The RIGHT Way To Use HttpClient In .NET

Get the source code for this video for FREE → the-dotnet-weekly.ck.page/htt...
☄️ Master the Modular Monolith Architecture: bit.ly/3SXlzSt
📌 Accelerate your Clean Architecture skills: bit.ly/3PupkOJ
🚀 Support me on Patreon to access the source code: / milanjovanovic
The easy way to make HTTP requests in .NET is to use the HttpClient to send those requests. And it's a great abstraction to work with, especially with the methods supporting JSON payloads and responses. Unfortunately, it's easy to misuse the HttpClient. Port exhaustion and DNS behavior are some of the most common problems. So in this video, I'll show what you need to know about working with HttpClient correctly.
The Right Way To Use HttpClient In .NET
www.milanjovanovic.tech/blog/...
Join my weekly .NET newsletter:
www.milanjovanovic.tech
Read my Blog here:
www.milanjovanovic.tech/blog
Subscribe for more:
/ @milanjovanovictech
Chapters
0:00 The naive way to use HttpClient
1:10 Why is this a problem?
2:05 Using the IHttpClientFactory
3:34 Adding configurability with named clients
5:54 Strong typing with typed clients
9:17 Why you should avoid typed clients in Singleton services

[@MilanJovanovicTech]

Get the source code for this video for FREE → the-dotnet-weekly.ck.page/httpclient

[@balakrishnan-gf9rw]

Hi Milan, I have a correlation Id in my web api. I used typed http client in mvc core 6 application. So in that case how to pass x-correlation Id in all my web API from client. Kindly suggest me..

[@davearkley7014]

I've been using named clients, but like the idea of strongly typed clients so may try them out. One advantage of being able to apply configuration once is that you can add Polly retry policies and they will be used for all instances of the named/typed client.

[@zfold4702]

Where is your post?

[@metaltyphoon]

Strongly typed and singletons have problems, so watch out.

[@davearkley7014]

@@zfold4702 ?

[@MilanJovanovicTech]

Named clients are awesome

[@hamitkarahan2973]

Last part was very nice Milan. Many thanks. I preferred named clients many times in projects. I didn't prefer typed clients because I was a little afraid to delve into this lifetime issues when my services are singleton. I am just basically calling CreateClient method in my infra layer with a using() block. This approach didn't give me any problem so far in production.

[@PippiTheLongSock]

I think you shouldn't manually dispose the instance. I am pretty sure the framework does that automatically.

[@MilanJovanovicTech]

I agree, if it works don't fix it. Right?

[@hamitkarahan2973]

@@MilanJovanovicTech Yesss 😁

[@csegyud]

It was a nicely built overview. Thank you!

[@MilanJovanovicTech]

Glad you liked it!

[@ArnonDanon]

Great content as always Milan...thank you🤴🏻

[@MilanJovanovicTech]

You're welcome!

[@niezebenmansour3736]

Thank you for the effort in this video!

[@MilanJovanovicTech]

You're welcome :)

[@vibes992000]

Must awaited video. Thanks.

[@MilanJovanovicTech]

Hope you enjoyed it!

[@kodindoyannick5328]

Very good content! Thanks Milan.

[@MilanJovanovicTech]

You're very welcome!

[@JamesSecretofski]

Thanks for the infiniteTimeSpan. Cool tip 👍

[@MilanJovanovicTech]

Don't mention it 😁

[@astroGTC]

Great !! keep the good work !

[@MilanJovanovicTech]

Thanks, will do!

[@mylesdavies9476]

Nice, have been working with typed clients for a while and find it to be a good approach. The only thing I do differently is inject configuration into the typed client and assign any required headers with the constructor, keeping the configure services nice and clean.

[@MilanJovanovicTech]

I think it's better to do the common things inside of AddHttpClient

[@mylesdavies9476]

@@MilanJovanovicTech fair enough, any particular reason?

[@JUTO_MTL]

I do exactly the same because i didnt figure out how to retrieve configuration with Option Pattern. I get null from the IConfiguration inside the HttpClient Configuration method. If someone have any idea

[@toragodzen]

Hi, great video! I've just subscribed and I want to ask about the difference between the EShop and Gatherly projects, which one is better to learn from?

[@MilanJovanovicTech]

They're slightly different, cant' say one is better than the other

[@Javilingys]

What about using Transient HttpClient service inside Scoped Services? Its ok?

[@MilanJovanovicTech]

No scoped services were used here

[@diogosilva3637]

Great explantion! Will you make a tutorial about Refit? Thanks!

[@MilanJovanovicTech]

Perhaps, I like that library. Maybe to compare it with HttpClient

[@d3vil5oldier]

That's exactly what i was looking for, thanks Milan. Just one question, where do you think those typed clients belong to - Application layer or whatever?

