Dependency Injection

§Dependency Injection in Lagom

When building services in Lagom, your code will have dependencies on Lagom APIs and other services that need to be satisfied by concrete implementations at runtime. Often, the specific implementations will vary between development, test and production environments, so it’s important not to couple your code tightly to a concrete implementation class. A class can declare constructor parameters with the abstract types of its dependencies — usually represented by traits in Scala — allowing the concrete implementations to be provided when the class is constructed. This pattern is called “dependency injection” and is fundamental to the way Lagom applications are assembled. Read “Dependency Injection in Scala using MacWire” for more background on this pattern and its benefits.

Your service’s dependencies will have their own dependencies on other APIs in turn. Taken all together, these form a dependency graph that must be constructed when your application starts up. This process is called “wiring” your application, and is performed by creating a subclass of LagomApplication that contains initialization code.

It is common to have clusters of interdependent classes that together form a larger logical component. It can be useful to modularize wiring code in a way that reflects these groupings. You can do so in Scala by defining a “component” trait that includes a combination of lazy val declarations that instantiate concrete implementations of interfaces that the component provides. Some of their constructor parameters may be declared as abstract methods, indicating a dependency that must be provided by another component or your application itself. Your application can then extend multiple components to mix them together into a fully assembled dependency graph. If you don’t fulfill all of the declared requirements for the components you include in your service, it will not compile.

In “Dependency Injection in Scala: guide” this method of mixing components together to form an application is referred to as the “thin cake pattern”. The guide also introduces Macwire which we’ll use in the next steps.

§Wiring together a Lagom application

In section Service descriptors and Implementing services we created a Service and its implementation, we now want to bind them and create a Lagom server with them. Lagom’s Scala API is built on top of Play Framework, and uses Play’s compile time dependency injection support to wire together a Lagom application. Play’s compile time dependency injection support is based on the thin cake pattern we just described.

Once you are creating your Application you will use Components provided by Lagom and obtain your dependencies from them (e.g., an Akka Actor System or a CassandraSession for your read-side processor to access the database). Although it’s not strictly necessary, we recommend that you use Macwire to assist in wiring dependencies into your code. Macwire provides some very lightweight macros that locate dependencies for the components you wish to create so that you don’t have to manually wire them together yourself. Macwire can be added to your service by adding the following to your service implementations dependencies:

libraryDependencies += "com.softwaremill.macwire" %% "macros" % "2.2.5" % "provided"

The simplest way to build the Application cake and then wire your code inside it is by creating an abstract class that extends LagomApplication:

import com.lightbend.lagom.scaladsl.server._
import play.api.libs.ws.ahc.AhcWSComponents
import com.softwaremill.macwire._

abstract class HelloApplication(context: LagomApplicationContext)
    extends LagomApplication(context)
    with AhcWSComponents {
  override lazy val lagomServer = serverFor[HelloService](wire[HelloServiceImpl])
}

In this sample, the HelloApplication gets AhcWSComponents mixed in using the cake pattern and implements the lazy val lagomServer using Macwire. The important method here is the lagomServer method. Lagom will use this to discover your service bindings and create a Play router for handling your service calls. You can see that we’ve bound one service descriptor, the HelloService, to our HelloServiceImpl implementation. The name of the Service Descriptor you bind will be used as the Service name, that is used in cross-service communication to identify the client making a request.

We’ve used Macwire’s wire macro to wire the dependencies into HelloServiceImpl - at the moment our service actually has no dependencies so we could just construct it manually ourselves, but it’s not likely that a real service implementation would have no dependencies.

The HelloApplication is an abstract class, the reason for this is that there is still one method that hasn’t been implemented, the serviceLocator method. The HelloApplication extends LagomApplication which needs a serviceLocator: ServiceLocator, a lagomServer: LagomServer and a wcClient: WSClient. We provide the wcClient: WSClient mixing in AhcWSComponents and we provide lagomServer: LagomServer programmatically. A typical application will use different service locators in different environments, in development, it will use the service locator provided by the Lagom development environment, while in production it will use whatever is appropriate for your production environment, such as the service locator implementation provided by Akka Discovery Service Locator. So our main application cake leaves this method abstract so that it can mix in the right one depending on which mode it is in when the application gets loaded.

Having created our application cake, we can now write an application loader. Play’s mechanism for loading an application is for the application to provide an application loader. Play will pass some context information to this loader, such as a classloader, the running mode, and any extra configuration, so that the application can bootstrap itself. Lagom provides a convenient mechanism for implementing this, the LagomApplicationLoader:

import com.lightbend.lagom.scaladsl.server._
import com.lightbend.lagom.scaladsl.api.ServiceLocator
import com.lightbend.lagom.scaladsl.devmode.LagomDevModeComponents

class HelloApplicationLoader extends LagomApplicationLoader {
  override def loadDevMode(context: LagomApplicationContext) =
    new HelloApplication(context) with LagomDevModeComponents

  override def load(context: LagomApplicationContext) =
    new HelloApplication(context) {
      override def serviceLocator = ServiceLocator.NoServiceLocator
    }

  override def describeService = Some(readDescriptor[HelloService])
}

The loader has two methods that must be implemented, load and loadDevMode. You can see that we’ve mixed in different service locators for each method, we’ve mixed in LagomDevModeComponents that provides the dev mode service locator and registers the services with it in dev mode, and in prod mode, for now, we’ve simply provided NoServiceLocator as the service locator - this is a service locator that will return nothing for every lookup. We’ll see in the deploying to production documentation how to select the right service locator for production.

A third method, describeService, is optional, but may be used by tooling to discover what service APIs are offered by this service. The metadata read from here may in turn be used to configure service gateways and other components.

Finally, we need to tell Play about our application loader. We can do that by adding the following configuration to application.conf:

play.application.loader = com.example.HelloApplicationLoader

§Defining your own components

To create a complex Application (one that uses Persistence, Clustering, the Broker API, etc…) you will need to mix in many Components. It is a good choice to create small custom traits that mix in some of those Components and build the Application by mixing in your small custom traits. That will let you test parts of the complete Application in isolation.

Imagine your Service consumes messages from a Broker topic where Orders are notified. Then your service stores that info into a database and does some processing with a final step of invoking a third party endpoint. If you only wanted to test the consuming of messages and proper storage you could create an OrderConsumingComponent trait and mix in LagomServiceClientComponents and CassandraPersistenceComponents so that you could consume the messages and store them. On your test you could extend your OrderConsumingComponent with TestTopicComponents that provides a mocked up broker so you didn’t need to start a broker to run the tests. Finally, on your Application you would mix in the tested OrderConsumingComponent and LagomKafkaClientComponents.

Lagom provides several Components. For this reference guide we grouped the components by feature:

Found an error in this documentation? The source code for this page can be found here. Please feel free to edit and contribute a pull request.