This is a collection of short recipes of “How to X in Finch”.

Fixing the .toService compile error

Finch promotes a type-full functional programming style, where an API server is represented as a coproduct of all the possible types it might return. That said, a Finch server is type-checked by the compiler to ensure that it’s known how to convert every part of coproduct into an HTTP response. Simply speaking, as a Finch user, you get a compile-time guarantee that for every endpoint in your application it’s possible to find an appropriate encoder. Otherwise you will get a compile error that looks like this.

<console>:34: error: An Endpoint you're trying to convert into a Finagle service is missing one or more encoders.

  Make sure shapeless.:+:[Foo,shapeless.:+:[Bar,shapeless.CNil]] is one of the following:

  * A com.twitter.finagle.http.Response
  * A value of a type with an io.finch.Encode instance (with the corresponding content-type)
  * A coproduct made up of some combination of the above

       (e :+: q).toServiceAs[Application.Json]

Which means: the compiler wasn’t able to find an instance of Encode.Json type-class for types Foo and Bar. To fix that you could either provide that instance (seriously, don’t do that unless you have an absolutely specific use case) or use one of the supported JSON libraries and get it for free (preferred).

For example, to bring the Circe support and benefit from its auto-derivation of codecs you’d only need to add two extra imports to the scope (file) where you call the .toService method.

import io.finch.circe._

Serving static content

Finch was designed with type-classes powered extensibility in mind, which means it’s possible to define an Endpoint of any type A as long as there is a type-class instance of Encode[A] available for that type. Needless to say, it’s pretty much straightforward to define a blocking instance of Encode[File] that turns a given File into a Buf. Although, it might be tricky to come up with a non-blocking way of serving static content with Finch, there is a way. The cornerstone idea is to return a Buf instance from the endpoint so we could use an identity Encode[Buf], thereby lifting the encoding part onto the endpoint itself (where it’s quite legal to return a Future[Buf]).

import io.finch._
import io.finch.syntax._
import{Reader, Buf}
import com.twitter.finagle.Http
import com.twitter.util.Await

val reader: Reader = Reader.fromFile(new File("/dev/urandom"))

val file: Endpoint[Buf] = get("file") {
val service = file.toServiceAs[Text.Plain]

// Await.ready(Http.server.serve(":8081", service))

Note: It’s usually not a great idea to use tools like Finch (or similar) to serve static content given their dynamic nature. Instead, a static HTTP server (i.e., Nginx) would be the perfect fit.

It’s also possible to stream the file content to the client using AsyncStream.

import io.finch._
import io.finch.syntax._
import com.twitter.concurrent.AsyncStream
import com.twitter.finagle.Http
import{Reader, Buf}


val reader: Reader = Reader.fromFile(new File("/dev/urandom"))

val file: Endpoint[AsyncStream[Buf]] = get("stream-of-file") {
  Ok(AsyncStream.fromReader(reader, chunkSize = 512.kilobytes.inBytes.toInt))

// Http.server
//   .withStreaming(enabled = true)
//   .serve(":8081", file.toServiceAs[Text.Plain])

Converting Errors into JSON

Finch’s own errors are often accumulated in the product Endpoint and represented as io.finch.Errors that wraps a[Error]. Writing an exception handling function for both Error (single error) and Errors (multiple errors) cases may not seem as a trivial thing to do.

With Circe the complete implementation might look like the following.

import io.circe.{Encoder, Json}
import io.finch._
import io.finch.circe._

def errorToJson(e: Error): Json = e match {
  case Error.NotPresent(_) =>
    Json.obj("error" -> Json.fromString("something_not_present"))
  case Error.NotParsed(_, _, _) =>
    Json.obj("error" -> Json.fromString("something_not_parsed"))
  case Error.NotValid(_, _) =>
    Json.obj("error" -> Json.fromString("something_not_valid"))

implicit val ee: Encoder[Exception] = Encoder.instance {
  case e: Error => errorToJson(e)
  case Errors(nel) => Json.arr( _*)

