Ember client doesn't shut down quickly when cancelled

To replicate, set fork := true, in your SBT config and compile the following:

import cats.data.Kleisli
import cats.effect.kernel.Resource
import cats.effect.{ExitCode, IO, IOApp}
import org.http4s.ember.server.EmberServerBuilder
import org.http4s.ember.client.EmberClientBuilder
import org.http4s.implicits.http4sLiteralsSyntax
import com.comcast.ip4s._

object TestEmberClientHang extends IOApp.Simple {
  def run: IO[Unit] = EmberClientBuilder.default[IO].build.use(_.statusFromUri(uri"http://localhost:8765")).void
}

object RunServer extends IOApp.Simple {
  def run: IO[Unit] =
    EmberServerBuilder.default[IO].withPort(port"8765").withHttpApp(Kleisli(_ => IO.never)).build.useForever
}

1. Run runMain RunServer
2. In a different SBT session, run runMain TestEmberClientHang
3. Hit ctrl+c in the second SBT session (with TestEmberClientHang)

Expected: the TestEmberClientHang process terminates immediately, like the version using the blaze client does. Actual: the TestEmberClientHang process continues to run for > 30 seconds

Tested on macOS 12.0.1 (21A559)

[stephenjudkins]

I can confirm that this fixes it for me!

FWIW, the nc version of the test doesn’t work and I had to do brew install netcat on macOS to get both the fixed and unfixed versions working as expected without throwing exceptions.

I’d like to express my gratitude for everyone who worked hard to reproduce/investigate/fix this issue

[stephenjudkins]

I think this might be possible to reproduce in a pure-scala context without any processes or signals, I might take a crack of that if I get a minute