[@jaomartins1994]

Infra

[@MilanJovanovicTech]

HttpClient could lead you to place it in Infrastructure You could create an interface in Application, however

[@user-sy8ld9be8j]

Great Video Milan! Do you have a video showing how you would unit test this pls? (unit test IHttpClientFactory).

[@MilanJovanovicTech]

I don't yet, sadly. But I would take a look at MockHttp: github.com/richardszalay/mockhttp

[@grm360]

Nice content as always! Could you make a video on how to use Swagger documentation with minimal api?

[@MilanJovanovicTech]

What's to be covered there?

[@Lazzerman42]

Thanks! Very well explained and clear code examples! I use the RestSharp for know - because it has support for oAuth1(I need it for an old smartHome device). Are there any helpers for oAuth1 and HttpClient?

[@MilanJovanovicTech]

I am not sure, to be completely honest

[@wdorsett84]

Great video! Thank you! Just to confirm when using named clients, the concern over transient vs singleton lifestyles isn't a concern? It's just with typed clients?

[@Sanabalis]

Correct. Since you use a factory to create the named client inside the method each time you invoke it, the lifetime of the class instance is not a factor at all. This is true only if the named client is created inside the method. However, should you create (and capture) named clients inside the constructors, you will effectively introduce the same issues as when using the typed client. Capture the factory instead.

[@MilanJovanovicTech]

Yes, since with named clients you create an instance and dispose of it. If you capture the client in a field, you have a similar issue

[@vishalsaini-py4yd]

Great video....can you please tell how I can configure multiple http client having different token mechanism and adding different authorisation header in each separate request.

[@MilanJovanovicTech]

You can use a delegating handler for each client

[@Ram92026z]

Hi @Milan. This was an absolute great video and I am learning a ton from this channel and so, thank you. Hey, how can I use the above method (the 3rd way you describe in the above video) if I am working off of a class library project and I need to use httpClient within it. Basically, all my API calls are in a class library project. Is there an optimal way of using httpClient in it because I am worried about port exhaustion? Thanks

[@MilanJovanovicTech]

If you can create one HttpClient, pass it a SocketsHttpHandler instance at creation with some connection lifetime, and store it in a static field so that it's long-lived - that should suffice

[@ramprahalad5579]

@@MilanJovanovicTech , got it! It is that oldish pattern of using the HttpClient I guess. Thanks a lot!

[@VascoFernandesBuenoJunior]

O think you need to make another video explaining how to add retry using Polly. I recomend to use Backoff.DecorrelatedJitterBackoffV2 in this case.

[@MilanJovanovicTech]

Will do, great suggestion

[@floralb5317]

Hello Milan, Thank you for amazing content. Something is bothering me regarding the configurations you need for a HttpClient. Let's say that you have some configuration stored in Azure Key Vault/Azure SQL Database, and each incoming request to our API has some kind of x-clientId header, how would you handle creating named HttpClient or strong typing HttpClient after retrieving some metadata on Azure Key Vault/Azure SQL? Thank you, keep up the good work :)

[@MilanJovanovicTech]

I would explore an option of using a DelegatingHandler, to access Key Vault or SQL and get the configuration value.

[@DrHeinzDoofenshmirtz]

Is the last solution for types clients with the lifetime settings a valid and good approach to use? or is it more on the hacky side? I have never stumpled upon it, and I knew almost everything else in the video. Though it was much more well explained than I have read.

[@MilanJovanovicTech]

MSFT recommendation: learn.microsoft.com/en-us/dotnet/fundamentals/networking/http/httpclient-guidelines

[@haskell3702]

Awesome explanation , but I have a doubt how can I Mock the http call using the typed approach? I mean, I want to write a test for GetByUserNameAsync method where the internal httpClient simulates returning 200 and a predefined json

[@MilanJovanovicTech]

Check this out: github.com/skwasjer/MockHttp

[@johntotetwoo]

great video but I am wondering how you can configure with token that expires?

[@MilanJovanovicTech]

DelegatingHandler

[@ershadnozari]

You can inject a httpclientfactory instead of httpclient in a typed client to get around the singleton issue.

[@MilanJovanovicTech]

What's the point of using a typed client then?

[@elfatel]

Thanks

[@MilanJovanovicTech]

You're welcome!

[@AmirHashemZadeh]

well done

[@MilanJovanovicTech]

Thanks!

[@pmsivakumar]

How do you do when you need to fetch the access token dynamically and refresh when it expires with this approach?

[@MilanJovanovicTech]

Check this out: kzbin.info/www/bejne/laaZp2d6a9aaeqs You need a DelegatingHandler on top of the HttpClient

[@WDGKuurama]

What i don't understand i why does it hate to make a new instance of the class injecting the httpclient if we could have a singleton do avoid being garbage collected all the time, is there issues with such techniques?

