Horace Williams

Using Cats Effect Resource in Non-Functional Contexts

Summary / TL;DR

Use Resource.allocated to integrate cats.effect.Resource into non-FP contexts such as a test framework setup/teardown methods or traditional callback-driven APIs. But be cautious in doing so, as it’s now your responsibility to make sure the provided finalizer gets called at the appropriate time.

Background

Cats Effect Resource is an excellent tool for managing resource lifecycles in Scala. It provides something similar to Python’s ‘with’ helper or Java’s Try With Resources but in a functional API that plays nice with cats.effect.IO or other pure effect types in Scala.

There are several ways to define a Resource but my preference is using Resource.make, which follows a familiar “acquire” / “release” pattern:

import cats.effect.IO
import java.io.FileWriter

// Resource's signature is Resource[+F[_], +A], containing 2 type parameters:
// an effect F (e.g. cats.effect.IO)
// and the resource type A
val myFile: Resource[IO, FileWriter] = Resource.make {
  // acquire the resource by providing an IO[A] which generates it
  IO(new FileWriter("pizza.txt"))
} { file: FileWriter =>
  // release the resource by providing an IO[Unit] which closes it
  // Note that this block receives the resource as a parameter
  IO(file.close())
}

This “double block” syntax can be unfamiliar at first, but it’s simply exploiting Scala’s multiple parameter lists to accept the acquire and release args in sequence without wrapping parentheses.

Once you’ve constructed a resource (or rather, defined the logic for constructing it, as the actual construction is deferred), you can access it via the use method:

val myFile: Resource[IO, FileWriter] = ???
val result: IO[Unit] = myFile.use { file =>
  IO(file.write("pepperoni"))
}

Integrating Resource into an App

Note that the block provided to use also expects an IO (its signature is (f: (A) => G[B])) and yields an IO[B]. Resource’s machinery folds the effectful acquire/release operations into the rest of your program so the whole thing distills into a single IO[Blah].

This works especially smoothly in contexts where you can put the resource-management on the “outside” of your program, such as in IOApp. It’s very common to see IOApp used along this pattern:

import cats.effect._

object Main extends IOApp {
  def context: Resource[IO, MyApp] = ???
  def run(args: List[String]): IO[ExitCode] = {
    context.use { app =>
      val res: IO[Unit] = app.doStuff
      res.map(_ => ExitCode.Success)
    }
  }
}

Resource in less friendly contexts

However, sometimes we don’t have total control over the layering of a program, and instead have to fit a functional core, which may be heavily IO-based, into an imperative shell, which is not. This comes up commonly in scenarious such as:

  • Test frameworks, which often use an imperative, stateful API for defining fixtures via a setup / teardown protocol.
  • Traditional hook-driven APIs, such as web frameworks that provide their own lifecycle methods. For example, I ran into this recently trying to integrate cats.effect.Resource into the ApplicationLifecycle of a Play application.

Luckily, Resource does provide an escape hatch for these situations: Resource.allocated.

Resource.allocated basically provides an unsafe method for “unnesting” the typical resource management flow. Instead of use-ing your resource by providing an inner IO, you retrieve the value of it immediately, along with a callback which you are responsible for invoking to trigger the necessary release steps:

val appResource: Resource[IO, MyApp] = ???
// this IO[Unit] is your deferred finalizer
val appLauncher: IO[(MyApp, IO[Unit])] = appResource.allocated
val (app, shutdownHook): (MyApp, IO[Unit]) = impLauncher.unsafeRunSync()

// Patch the provided finalizer into the application's shutdown lifecycle
myLifeCycle.addStopHook(() => shutdownHook.unsafeToFuture())

This is certainly not the ideal way to use cats.effect.Resource, but it’s a useful escape hatch to have when needed. As always, heed the warnings from the docs:

For this reason, this is an advanced and potentially unsafe api which can cause a resource leak if not used correctly, please prefer use as the standard way of running a Resource program.

So if you do resort to manual use of .allocated, just make sure you’re careful about cleaning up your resources at the appropriate time.