Defining endpoints returning empty responses

Just like in any Scala program you can define a function returning an empty result (a unit value), in Finch, you can define an endpoint returning an empty response (an empty/unit output). An Endpoint[Unit] represents an endpoint that doesn’t return any payload in the response.

import io.finch._

val empty: Endpoint[Unit] = get("empty" :: path[String]) { s: String =>
  NoContent[Unit].withHeader("X-String" -> s)

There are also cases when an endpoint returns either a payload or an empty response. While it’s probably a better idea to use failures in order to explain to the remote client why there is no payload in the response, it’s totally possible to send empty ones instead.

import io.finch._
import com.twitter.finagle.http.Status

case class Foo(s: String)

// This is possible
val fooOrEmpty: Endpoint[Foo] = get("foo" :: path[String]) { s: String =>
  if (s != "") Ok(Foo(s))
  else NoContent

// This is recommended
val fooOrFailure: Endpoint[Foo] = get("foo" :: path[String]) { s: String =>
  if (s != "") Ok(Foo(s))
  else BadRequest(new IllegalArgumentException("empty string"))

Defining redirecting endpoints

Redirects are still weird in Finch. Until reversed routes/endpoints are shipped, the reasonable way of defining redirecting endpoints is to represent them as Endpoint[Unit] (empty output) indicating that there is no payload returned.

import io.finch._
import com.twitter.finagle.http.Status

val redirect: Endpoint[Unit] = get("redirect" :: "from") {
  Output.unit(Status.SeeOther).withHeader("Location" -> "/redirect/to")

Defining custom endpoints

“Custom endpoints” isn’t probably a good definition for these since once you called map* or as* on a predefined endpoint it becomes “custom”. Anyways, there are endpoints (at least some part of more complex endpoints) that might be decoupled and shared across other endpoints in your application.

Finch is a library promoting functional programming, which means it prefers composition over inheritance. Thus, building new instances in Finch is never about extending some base class, but about composing existing instances together.

Example 1: aka request reader

Before 0.10, there was a RequestReader abstraction in Finch that has been replaced with evaluating endpoints. Even though the name was changed, the request-reader-flavored API (and behavior) wasn’t touched at all.

In the following example, we define a new endpoint foo that reads an instance of the case class Foo from the request during the evaluation stage. So it won’t affect matching.

import io.finch._

case class Foo(i: Int, s: String)

val foo: Endpoint[Foo] = (param[Int]("i") :: param("s")).as[Foo]

val getFoo: Endpoint[Foo] = get("foo" :: foo) { f: Foo =>
  println(s"Got foo: $f")
  Ok(f) // echo it back

Note: The endpoint body from the example above will never be evaluated if the foo endpoint fails (e.g., one of the params is missing). This shouldn’t be a big surprise given that such behavior is quite natural for a functor (i.e., map function) - an endpoint on which mapOutput is called (via the syntactic sugar around apply) might have already failed.

Example 2: authentication

Since endpoints provide more control over the output (i.e., via io.finch.Output), it’s now possible to define self-contained instances that also handle exceptions (convert them to appropriate outputs).

In this example, we define an evaluating endpoint auth that takes a request and tries to authenticate it by the user name passed in the User header. If the header is missing, the request is considered unauthorized.

import io.finch._

case class User(id: Int)

val auth: Endpoint[User] = header("User").mapOutput(u =>
  if (u == "secret user") Ok(User(10))
  else Unauthorized(new Exception(s"User $u is unknown."))
).handle {
  // if header "User" is missing we respond 401
  case e: Error.NotPresent => Unauthorized(e)

val getCurrentUser: Endpoint[User] = get("user" :: auth) { u: User =>
  println(s"Got user: $u")
  Ok(u) // echo it back

Note: Even though an endpoint auth can’t fail, since we explicitly handled its only possible exception, the body of the getCurrentUser endpoint will only be evaluated if the incoming request contains a header User: secret user and a path /user. This comes from io.finch.Output, which provides a monadic API over the three cases (payload (i.e., Ok), failure (i.e., BadRequest) and empty) and only Output.Payload is considered a success. Simply speaking, calling map* on either Output.Failure or Output.Empty is the same as calling map* on None: Option[Nothing]. Thus, an endpoint returning non-Output.Payload output is considered failed and its map* call won’t be evaluated.