[@MilanJovanovicTech]

A new instance of which class?

[@ALOKSHARMAMD]

you said using httpclient in traditional way can exhaust ports in long run, but is this true if we are doing proper dispose by applying using statements?

[@MilanJovanovicTech]

Yes - you'd end up creating too many handler instances. You want it to be long-lived

[@KhinSandarMyint-eq2ks]

May I know how to add dynamic certificate using HttpClientFactory ? I have to use different certs.

[@MilanJovanovicTech]

learn.microsoft.com/en-us/dynamics365/business-central/dev-itpro/developer/methods-auto/httpclient/httpclient-addcertificate-method

[@coder3123]

If I were to call API2 from my API1 and the API2 accepts the bearer token that API1 call has is there a way to pass that as header using this method?

[@MilanJovanovicTech]

You can either create a request object, and set the header. Or set the DefaultRequestHeaders on the HttpClient

[@VijayKumar-wy9ku]

If I'm using 10 different named clients, then 10 object of HttpClient is created using IHttpClientFactory. How can we configure a single instance of HttpClient throughout whole application?

[@MilanJovanovicTech]

Well if you want to 10 _DIFFERENT_ named clients, you have no other choice

[@pitthp]

What is good configuration for PooledConnectionLifeTime for Integration test project, that have a hundred test with API calls?

[@MilanJovanovicTech]

How long does it take for your test suite execute?

[@pigrebanto]

ok but what if the token expires and you need to renew it / pass-in it again? to me it seems that with Factory approach you configure it statically at application startup phase.

[@MilanJovanovicTech]

Use a delegating handler

[@codingbloke]

Typed clients has too many gotcha's. In addition to the messy stuff you added at the end, what if you have multiple services that require the same HttpClient configuration? I just used named configurations and only inject the factory, simples.

[@MilanJovanovicTech]

You'd configure the services separately

[@codingbloke]

@@MilanJovanovicTech Indeed but then what you would have is all that malarkey to provide a specific configuration of a HttpClient which is only ever used by a single type and that type's only job is to wrap a specific configuration of HttpClient. In that case again you might as well have that service type do all that work. It would take the factory and what other config services it needed, and have a private method that returns a configured HttpClient

[@palpie]

You should show the strongly typed version using an interface instead.

[@MilanJovanovicTech]

You don't need interfaces for everything 🤷‍♂️

[@semen083]

It is need to add "using" for httpClient, received from IhttpClientFactory?

[@MilanJovanovicTech]

No need, it'll be garbage collected

[@bahmansani3408]

why can not find video about creating read models eventualy with eventual consistency?

[@MilanJovanovicTech]

It's a recent video, look for Materialized views

[@lucid111lucid]

My use case is a DLL which will be used with a winforms program. I expect that my DLL will be called periodically (hourly) and occasionally for a short burst of http activity. What video do you have that might help me choose the best implementation method? /Rob

[@MilanJovanovicTech]

"short burst of http activity" - it doesn't look like you can do much harm whichever implementation you opt for. You could create a static HttpClient in the DLL and reuse it.

[@nanvlad]

What if I just call AddHttpClient() and inject HttpClient into services contstructors? Do they share the same instance or every service gets its own copy?

[@TheMichaelos]

I wanted to ask the same question! @MilanJovanovicTech I would really appreciate if you could elaborate on differences between factory injection and direct client injection 🙏

[@MilanJovanovicTech]

In theory, it shouldn't work - because no HttpClient would be wired up for that service

[@juke-duke]

What if you make the typed client a static class with a static HttpClient?

[@MilanJovanovicTech]

That works cool, since clients should be long-lived (at least the underlying message handler)

[@adriancamposteixeira5483]

why not the "using" or "dispose"?

[@MilanJovanovicTech]

Not needed

[@user-tk2jy8xr8b]

It's a pity that AddHttpClient forces you to use service locator pattern

[@MilanJovanovicTech]

You win some, you lose some

[@zfold4702]

Now extend this for clients that need to send Jwt Token in headers with refresh token capability

[@MilanJovanovicTech]

Will do, that requires Delegating Handlers

[@thecodeman_]

Admit that you saw my post a few minutes ago, and in 20 minutes you recorded a 12-minute video, edited and uploaded it. Just admit it, I'll forgive you..

[@zfold4702]

Where is your post?

[@MilanJovanovicTech]

Yes!

[@Cesar-qi2jb]

If it was true, nothing wrong with this.

[@PauloMorgado]

You forgot to dispose the clients.

[@MilanJovanovicTech]

You would only need to do it in the first example