Cancellable HTTP requests via Scala's Tapir
This sample shows an HTTP server and an HTTP client, that can cleanly cancel requests, both on the client-side and on the server-side. This is needed for safe disposal of resources.
On the client-side, it’s important to close or reuse the connection early, to avoid connection leaks. And on the server-side, the cancellation signal can be received when the client closes its connection, so the server may choose to cancel the processing of the request, since there’s no longer a client waiting for the response.
This sample is using Tapir for describing HTTP endpoints. For the server backend, Tapir is configured to use Netty. And for the client making HTTP requests, it’s using Sttp, powered by the standard async-http-client. And I’m also using Cats-Effect, as the effect system because it rocks, with its interruption model being best in class.
#!/usr/bin/env -S scala shebang
//> using scala "3.3.4"
//> using dep "com.softwaremill.sttp.tapir::tapir-netty-server-cats:1.11.5"
//> using dep "com.softwaremill.sttp.tapir::tapir-sttp-client:1.11.5"
//> using dep "com.softwaremill.sttp.client3::async-http-client-backend-cats:3.9.8"
//> using dep "org.typelevel::cats-effect:3.5.4"
//> using dep "org.slf4j:slf4j-nop:2.0.16"
import cats.effect.*
import cats.effect.std.Dispatcher
import sttp.tapir.*
import sttp.tapir.server.netty.cats.NettyCatsServer
import sttp.client3.*
import sttp.client3.asynchttpclient.cats.AsyncHttpClientCatsBackend
import sttp.tapir.client.sttp.SttpClientInterpreter
import scala.concurrent.duration.*
import scala.concurrent.TimeoutException
object Endpoints:
type Type = Endpoint[Unit, Unit, Unit, String, Any]
val slow: Type = endpoint.get
.in("slow")
.out(stringBody)
val fast: Type = endpoint.get
.in("fast")
.out(stringBody)
def startServer: Resource[IO, Unit] =
val fastEndpoint = Endpoints.fast.serverLogic: _ =>
IO.pure(Right("fast response"))
val slowEndpoint = Endpoints.slow.serverLogic: _ =>
IO.monotonic.flatMap: startedAt =>
val task =
for
_ <- IO.println("[Server] Received ping request!")
_ <- IO.sleep(10.seconds)
yield Right("slow response")
task.onCancel:
for
now <- IO.monotonic
elapsedSecs = (now - startedAt).toNanos / 1_000_000_000.0
_ <- IO.println(
f"[Server] Request cancelled after $elapsedSecs%.2f seconds!"
)
yield ()
for
d <- Dispatcher.parallel[IO]
_ <- Resource
.make:
NettyCatsServer[IO](d)
.port(8080)
.addEndpoints(List(fastEndpoint, slowEndpoint))
.start()
.apply: s =>
for
_ <- IO.println("[Server] Shutting down...")
_ <- s.stop()
_ <- IO.println("[Server] Bye, bye 👋")
yield ()
yield ()
def makeRequest(e: Endpoints.Type, backend: SttpBackend[IO, Any]): IO[Unit] =
val send = SttpClientInterpreter()
.toRequest(e, Some(uri"http://localhost:8080"))
.apply(())
.response(asStringAlways)
.send(backend)
.flatMap: response =>
IO.println(s"[Client] Response: ${response.body}")
.timeout(1.second)
.recoverWith:
case _: TimeoutException =>
IO(System.err.println("[Client] ERROR: Request timed out!"))
case e =>
IO.apply:
System.err.println(s"[Client] ERROR Request failed!")
e.printStackTrace()
for
_ <- IO.println(s"[Client] Sending request to ${e.show}")
_ <- send
yield ()
object Main extends IOApp.Simple:
override def run: IO[Unit] =
val resources =
for
_ <- startServer
clientBackend <- AsyncHttpClientCatsBackend.resource[IO]()
yield clientBackend
resources.use: clientBackend =>
for
_ <- makeRequest(Endpoints.fast, clientBackend)
_ <- makeRequest(Endpoints.slow, clientBackend)
yield ()
To run this sample, you can use Scala CLI, which is the default scala
launcher since Scala 3.5.0. For macOS:
# Installs the latest Scala via Homebrew
brew install scala
# Makes the above script executable
chmod +x ./tapir-client-server.scala
# Runs the above script
./tapir-client-server.scala