Finagle Integration

You can generate Finagle binding code by passing the –finagle option to scrooge. For each Thrift service, Scrooge will generate classes that build Finagle Services for both client and server.

The following examples are for Scala generated code and use the following Thrift IDL:

service BinaryService {
  binary fetchBlob(1: i64 id)
}

For both clients and servers you have three API choices to pick from:

  • MethodPerEndpoint

  • ServicePerEndpoint

  • ReqRepServicePerEndpoint

MethodPerEndpoint represents each Thrift service endpoint, or function, as a method of the generated service class. The signature of each method will closely match the corresponding function as defined in the Thrift IDL. One difference though is that the return type of each method will be modified to return a Future of the expected value, instead of the value directly. As an example, the fetchBlob function of the BinaryService would be a method which takes a Long named ‘id’ as an argument, and returns a Future of a Java NIO ByteBuffer, with ByteBuffer being used to represent a binary value.

ServicePerEndpoint represents each Thrift service endpoint, or function, as a Finagle Service, instead of a method. This allows methods to be filtered, either individually or collectively, as a Filter can be composed with a Service but not a method. At its base, a Finagle Service is defined as a function from a RequestType to a Future of a ResponseType. Therefore, a call to a Service requires a single argument. To handle the discrepancy with Thrift methods, where endpoints can require multiple arguments, an Args class is generated, with each value being a named endpoint parameter. For the return type of the Service, a type alias is used named SuccessType, which represents the return type of the Thrift service endpoint. As an example, the fetchBlob function of the BinaryService would be a Service[FetchBlob.Args, FetchBlob.SuccessType]. The FetchBlob.Args class would contain an ‘id’ value of type Long. Whereas FetchBlob.SuccessType would be a type alias for Java NIO’s ByteBuffer.

ReqRepServicePerEndpoint is similar to ServicePerEndpoint, except that each endpoint is represented by a Service from a |c.t.scrooge.Request|_ to a |c.t.scrooge.Response|_. These envelope types allow for the passing of header information between clients and servers, similar to the way HTTP request and response headers work. Note that exchanging header information only works when using the com.twitter.finagle.ThriftMux protocol.

Terminology

If you are still using the deprecated API constructs, we recommend switching to the non-deprecated code as it attempts to add clarity to the API.

Below is a guide which may be helpful for translating from the deprecated API to the non-deprecated.

Deprecated

Replacement

FutureIface

MethodPerEndpoint

MethodIface

MethodPerEndpoint.apply()

MethodIfaceBuilder

MethodPerEndpointBuilder

BaseServiceIface

ServicePerEndpoint

ServiceIface

ServicePerEndpoint.apply()

ServiceIfaceBuilder

ServicePerEndpointBuilder

ReqRepServicePerEndpoint

ReqRepServicePerEndpoint.apply()

ReqRepServicePerEndpointBuilder

Creating a Server

MethodPerEndpoint

To create a server, you need to provide an implementation of the MethodPerEndpoint interface and use it with Finagle’s com.twitter.finagle.Thrift.Server or com.twitter.finagle.ThriftMux.Server class.

class ServerImpl extends BinaryService.MethodPerEndpoint {
  def fetchBlob(id: Long): Future[ByteBuffer] = {
    ??? // your implementation here
  }
}
val server = Thrift.server.serveIface("host:port", new ServerImpl)

ServicePerEndpoint

Additionally, Scrooge generates a ServicePerEndpoint which is a trait defining a Service for each Thrift method.

trait ServicePerEndpoint {
    def fetchBlob: Service[BinaryService.FetchBlob.Args, BinaryService.FetchBlob.SuccessType])

Note that every method in the IDL becomes a Service for the corresponding Args and SuccessType structures. The wrappers are needed to wrap multiple method arguments into one type. Instead of implementing the MethodPerEndpoint interface directly, you can provide an instance of the ServicePerEndpoint and convert it to the MethodPerEndpoint interface using the toThriftService function.

val servicePerEndpoint: BinaryService.ServicePerEndpoint = new BinaryService.ServicePerEndpoint {
  def fetchBlob: Service[BinaryService.FetchBlob.Args, BinaryService.FetchBlob.SuccessType] = ???
}

val methodPerEndpoint: BinaryService.MethodPerEndpoint = servicePerEndpoint.toThriftService
val service = Thrift.server.serveIface("host:port", methodPerEndpoint)

The advantage of this approach is that the Services can be decorated with Filters.

val servicePerEndpoint: BinaryService.ServicePerEndpoint = new BinaryService.ServicePerEndpoint {
  def fetchBlob: Service[BinaryService.FetchBlob.Args, BinaryService.FetchBlob.SuccessType] = ???
}

val loggingFilter: Filter[BinaryService.FetchBlob.Args, BinaryService.FetchBlob.SuccessType] = ???
val filteredServicePerEndpoint: BinaryService.ServicePerEndpoint =
  servicePerEndpoint
    .withFetchBlob(
      fetchBlob = loggingFilter.andThen(servicePerEndpoint.fetchBlob)
    )

val methodPerEndpoint: BinaryService.MethodPerEndpoint = filteredServicePerEndpoint.toThriftService
val service = Thrift.server.serveIface("host:port", methodPerEndpoint)

ReqRepServicePerEndpoint

Note

Only the ThriftMux protocol supports passing header information.

