Hey everyone :wave:, I was playing with Ktor to bu...

# ktor

[simon.vergauwen]

Hey everyone 👋,
I was playing with Ktor to build a native server and was running it locally on my macArm64 (have not tried linux yet), but when I call engine.stop(5000, 10_000) on the ApplicationEngine it keeps the port occupied, and I get following exception when I try to re-run the server. I'm running the hello world example from the samples repo. Any ideas on how to fix this? Or is this a bug? I couldn't find any bug report on YouTrack, but happy to add one if needed ☺️

Uncaught Kotlin exception: io.ktor.utils.io.errors.PosixException.AddressAlreadyInUseException: EADDRINUSE (48): Address already in use

[simon.vergauwen]

It seems to depend on the values being passed to stop 🤔

It noticed that after some time the port became available again, and decreasing the duration of the parameters passed to stop seem to increase the speed at which the port becomes available again..

[simon.vergauwen]

It's to me not entirely clear why that is looking at the internal code of CIOApplicationEngine.shutdownServer.

[ribesg]

I've noticed native ktor server not releasing port immediately when being killed too (when run from IDE)

[hfhbd]

Not related to Ktor, but native (posix) port handling, that closes the port after 2 minutes to answer incoming requests

[simon.vergauwen]

Hey @hfhbd, could you point me in any direction explaining this? I'm seeing different behavior depending on the parameters passed to ApplicationEngine#stop.

Btw really awesome work on the NativePostgres engine & Postgres work for SqlDelight 👏 I am playing around with it, and it's a really promising project.

[hfhbd]

Thanks, using Ktor native + postgresql is my usecase too 😄
I noticed the same behavior from Ktor and some other web servers too, after shutting down the TCP server it takes some time to free the port. And I just found this on StackOverflow, which makes sense to me: https://unix.stackexchange.com/questions/68311/why-does-it-take-up-to-several-minutes-to-clean-a-listening-tcp-port-after-a-pro

[simon.vergauwen]

Thanks for sharing 🙏 If you're interested, I am developing my current experiment in the open here https://github.com/nomisRev/ktor-native-server

I am still tipping my toes in Kotlin Native, but I am very interested in helping out making SqlDelight better for native and SqlDelight. So far Arrow, and other OSS projects have taken most of my time though..

[napperley]

Unfortunately Ktor Server isn't available for the linuxArm32Hfp and linuxArm64 targets, which is mainly due to the KotlinX Coroutines and KotlinX AtomicFu libraries not being available (Ktor Server uses these libraries as dependencies). The KotlinX Coroutines team have no intention of porting the library to the Linux ARM targets, even though there is large demand to have the issue resolved ( https://github.com/Kotlin/kotlinx.coroutines/issues/855 ), and the issue is nearly four years old!

[hfhbd]

Yeah, and I don't get the reason because other libraries and the std lib "support" these targets, without testing too.

[napperley]

What I do know is that the Ktor team would be very eager to jump at the chance to port Ktor Server to the Linux ARM targets once the KotlinX Coroutines and KotlinX AtomicFu libraries are made available.

[napperley]

KotlinX Serialization (one of the most sophisticated/advanced KotlinX libraries) is one of the KotlinX libraries that is available for the Linux ARM targets, which makes the resistance from the KotlinX Coroutines team very odd. In my own opinion the KotlinX Coroutines team, "don't have a leg to stand on" when it comes to not porting the library to the Linux ARM targets, especially when the library is already available for the linuxX64 target. One can say that the team is facing a "resistance is futile" scenario.

[simon.vergauwen]

KotlinX AtomicFu

This dependency could be removed from Ktor 🤔
AtomicReference from Kotlin Std is supported for all native targets, or is the performance benefit of AtomicFieldUpdater on the JDK so big and so widely used in Ktor?

[simon.vergauwen]

Perhaps if Atomic FU stabilises (soon-ish) KotlinX Coroutines will support it. I've followed Atomic FU for a long time, and they're basically waiting for K2 / FIR (like almost everything at this point) 🤞

Having KotlinX Coroutines for all platforms seems vital for the MPP eco-system

[Sam]

Just wondering if anyone found a solution for this? I'm getting the same issue running a ktor server on iOS 😞

[Skaldebane]

It's been a couple years... but I found the solution to this, simply by configuring reuseAddress = true in the CIO engine configuration like so:

embeddedServer(factory = CIO, configure = {
    connector { port = 8080 }
    reuseAddress = true
}) {
    // ...
}

I assume this sets the SO_REUSEADDR flag on the socket as the Stack Exchange answer linked by @hfhbd states.

[Sam]

Yeah sorry should have posted that here, we PRd the library to expose this property!

[Skaldebane]

Ahh nice, it's super convenient, thank you! 💜