Example 3: asynchronous authentication

Sometimes authenticating a request requires an asynchronous call (e.g., to a database or another HTTP service). Luckily, in addition to the mapOutput method used in the previous example, which takes a function of type (A) => Output[B], Endpoints also have a method called mapOutputAsync that takes a function of type (A) => Future[Output[B]].

The previous example’s auth endpoint can be updated as follows:

import com.twitter.util.Future

def fetchUserForToken(token: String): Future[Option[User]] = ???

val auth: Endpoint[User] = header("User").mapOutputAsync(u =>
  if (u == "secret user") Future.value(Ok(User(10)))
  else fetchUserForToken(u).map {
    case Some(user) => Ok(user)
    case None => Unauthorized(new Exception(s"Invalid token: $u"))
).handle {
  // if header "User" is missing we respond 401
  case e: Error.NotPresent => Unauthorized(e)

The getCurrentUser endpoint doesn’t need to change at all, since auth is still an Endpoint[User].

Example 4: custom path matcher

Let’s say you want to write a custom matching endpoint that only matches requests whose current path segment might be extracted as (converted to) Java 8’s LocalDateTime.

import io.finch._
import com.twitter.util.Try
import java.time.LocalDateTime

implicit val e: DecodePath[LocalDateTime] =
  DecodePath.instance(s => Try(LocalDateTime.parse(s)).toOption)

val dateTime: Endpoint[LocalDateTime] = get("time" :: path[LocalDateTime]) { t: LocalDateTime =>
  println(s"Got time: $t")
  Ok(t) // echo it back

Note: io.finch.DecodePath is an experimental API that will be (or not) eventually promoted to non-experimental.

Example 5: get all parameters or headers

In case if you want to operate with a whole request, you could use root endpoint. But if there is need to get only query parameters or headers, it’s easy to reuse this endpoint to get something like following:

scala> import io.finch._
import io.finch._

scala> val headersAll =
headersAll: io.finch.Endpoint[scala.collection.immutable.Map[String,String]] = root

scala> val headers = get("hello" :: headersAll) {headers: Map[String, String] =>
     |   Ok(s"Headers: $headers")
     | }
headers: io.finch.Endpoint[String] = GET /hello :: root

scala> headers(Input.get("/hello").withHeaders("foo" -> "bar")).awaitValueUnsafe()
res31: Option[String] = Some(Headers: Map(foo -> bar))

CORS in Finch

There is a Finagle filter which, when applied, enriches a given HTTP service with CORS behavior. The following example builds a CORS filter that allows GET and POST requests with an Accept header from any origin.

import com.twitter.finagle.http.filter.Cors
import com.twitter.finagle.http.{Request, Response}
import com.twitter.finagle.Service
import io.finch._

val service: Service[Request, Response] = Endpoint.liftOutput(Ok("Hello, world!")).toServiceAs[Text.Plain]

val policy: Cors.Policy = Cors.Policy(
  allowsOrigin = _ => Some("*"),
  allowsMethods = _ => Some(Seq("GET", "POST")),
  allowsHeaders = _ => Some(Seq("Accept"))

val corsService: Service[Request, Response] = new Cors.HttpFilter(policy).andThen(service)

Converting between Scala futures and Twitter futures

Since Finch is built on top of Finagle, it shares its utilities, including futures. While there is already an official tool for performing conversions between Scala futures and Twitter futures (i.e., Twitter Bijection), it usually makes sense to avoid an extra dependency because of a couple of functions which are fairly easy to implement.

