The problem with HttpClient
When you create lots of HttpClient
instances, you may incur Socket Exhaustion.
This happens because sockets are a finite resource, and they are not released exactly when you 'Dispose' them, but a bit later. So, when you create lots of clients, you may terminate the available sockets.
Even with using
statements you may end up with Socket Exhaustion.
class ResourceChecker
{
public async Task<bool> ResourceExists(string url)
{
using (HttpClient client = new HttpClient())
{
var response = await client.GetAsync(url);
return response.IsSuccessStatusCode;
}
}
}
Actually, the real issue lies in the disposal of HttpMessageHandler
instances. With simple HttpClient
objects, you have no control over them.
Introducing HttpClientFactory
The HttpClientFactory
class creates HttpClient
instances for you.
class ResourceChecker
{
private IHttpClientFactory _httpClientFactory;
public ResourceChecker(IHttpClientFactory httpClientFactory)
{
_httpClientFactory = httpClientFactory;
}
public async Task<bool> ResourceExists(string url)
{
HttpClient client = _httpClientFactory.CreateClient();
var response = await client.GetAsync(url);
return response.IsSuccessStatusCode;
}
}
The purpose of IHttpClientFactory
is to solve that issue with HttpMessageHandler
.
An interesting feature of IHttpClientFactory
is that you can customize it with some general configurations that will be applied to all the HttpClient
instances generated in a certain way. For instance, you can define HTTP Headers, Base URL, and other properties in a single point, and have those properties applied everywhere.
How to add it to .NET Core APIs or Websites
How can you use HttpClientFactory in your .NET projects?
If you have the Startup class, you can simply add an instruction to the ConfigureServices
method:
public void ConfigureServices(IServiceCollection services)
{
services.AddHttpClient();
}
You can find that extension method under the Microsoft.Extensions.DependencyInjection
namespace.
Wrapping up
In this article, we've seen why you should not instantiate HttpClients manually, but instead, you should use IHttpClientFactory
.
Happy coding!
🐧
Top comments (6)
Since HttpClient instances are thread-safe and don't hold much in the way of state (except if you're setting up, eg, default headers or base urls), you can also use a singleton pattern with them - this is a performant way to do a lot of concurrent http requests.
On one hand you're right.
On the other hand, using a single instance of HttpClient creates a bottleneck.
Creating them using HttpClientFactory allows you to use on of the clients currently idle in the connections pool.
Incorrect - unless you want to set default headers or connection handlers (docs.microsoft.com/en-us/aspnet/co...), HttpClient is thread-safe (docs.microsoft.com/en-us/dotnet/ap...) and does not block on requests.
In fact, we use this static pattern a lot in a highly-concurrent environment, and have been doing so for years.
TL;DR: if you're willing to specify all requests fully (ie, full url, required headers (if any)) then a static HttpClient is just fine. If you aren't doing tricks with headers, then your only requirement here is to fully qualify request urls, which isn't just a big requirement.
Here's a demo to prove it: github.com/fluffynuts/ConcurrentHt... - if you run that, I see 10 sequential requests to google.com taking around 5 seconds, and 10 parallel requests taking around 1 second.
Ah, I see. Interesting.
Silly question: if it works, why did the .NET team created the HttpClientFactory? Only for customizing single HTTP clients with stuff like BaseUrl and headers?
docs.microsoft.com/en-us/dotnet/ar...
Mostly because most people often use HttpClient incorrectly - disposing and recreating clients, which is expensive. Hiding this behind a factory means you don't see that you're getting the same instance, and you have an injectable service to provide this instance. Also, you can create clients with common config (headers, baseurl, etc) and just rely on getting the "correct one" when you ask for it by name or as an injectable constructor parameter for a service of your own. The factory should ensure better life-cycles for your consumed HttpClient instances because they can (as demonstrated) be re-used and used in parallel, without issue.
I must be clear that there's nothing wrong with using IHttpClientFactory - just that it's not the only way to work with HttpClient instances performantly.