IHttpClientFactory in ASP.NET Core — why you should never new HttpClient()

new HttpClient() every call exhausts sockets (TIME_WAIT). Holding a single static HttpClient ignores DNS changes. IHttpClientFactory solves both.

What it does

1. Pools HttpMessageHandler instances (default 2-minute lifetime).
2. Rotates handlers periodically so DNS refresh actually kicks in.
3. Wires Polly policies, headers, and base URLs into named/typed clients.

Three usage patterns

1. Named client

builder.Services.AddHttpClient("github", c => {
    c.BaseAddress = new Uri("https://api.github.com");
    c.DefaultRequestHeaders.UserAgent.ParseAdd("MyApp/1.0");
});

// Consume
public class GhController(IHttpClientFactory factory) : ControllerBase
{
    public async Task<IActionResult> Repos()
    {
        var c = factory.CreateClient("github");
        return Ok(await c.GetFromJsonAsync<object>("/repos"));
    }
}

2. Typed client (preferred)

public class GitHubClient(HttpClient http)
{
    public Task<IList<Repo>?> ListAsync() => http.GetFromJsonAsync<IList<Repo>>("/repos");
}

builder.Services.AddHttpClient<GitHubClient>(c => c.BaseAddress = new("https://api.github.com"));

3. With Polly resilience

builder.Services.AddHttpClient<GitHubClient>()
    .AddStandardResilienceHandler();  // .NET 8+, replaces the old AddTransientHttpErrorPolicy

Why static HttpClient is also wrong

A singleton HttpClient reuses the same socket connection forever. If the DNS record changes (failover, scale-out, blue/green), your app keeps talking to the dead IP until the process restarts. IHttpClientFactory rotates handlers so DNS does refresh.

Lifetime cheat sheet

Approach	Sockets exhausted?	DNS refresh?	Use?
new HttpClient() per call	❌ yes	✅ yes	❌ never
Static HttpClient	✅ no	❌ no	❌ rarely
IHttpClientFactory	✅ no	✅ yes	✅ always