Home > Articles > Software Development & Management

  • Print
  • + Share This
This chapter is from the book

Service Layer

by Randy Stafford

Defines an application's boundary with a layer of services that establishes a set of available operations and coordinates the application's response in each operation.

Enterprise applications typically require different kinds of interfaces to the data they store and the logic they implement: data loaders, user interfaces, integration gateways, and others. Despite their different purposes, these interfaces often need common interactions with the application to access and manipulate its data and invoke its business logic. The interactions may be complex, involving transactions across multiple resources and the coordination of several responses to an action. Encoding the logic of the interactions separately in each interface causes a lot of duplication.

A Service Layer defines an application's boundary [Cockburn PloP] and its set of available operations from the perspective of interfacing client layers. It encapsulates the application's business logic, controlling transactions and coordinating responses in the implementation of its operations.

How It Works

A Service Layer can be implemented in a couple of different ways, without violating the defining characteristics stated above. The differences appear in the allocation of responsibility behind the Service Layer interface. Before I delve into the various implementation possibilities, let me lay a bit of groundwork.

Kinds of “Business Logic”

Like Transaction Script (110) and Domain Model (116), Service Layer is a pattern for organizing business logic. Many designers, including me, like to divide “business logic” into two kinds: “domain logic,” having to do purely with the problem domain (such as strategies for calculating revenue recognition on a contract), and “application logic,” having to do with application responsibilities [Cockburn UC] (such as notifying contract administrators, and integrated applications, of revenue recognition calculations). Application logic is sometimes referred to as “workflow logic,” although different people have different interpretations of “workflow.”

Domain Models (116) are preferable to Transaction Scripts (110) for avoiding domain logic duplication and for managing complexity using classical design patterns. But putting application logic into pure domain object classes has a couple of undesirable consequences. First, domain object classes are less reusable across applications if they implement application-specific logic and depend on application-specific packages. Second, commingling both kinds of logic in the same classes makes it harder to reimplement the application logic in, say, a workflow tool if that should ever become desirable. For these reasons Service Layer factors each kind of business logic into a separate layer, yielding the usual benefits of layering and rendering the pure domain object classes more reusable from application to application.

Implementation Variations

The two basic implementation variations are the domain facade approach and the operation script approach. In the domain facade approach a Service Layer is implemented as a set of thin facades over a Domain Model (116). The classes implementing the facades don't implement any business logic. Rather, the Domain Model (116) implements all of the business logic. The thin facades establish a boundary and set of operations through which client layers interact with the application, exhibiting the defining characteristics of Service Layer.

In the operation script approach a Service Layer is implemented as a set of thicker classes that directly implement application logic but delegate to encapsulated domain object classes for domain logic. The operations available to clients of a Service Layer are implemented as scripts, organized several to a class defining a subject area of related logic. Each such class forms an application “service,” and it's common for service type names to end with “Service.” A Service Layer is comprised of these application service classes, which should extend a Layer Supertype (475), abstracting their responsibilities and common behaviors.

To Remote or Not to Remote

The interface of a Service Layer class is coarse grained almost by definition, since it declares a set of application operations available to interfacing client layers. Therefore, Service Layer classes are well suited to remote invocation from an interface granularity perspective.

However, remote invocation comes at the cost of dealing with object distribution. It likely entails a lot of extra work to make your Service Layer method signatures deal in Data Transfer Objects (401). Don't underestimate the cost of this work, especially if you have a complex Domain Model (116) and rich editing UIs for complex update use cases! It's significant, and it's painful—perhaps second only to the cost and pain of object-relational mapping. Remember the First Law of Distributed Object Design (page 89).

My advice is to start with a locally invocable Service Layer whose method signatures deal in domain objects. Add remotability when you need it (if ever) by putting Remote Facades (388) on your Service Layer or having your Service Layer objects implement remote interfaces. If your application has a Web-based UI or a Web-services-based integration gateway, there's no law that says your business logic has to run in a separate process from your server pages and Web services. In fact, you can save yourself some development effort and runtime response time, without sacrificing scalability, by starting out with a colocated approach.

Identifying Services and Operations

Identifying the operations needed on a Service Layer boundary is pretty straightforward. They're determined by the needs of Service Layer clients, the most significant (and first) of which is typically a user interface. Since a user interface is designed to support the use cases that actors want to perform with an application, the starting point for identifying Service Layer operations is the use case model and the user interface design for the application.

