The following example route, does an http call to ...

# ktor

[Jamie Craane]

The following example route, does an http call to one of its own routes. This http call goes through the full network stack. What are the options
for bypassing the http stack when executing requests against itself, like testApplication: https://ktor.io/docs/server-testing.html#add-dependencies

fun Application.configureRouting() {
val client = HttpClient {
defaultRequest {
host = “0.0.0.0”
port = 8080
}
engine {
proxy = ProxyBuilder.http(“http://127.0.0.1:8888”)
}
}

routing {
get(“/”) {
val t = client.get(“https://www.google.com”).bodyAsText()
call.respondText(t)
}

get(“/sayHello/{name}“) {
call.respondText(“Hello ${call.parameters[“name”]}“)
}

get(“/helloService/{name}“) {
// This httpcall calls the above route (a call within the same process)
val response = client.get(“/sayHello/${call.parameters[“name”]}“).bodyAsText()
call.respondText(response)
}
}
}

[Aleksei Tirman [JB]]

You can use the external services API to mock the external endpoints. However, you need to reorganize your code to make the client instance injectable into the configureRouting method to pass the test client within the testApplication.

[Jamie Craane]

Thanks. I looked at the external service api but in this case I do not want to mock the endpoint but actually call the internal endpoint in mu own application using http (and not a direct dependency on a class/interface which is of course possible because it is the same application).

The http call works but does a network roundtrip (on localhost), when inspecting the request with for example Charles proxy.

I wonder if it is possible to execute this http request which is internal to my application and bypass the network stack (just like testApplication does).

[Aleksei Tirman [JB]]

Then, you need to simply use the test client within the route handlers.

[Jamie Craane]

I am going to take a look at that! Any idea how to obtain this testclient within main application code?

[Aleksei Tirman [JB]]

You can pass the instance from the testApplication block down to the method that configures the routing.

[Jamie Craane]

Something like this you mean?

fun main() {
    embeddedServer(CIO, port = 8082, host = "0.0.0.0", module = Application::module)
        .start(wait = true)
}

fun Application.module() {
    testApplication {
        [e-mail](client)
    }
}

I do get an error if I try it like this:
2024-10-07 10:52:08.790 [DefaultDispatcher-worker-9] DEBUG io.ktor.server.Application - Unhandled: GET - /helloService/jamie. Exception class kotlinx.coroutines.JobCancellationException: Parent job is Completed
kotlinx.coroutines.JobCancellationException: Parent job is Completed

[Aleksei Tirman [JB]]

No. First, you need to define a parameter of the HttpClient type for the Application.configureRouting method. Second, when using embeddedServer , you need to call the configureRouting method with the HttpClient instance for the production environment. Within the test, you can pass the test client to the configureRouting method.

[Jamie Craane]

In this case, I don’t want to use this client in a unit test but in the actual application itself (outside of the unit test context). Like calling the http endpoint from within the producten app, so that it bypasses the network stack in the production app itself.

[Aleksei Tirman [JB]]

You can't bypass the network stack using the HTTP client within the production app. What is the purpose of making a request to the server within the server?

[Jamie Craane]

I have a multi-module Gradle project build with Ktor. Each submodule is independent of each other (except some common modules). You can think of it as a domain oriented micro-services approach
within a monolithic application. The application is deployed as a single deployable unit (in a container). Dependencies between modules can exists on the external interface level.
By leveraging the same transport as clients do, HTTP in this case, I get the cleanest separation. With this in place (every module also has its own database schema with no foreign key
constraints between each schema), I have the option to deploy a module in its own container. Effectively breakup u the monilith in multiple deployable units without any code changes.

It does work atm but the ‘local’ http call does has some network overhead. This is a tradoff (compared to calling a Koltin interface directly.), so trade-in some performance for flexibility.
My next option was to research the option to bypass the network stack when modules are deployed within the same container. Of course, when a module is deployed as a separate container, the http call
will be a network call.

Hope this clarifies.