Lastly, Scrooge generates a ReqRepServicePerEndpoint which is also a trait defining a Service for each Thrift method.

trait ReqRepServicePerEndpoint {
    def fetchBlob: Service[Request[BinaryService.FetchBlob.Args], Response[BinaryService.FetchBlob.SuccessType]])

Note that every method in the IDL becomes a Service over c.t.scrooge.Request[Args] and c.t.scrooge.Response[SuccessType] structures. These wrappers allow for the getting and setting of header data that can be exchanged between a client and server. Again, instead of implementing the MethodPerEndpoint interface directly, you can also provide an instance of the ReqRepServicePerEndpoint and convert it to the MethodPerEndpoint interface using the toThriftService function.

val reqRepServicePerEndpoint: BinaryService.ReqRepServicePerEndpoint = new BinaryService.ReqRepServicePerEndpoint {
  def fetchBlob: Service[Request[BinaryService.FetchBlob.Args], Response[BinaryService.FetchBlob.SuccessType]] = ???
}

val methodPerEndpoint: BinaryService.MethodPerEndpoint = reqRepServicePerEndpoint.toThriftService
val service = ThriftMux.server.serveIface("host:port", methodPerEndpoint)

As true with the previous ServicePerEndpoint approach, this can also be decorated with Filters.

val reqRepServicePerEndpoint: BinaryService.ReqRepServicePerEndpoint = new BinaryService.ReqRepServicePerEndpoint {
  def fetchBlob: Service[Request[BinaryService.FetchBlob.Args], Response[BinaryService.FetchBlob.SuccessType]] = ???
}

val loggingFilter: Filter[Request[BinaryService.FetchBlob.Args], Response[BinaryService.FetchBlob.SuccessType]] = ???
val filteredReqRepServicePerEndpoint: BinaryService.ReqRepServicePerEndpoint =
  reqRepServicePerEndpoint
    .withFetchBlob(
      fetchBlob = loggingFilter.andThen(reqRepServicePerEndpoint.fetchBlob)
    )

val methodPerEndpoint: BinaryService.MethodPerEndpoint = filteredReqRepServicePerEndpoint.toThriftService
val service = ThriftMux.server.serveIface("host:port", methodPerEndpoint)

Important

It is expected that a MethodPerEndpoint is what is eventually served to create a ListeningServer, e.g.,

Protocol.server.serveIface(…, methodPerEndpoint)

Thus, even when implementing a ServicePerEndpoint or a ReqRepServicePerEndpoint, the implementation should be converted to a MethodPerEndpoint (via #toThriftService) then served.

Creating a Client

MethodPerEndpoint

Creating a client works similarly; you provide Finagle’s Thrift.Client or ThriftMux.Client object the MethodPerEndpoint.

val methodPerEndpoint: BinaryService.MethodPerEndpoint =
  Thrift.client.build[BinaryService.MethodPerEndpoint]("host:port")

val result: Future[ByteBuffer] = methodPerEndpoint.fetchBlob(12525L)

ServicePerEndpoint

Alternatively, you can request a ServicePerEndpoint instead.

val servicePerEndpoint: BinaryService.ServicePerEndpoint =
  Thrift.client.servicePerEndpoint[BinaryService.ServicePerEndpoint]("host:port")

val result: Future[ByteBuffer] =
  servicePerEndpoint.fetchBlob(BinaryService.FetchBlob.Args(12525L))

As in the server case, this allows you to decorate your client with Filters and convert to the MethodPerEndpoint.

val timeoutFilter: Filter[BinaryService.FetchBlob.Args, BinaryService.FetchBlob.SuccessType] = ???
val filteredServicePerEndpoint: BinaryService.ServicePerEndpoint =
  servicePerEndpoint
    .withFetchBlob(
      fetchBlob = timeoutFilter.andThen(servicePerEndpoint.fetchBlob)
    )

val methodPerEndpoint: BinaryService.MethodPerEndpoint =
  Thrift.Client.methodPerEndpoint(filteredServicePerEndpoint)

ReqRepServicePerEndpoint

Note

Only the ThriftMux protocol supports passing header information.

Lastly, you can request a ReqRepServicePerEndpoint.

val reqRepServicePerEndpoint: BinaryService.ReqRepServicePerEndpoint =
   ThriftMux.client.servicePerEndpoint[BinaryService.ReqRepServicePerEndpoint]("host:port")

 val result: Future[ByteBuffer] =
   reqRepServicePerEndpoint.fetchBlob(
     Request(BinaryService.FetchBlob.Args(12525L))
       .setHeader("foo", "bar"))

As in the server case, this also allows you to decorate your client with Filters and convert to the MethodPerEndpoint.

val timeoutFilter: Filter[Request[BinaryService.FetchBlob.Args], Response[BinaryService.FetchBlob.SuccessType]] = ???
val filteredReqRepServicePerEndpoint: BinaryService.ReqRepServicePerEndpoint =
  reqRepServicePerEndpoint
    .withFetchBlob(
      fetchBlob = timeoutFilter.andThen(reqRepServicePerEndpoint.fetchBlob)
    )

val methodPerEndpoint: BinaryService.MethodPerEndpoint =
  ThriftMux.Client.methodPerEndpoint(filteredReqRepServicePerEndpoint)

Configuration

In both the server and client cases, you probably want to pass more configuration parameters to the Finagle client (e.g. Thrift.client.withXXX), please check the Finagle documentation for tweaks once you get things to compile.

A common configuration is the Thrift TProtocolFactory which can be set with com.twitter.finagle.Thrift.Server.withProtocolFactory and com.twitter.finagle.Thrift.Client.withProtocolFactory along with the same methods available for ThriftMux.