7.3 Service Loose Coupling
In a service-oriented environment, components can be coupled at a number of different levels. Coupling can occur at the level of the service contract, the service implementation logic, or the underlying technology platform the service is running on if a service consumer is coupled with a particular service. In general, SOA promotes the notion of decoupled systems by which the parts of a service-oriented solution are as decoupled as possible. Reducing the dependency of one part on another allows one area, such as a service or an aggregation, to be changed without requiring changes to other areas.
The following design characteristics are helpful in the creation of decoupled systems:
- separation of contract and implementation
- functional service context with no dependency on outside logic
- minimal requirements for the service consumer
When service logic is implemented in Java and executed in a JRE, a coupling is created between the logic and the underlying technology platform, Java. Additional dependencies arise when Java EE is used as the hosting platform. For an in-depth explanation of the variations of positive and negative coupling, see Chapter 7 in SOA Principles of Service Design.
Separation of Contract and Implementation
The separation of a service contract from its implementation is a key characteristic of service-oriented design that was first established as part of the definition of remote procedure calls and then in object-oriented programming. The interface of a component is exposed to a point that allows other logic to invoke this component, but no details about the implementation are added. Changes to the implementation can then be made without affecting the client. SOA took this concept further by establishing the notion of a standards-based contract. For example, the service interface can be expressed in a programming language-neutral way to allow for the integration of logic written in different languages and running on different platforms.
The choice of programming language can be exposed in the service contract. In the context of a SOAP Web service from the Top-Down vs. Bottom-Up section, a common approach is to generate the service contract from existing Java logic. In JAX-WS, the wsgen tool can be used to generate a complete WSDL document, including the XML schema definition from an existing JavaBean or Java interface. However, the existing Java code may not be easily mapped into XML schema.
For example, assume that the public java.util.Hashtable<String, String> getTable(int i); method is part of an interface to be turned into a service. Using the wsgen tool to generate both the WSDL and XML schema definition from this interface creates the type definitions presented in Example 7.33.
<xs:element name="getTableResponse" type="ns2:getTableResponse" xmlns:ns2="http://the.package/"/> <xs:complexType name="getTableResponse"> <xs:sequence> <xs:element name="return" type="ns3:hashtable" minOccurs="0" xmlns:ns3="http://the.package/"/> </xs:sequence> </xs:complexType> <xs:complexType name="hashtable"> <xs:complexContent> <xs:extension base="ns4:dictionary" xmlns:ns4= "http://the.package/"> <xs:sequence/> </xs:extension> </xs:complexContent> </xs:complexType> <xs:complexType name="dictionary" abstract="true"/>
The result is a valid XML schema definition, although a Java Hashtable is exposed in the service interface which renders the schema not useful or usable outside of Java. The same is true for many other classes and interfaces in Java, which, while useful as part of the implementation, do not map well into XML schema and non-Java environments. Many of the more technically-oriented classes, including the members of the java.io package, fall into this category.
The use of internal, platform-specific information carried in a generic type is another aspect of decoupling the contract and implementation. For example, the name of a file usually stored as a string can be mapped into XML schema and added to a service contract. However, exposing file names on a service contract is considered poor practice, and revealing details of the service implementation to the service consumer should be avoided. The same holds true for names of databases, database tables, machine names, and addresses.
Using the reverse approach is another way of creating this coupling, particularly when generating a service logic skeleton directly from a service contract. In JAX-WS, the wsimport tool is used to create Java code skeletons representing a given WSDL definition. Despite being the recommended approach, be aware that the generated Java code is now tightly coupled to its contract, which prevents the code from being easily reused to serve other types of service contracts or updated versions of the current contract. In most cases, the tight coupling is acceptable because the logic is created for a particular contract and nothing else.
However, instances occur where service logic must be created for reuse despite changes to the underlying contract changes or the need to concurrently serve multiple versions of the same contract. The JAX-WS Provider API is equipped for such instances. In this model, the service logic parses the incoming message at runtime for dynamic processing with no direct dependency on the types and elements defined in the WSDL contract. Use of the JAX-WS Provider API is detailed in Chapter 8.
For a REST service, coupling the code to the service contract is inconsequential because generating implementation artifacts from a machine-readable contract, unless WADL is being used, is uncommon. Implementing a Web resource for a platform, such as JAX-RS, out of the box supports only a finite subset of Java classes that can be automatically mapped to appropriate content-types. For custom types not mapped automatically by the built-in content handlers, developers must provide implementations of javax.ws.rs.core.MessageBodyReader/MessageBodyWriter interfaces to map such types to a known set of resource representations or media types.
Independent Functional Contexts
Besides the direct compile-time of coupling services, consider the coupling of a service to its outside functional context. Service invocations happen as part of a business transaction or process to establish a context that effectively binds the services, which are invoked downstream, into an aggregated set. Having a set of services as part of the same process establishes a type of coupling between the services, a coupling that should always be top-down. Particularly when leveraging other services to fulfill functionality, a service implementation can be coupled with or have dependency on those services. However, the service should not have a dependency on any services at a higher level of the process hierarchy, or be coupled with the invoking service or with peer services.
For example, assume that a service offers Credit Check functionality. The service is implemented as a business process that invokes a number of other finer-grained services, such as Credit Lookup, Credit Eligibility, Update Profile, and Notification. All four downstream services are peers within the Credit Check business process. Service peers should have no dependency on each other, or be aware of or coupled with their upstream Credit Check service.
Services representing business processes, such as Credit Check and Maintain Customer Information, are decomposed into a set of fine-grained services to create a downstream dependency. A service should not introduce a dependency on another service that is higher level. For example, the implementation of the Update Profile entity service should not introduce any dependency on the Maintain Customer Information task service.
In Java, the same principles of downstream dependency hold true. Unwanted dependencies can be detected by examining the classes that are used by a piece of Java logic. Organizing classes into packages directly identifying the service and/or affiliated business process and ensuring that logically decoupled functions are not packaged together is recommended. A package name should be selected with consideration for possible reuse opportunities. For Web services, the same requirements for namespaces are used in the service contract.
Service Consumer Coupling
For SOAP-based Web services, a service consumer will often be tightly coupled with the service contract and not the implementation of the service being invoked. However, a looser coupling lessens the impact when the service contract changes.
Using the JAX-WS Dispatch API, service consumer logic can dynamically assemble a request message at runtime and manage any response messages. Additional effort is required to build service consumer logic that can build request messages that the intended service can process.
Another way of further decoupling a service consumer from the service is to insert an intermediary between the two. The intermediary, generally deployed as part of an ESB, can mediate the differences in message format and network protocol to further decouple the service consumer and service. Chapter 12 explores this further as part of its coverage of ESBs.