Disappointing as it is, many of the use cases in an enterprise application are fairly boring “CRUD” (create, read, update, delete) use cases on domain objects—create one of these, read a collection of those, update this other thing. My experience is that there's almost always a one-to-one correspondence between CRUD use cases and Service Layer operations.

The application's responsibilities in carrying out these use cases, however, may be anything but boring. Validation aside, the creation, update, or deletion of a domain object in an application increasingly requires notification of other people and other integrated applications. These responses must be coordinated, and transacted atomically, by Service Layer operations.

If only it were as straightforward to identify Service Layer abstractions to group related operations. There are no hard-and-fast prescriptions in this area; only vague heuristics. For a sufficiently small application, it may suffice to have but one abstraction, named after the application itself. In my experience larger applications are partitioned into several “subsystems,” each of which includes a complete vertical slice through the stack of architecture layers. In this case I prefer one abstraction per subsystem, named after the subsystem. Other possibilities include abstractions reflecting major partitions in a domain model, if these are different from the subsystem partitions (e.g., ContractsService, ProductsService), and abstractions named after thematic application behaviors (e.g., RecognitionService).

Java Implementation

In both the domain facade approach and the operation script approach, a Service Layer class can be implemented as either a POJO (plain old Java object) or a stateless session bean. The trade-off pits ease of testing against ease of transaction control. POJOs might be easier to test, since they don't have to be deployed in an EJB container to run, but it's harder for a POJO Service Layer to hook into distributed container-managed transaction services, especially in interservice invocations. EJBs, on the other hand, come with the potential for container-managed distributed transactions but have to be deployed in a container before they can be tested and run. Choose your poison.

My preferred way of applying a Service Layer in J2EE is with EJB 2.0 stateless session beans, using local interfaces, and the operation script approach, delegating to POJO domain object classes. It's just so darned convenient to implement a Service Layer using stateless session bean, because of the distributed container-managed transactions provided by EJB. Also, with the local interfaces introduced in EJB 2.0, a Service Layer can exploit the valuable transaction services while avoiding the thorny object distribution issues.

On a related Java-specific note, let me differentiate Service Layer from the Session Facade pattern documented in the J2EE patterns literature [Alur et al.] and [Marinescu]. Session Facade was motivated by the desire to avoid the performance penalty of too many remote invocations on entity beans; it therefore prescribes facading entity beans with session beans. Service Layer is motivated instead by factoring responsibility to avoid duplication and promote reusability; it's an architecture pattern that transcends technology. In fact, the application boundary pattern [Cockburn PloP] that inspired Service Layer predates EJB by three years. Session Facade may be in the spirit of Service Layer but, as currently named, scoped, and presented, is not the same.

When to Use It

The benefit of Service Layer is that it defines a common set of application operations available to many kinds of clients and it coordinates an application's response in each operation. The response may involve application logic that needs to be transacted atomically across multiple transactional resources. Thus, in an application with more than one kind of client of its business logic, and complex responses in its use cases involving multiple transactional resources, it makes a lot of sense to include a Service Layer with container-managed transactions, even in an undistributed architecture.

The easier question to answer is probably when not to use it. You probably don't need a Service Layer if your application's business logic will only have one kind of client—say, a user interface—and its use case responses don't involve multiple transactional resources. In this case your Page Controllers can manually control transactions and coordinate whatever response is required, perhaps delegating directly to the Data Source layer.

But as soon as you envision a second kind of client, or a second transactional resource in use case responses, it pays to design in a Service Layer from the beginning.

Further Reading

There's not a great deal of prior art on Service Layer, whose inspiration is Alistair Cockburn's application boundary pattern [Cockburn PloP]. In the remotable services vein [Alpert, et al.] discuss the role of facades in distributed systems. Compare and contrast this with the various presentations of Session Facade [Alur et al.] and [Marinescu]. On the topic of application responsibilities that must be coordinated within Service Layer operations, Cockburn's description of use cases as a contract for behavior [Cockburn UC] is very helpful. An earlier background reference is the Fusion methodology's recognition of “system operations” [Coleman et al.].

Example: Revenue Recognition (Java)

This example continues the revenue recognition example of the Transaction Script (110) and Domain Model (116) patterns, demonstrating how Service Layer is used to script application logic and delegate for domain logic in a Service Layer operation. It uses the operation script approach to implement a Service Layer, first with POJOs and then with EJBs.

To make the demonstration we expand the scenario to include some application logic. Suppose the use cases for the application require that, when the revenue recognitions for a contract are calculated, the application must respond by sending an e-mail notification of that event to a designated contract administrator and by publishing a message using message-oriented middleware to notify other integrated applications.

