client => {}) as taking up 5% of the startup cost of the app.

Configuration

• Which version of .NET is the code running on? 6.0
• What OS version, and what distro if applicable? Android 12
• What is the architecture (x64, x86, ARM, ARM64)? ARM64
• If relevant, what are the specs of the machine? Google Pixel 5a

Regression?

No

Data

Using the startup profiling script at https://github.com/jonathanpeppers/maui-profiling#faq, I'm getting the following results:

(using a full AOT of the app, so to eliminate any JIT time)

Startup time with AddHttpClient

Average(ms): 838.2
Average(ms): 832.4
Average(ms): 835.2

Startup time with just creating a shared HttpClient without DI

Average(ms): 795.5
Average(ms): 799.8
Average(ms): 797.9

Analysis

dotnet-podcasts.speedscope.zip

1. Looking through the attached speedscope, a large chunk of time is spent creating the named Http client service in AddTransientHelper:

A big chunk of that time is due to calling ActivatorUtilities.CreateFactory, which uses System.Linq.Expressions and IL Ref.Emit.

2. The Logging in Microsoft.Extensions.Http also seems to be adding a decent amount of overhead:

Discussion

I wonder if using Microsoft.Extensions.Http on a mobile application is recommended at all. Or if a mobile app should just be following the guidance we give in our docs:

HttpClient is intended to be instantiated once and re-used throughout the life of an application.

@davidfowl @jonathanpeppers @dotnet/ncl