import com.twitter.util.{Future => TFuture, Promise => TPromise, Return, Throw}
import scala.concurrent.{Future => SFuture, Promise => SPromise, ExecutionContext}
import scala.util.{Success, Failure}

implicit class RichTFuture[A](f: TFuture[A]) {
  def asScala(implicit e: ExecutionContext): SFuture[A] = {
    val p: SPromise[A] = SPromise()
    f.respond {
      case Return(value) => p.success(value)
      case Throw(exception) => p.failure(exception)


implicit class RichSFuture[A](f: SFuture[A]) {
  def asTwitter(implicit e: ExecutionContext): TFuture[A] = {
    val p: TPromise[A] = new TPromise[A]
    f.onComplete {
      case Success(value) => p.setValue(value)
      case Failure(exception) => p.setException(exception)


Also note that as of Finch 0.16-M3 there is a Scala Futures syntax support for endpoints.

import io.finch._, io.finch.syntax.scalaFutures._
import scala.concurrent.Future

val e = get("foo") { Future.successful(Ok("bar")) }


Server Sent Events

Finch offers support for Server Sent Events through the finch-sse sub-project. Server Sent Events are represented as AsyncStreams and streamed over the chunked HTTP transport.

The ServerSentEvent case class carries an arbitrary data field and it’s possible to encode any ServerSentEvent[A] for which cats.Show[A] is defined.

In this example, every next second we stream instances of java.util.Date as server sent events on the time endpoint.

NOTE: SSE requires Cache-Control to be disabled.

import cats.Show
import com.twitter.concurrent.AsyncStream
import com.twitter.conversions.time._
import com.twitter.finagle.Http
import com.twitter.finagle.util.DefaultTimer
import com.twitter.util.{ Await, Future, Timer}
import io.finch._
import io.finch.sse._
import java.util.Date

implicit val showDate: Show[Date] = Show.fromToString[Date]

implicit val timest: Timer = DefaultTimer.twitter

def streamTime(): AsyncStream[ServerSentEvent[Date]] =
          .map(_ => new Date())
  ) ++ streamTime()

val time: Endpoint[AsyncStream[ServerSentEvent[Date]]] = get("time") {
    .withHeader("Cache-Control" -> "no-cache")

//Await.ready(Http.server.serve(":8081", time.toServiceAs[Text.EventStream]))


Not going into the details on why JSONP considered insecure, there is a Finagle filter JsonpFilter that can be applied to an HTTP service returning JSON to “upgrade” it to JSONP.

Here is a small example on how to wire this filter with Finch’s endpoint.

import com.twitter.finagle.Http
import com.twitter.finagle.http.filter.JsonpFilter
import io.finch._
import io.finch.circe._

val endpoint: Endpoint[Map[String, String]] = get("jsonp") {
  Ok(Map("foo" -> "bar"))

val service = endpoint.toServiceAs[Application.Json]

// Http.server.serve(":8080", JsonpFilter.andThen(service))

JsonpFilter is dead simple. It checks the returned HTTP payload and if it’s a JSON string, wraps it with a call to a function whose name is passed in the callback query-string param (and changes the content-type to application/javascript correspondingly). Using HTTPie, this would look as:

$ http :8081/jsonp
HTTP/1.1 200 OK
Content-Encoding: gzip
Content-Length: 39
Content-Type: application/json

    "foo": "bar"

$ http :8080/jsonp?callback=myfunction
HTTP/1.1 200 OK
Content-Encoding: gzip
Content-Length: 56
Content-Type: application/javascript



There is finagle-oauth2 server-side provider that is supported in Finch via the finch-oauth2 package.

Basic HTTP Auth

Basic HTTP Auth support could be added via the finagle-http-auth project that provides Finagle filters implementing authentication for clients and servers. In Finch this would look like a BasicAuth.Server filter applied to Service returned from the .toService call. See finagle-http-auth’s README for more usage details.