We start by changing the RecognitionService class from the Transaction Script (110) example to extend a Layer Supertype (475) and to use a couple of Gateways (466) in carrying out application logic. This yields the class diagram of Figure 9.7. RecognitionService becomes a POJO implementation of a Service Layer application service, and its methods represent two of the operations available at the application's boundary.

The methods of the RecognitionService class script the application logic of the operations, delegating to domain object classes (of the example from Domain Model (116)) for domain logic.

public class ApplicationService { 
   protected EmailGateway getEmailGateway() {
      //return an instance of EmailGateway
   protected IntegrationGateway getIntegrationGateway() {
      //return an instance of IntegrationGateway
public interface EmailGateway {
   void sendEmailMessage(String toAddress, String subject, String body);
public interface IntegrationGateway {
   void publishRevenueRecognitionCalculation(Contract contract);
public class RecognitionService
extends ApplicationService {
   public void calculateRevenueRecognitions(long contractNumber) {
      Contract contract = Contract.readForUpdate(contractNumber);
         "RE: Contract #" + contractNumber,
         contract + " has had revenue recognitions calculated.");
   public Money recognizedRevenue(long contractNumber, Date asOf) {
      return Contract.read(contractNumber).recognizedRevenue(asOf);

Persistence details are again left out of the example. Suffice it to say that the Contract class implements static methods to read contracts from the Data Source layer by their numbers. One of these methods has a name revealing an intention to update the contract that's read, which allows an underlying Data Mapper (165) to register the read object(s) with for example, a Unit of Work (184).

Transaction control details are also left out of the example. The calculateRevenueRecognitions() method is inherently transactional because, during its execution, persistent contract objects are modified via addition of revenue recognitions; messages are enqueued in message-oriented middleware; and e-mail messages are sent. All of these responses must be transacted atomically because we don't want to send e-mail and publish messages to other applications if the contract changes fail to persist.

In the J2EE platform we can let the EJB container manage distributed transactions by implementing application services (and Gateways (466)) as stateless session beans that use transactional resources. Figure 9.8 shows the class diagram of a RecognitionService implementation that uses EJB 2.0 local interfaces and the “business interface” idiom. In this implementation a Layer Supertype (475) is still used, providing default implementations of the bean implementation class methods required by EJB, in addition to the application-specific methods. If we assume that the EmailGateway and IntegrationGateway interfaces are also “business interfaces” for their respective stateless session beans, then control of the distributed transaction is achieved by declaring the calculateRevenueRecognitions, sendEmailMessage, and publishRevenueRecognitionCalculation methods to be transactional. The RecognitionService methods from the POJO example move unchanged to RecognitionServiceBeanImpl.

The important point about the example is that the Service Layer uses both operation scripting and domain object classes in coordinating the transactional response of the operation. The calculateRevenueRecognitions method scripts the application logic of the response required by the application's use cases, but it delegates to the domain object classes for domain logic. It also presents a couple of techniques for combating duplicated logic within operation scripts of a Service Layer. Responsibilities are factored into different objects (e.g., Gateways (466)) that can be reused via delegation. A Layer Supertype (475) provides convenient access to these other objects.

Some might argue that a more elegant implementation of the operation script would use the Observer pattern [Gang of Four], but Observer is difficult to implement in a stateless, multithreaded Service Layer. In my opinion the open code of the operation script is clearer and simpler.

Some might also argue that the application logic responsibilities could be implemented in domain object methods, such as Contract.calculateRevenueRecognitions(), or even in the data source layer, thereby eliminating the need for a separate Service Layer. However, I find those allocations of responsibility undesirable for a number of reasons. First, domain object classes are less reusable across applications if they implement application-specific logic (and depend on application-specific Gateways (466), and the like). They should model the parts of the problem domain that are of interest to the application, which doesn't mean all of the application's use case responsibilities. Second, encapsulating application logic in a “higher” layer dedicated to that purpose (which the data source layer isn't) facilitates changing the implementation of that layer—perhaps to use a workflow engine.

As an organization pattern for the logic layer of an enterprise application, Service Layer combines scripting and domain object classes, leveraging the best aspects of both. Several variations are possible in a Service Layer implementation—for example, domain facades or operation scripts, POJOs or session beans, or a combination of both. Service Layer can be designed for local invocation, remote invocation, or both. Most important, regardless of these variations, this pattern lays the foundation for encapsulated implementation of an application's business logic and consistent invocation of that logic by its various clients.

  • + Share This
  • 🔖 Save To Your Account