This chapter is from the book
This chapter is from the book
This chapter is from the book
In addition to accurately reflecting the domain model of the enterprise, an application design model must also address application characteristics that could cause problems at run time, as well as when maintaining and modifying the application. The following three design guidelines help to avoid major problems that usually appear in applications with legacy system integration unless addressed during the design phase
-
Performance: Minimize access to the persistent store to mitigate the performance impact of accessing legacy systems.
-
Flexibility and reuse: Design the integration tier so that component reuse and re-implementation become easy and straightforward.
-
Integration: Overcome the impedance mismatch between an object-oriented data representation and its corresponding representation in a legacy system.
These guidelines are based on the assumption that the application is structured in tiers. Note, however, that only software tiers are considered here; in this context, an application tier can reside on any physical tier. For example, all application tiers can reside on the same physical tier, or one application tier can span multiple physical tiers. Figure 2.6 shows the application tiers as defined in Core J2EE Patterns [D. Alur, J. Crupi, D. Malks, Prentice-Hall 2001].
)
Figure 2.6. A five-tiered architecture
From this figure, it is clear that the role played by the integration tier of an application is to isolate the business tier from the resource tier, in order to make the business tier resilient and minimize the impact of changes in the resource tier.
In creating a design model from a domain model, the core business entities, types, utilities, and relations of the domain model are mapped to elements such as Transfer Objects and Data Access Objects, from which the integration tier is designed. The following sections discuss these concepts and how to apply them to an integration tier design.
Managing Entity-to-Entity Relations
Overly fine-grained object-oriented applications tend to have a performance impact on large-scale distributed systems. Increased component granularity provides a means of controlling that impact. Communication and data access, the main bottlenecks in any distributed system, are normally the focal points for any optimizations made in the application architecture.
In order to minimize communication between the business tier and the resource tier, two principles are essential:
-
The business tier should access only large chunks of data, that is, whole objects rather than single attributes.
-
The business tier should resolve only those entity-to-entity relations that are required by the application.
Applications often need to retrieve collections of business entities from the persistent store. For instance, an application might provide a method to retrieve a list of all the Arrangements of a Party (see Figure 1.1). The representation of Transfer Object collections is an area in which performance can be improved by careful design. How should the Transfer Objects Party and Arrangement be designed to represent the relationship between them?
){kind=link}
Using the traditional object-oriented approach, the class representing Party would be designed to have a collection of references to Arrangement objects. However, this approach sometimes leads to complex situations in which data not actually required by the application is retrieved from persistent store, or in which circular references must be addressed (see Figure 2.7).
)
Figure 2.7. Transfer Object containing object references
To avoid these problems, Transfer Objects should not contain object references to other Transfer Objects. Associations should instead be represented as object identities or primary keys. To continue our example, the Party contains an array of identifiers representing primary keys of Arrangements (see Figure 2.8).
)
Figure 2.8. Transfer Objects containing ID attributes
Composite Transfer Objects
A Composite Transfer Object is a representation of an association between two or more Transfer Objects. In our example, the association between a Party and its Arrangements can be represented by a class called, for example, PartyArrangements, that contains an attribute of type Party and a second attribute consisting of a sequence of Arrangement objects (see Figure Figure 2.9).
)
Figure 2.9. A Composite Transfer Object containing Party and Arrangements
The purpose of representing associations in separate objects is to prevent the application from reading the entire database when instantiating a Transfer Object. For example, in the domain model depicted in Figure 1.1, a Party has associations to Arrangements that have associations to Products that have associations to Customers that have associations to a Party. The risk of reading the entire database is apparent, since domain models often tend to relate everything to everything.
By using Composite Transfer Objects, an application can resolve only the needed associations (in our example, only the association between Party and Arrangement), leaving the remaining associations unresolved (in our example, this would mean that the association between Arrangement and Product would remain unresolved, among others).
This approach is sometimes referred to as lazy evaluation. The motivation for this design can also be found in Design Patterns [E. Gamma et al., Addison-Wesley 1994], in which it is shown that decoupling objects allows varying their interactions without having to subclass or modify them. Furthermore, when the domain model changes, only the mediator class (that is, the Composite Transfer Object) must be redesigned.
The disadvantage with this approach is that unresolved associations lose the type information normally contained in object references. For example, leaving the association from Arrangement to Product unresolved means that when the application wants to access this association, the type information is not available. The application itself, therefore, must be aware of which Transfer Object class (for example, Product) to retrieve from persistent store in order to resolve the association. This domain knowledge can, however, be contained in a Business Rule Object, as discussed later in this chapter.
Design for Flexibility and Reuse
Making the integration tier components flexible and reusable is a common problem. Component-driven design is the key technique in obtaining flexibility and reuse. Any component should be designed according to the following principles:
-
A well-defined interface that hides the component's implementation, providing flexibility
-
Service-driven design, in which the component is designed to provide a service and does not care to whom this service is provided, thereby promoting reuse
Applying these principles to the integration tier gives it a good chance of becoming truly flexible and reusable. Furthermore, in order to make the integration tier's components usable by any device or application, they should be designed to make no assumptions about the presentation format of data retrieved from the resource tier.
Composite Transfer Objects, Transfer Objects, and Data Access Objects provide excellent means for achieving flexibility and reusability.
The use of Composite Transfer Objects, as discussed earlier, allows the associations between Transfer Objects to be created dynamically and managed in a flexible manner. By using Transfer Objects to represent data, a normalized data representation is provided within the business tier and toward the presentation tier. (Note that this use of Transfer Objects is an internal representation only. For external representation—for example, toward other, possibly non-Java applications—it must be possible to transform data from the Transfer Object format to other formats, XML being the prevailing choice.)
Data Access Objects integrate the business tier with the resource tier, providing the means for the business tier to access persistent data without knowledge of the implementation of the persistent store. In this way, the business tier is shielded from changes in underlying resources.
Designing the integration tier for flexibility also requires that logic concerning entities, relationships, and data access is located in the appropriate components. For example, logic concerning a calculation or a constraint based on the attributes of a single entity, such as a Party, could very well be located in a Transfer Object. Transfer Objects, however, should never directly access a persistent store or a remote service. If a Transfer Object is to access a persistent store or a remote service directly, it must contain logic that depends on the infrastructure of a particular application, which could make the Transfer Object less reusable in the context of another application running on a different infrastructure.
Generally, in order to make the application as flexible as possible, it is advisable to locate business logic outside plain data carriers such as Transfer Objects, by using Business Rule Objects, for example, as discussed in the following section.
Business Rule Objects
A Business Rule Object is a design model element that represents enterprise business logic. For example, an application can call a sequence of Business Rule Objects in a certain order to verify that the business rules of a particular method are followed.
Business Rule Objects should operate primarily on objects that are passed to it as method parameters. As with Transfer Objects, therefore, and for the same reasons, Business Rule Objects should not normally communicate with persistent stores or with remote services. It is possible, however, for a Business Rule Object to operate on Transfer Objects, as well as on Supporting Objects and Composite Transfer Objects.
A Validator is the interface of a Business Rule Object that is used to validate the contents of an object according to constraints set on the application. For example, this interface is used by the application to validate input parameters passed to the application, as well as the contents of Transfer Objects returned from Data Access Objects.
Other types of Business Rule Objects, given a Transfer Object and a relationship name, can return a Transfer Object class. For example, a Business Rule Object could be given a Party and the relationship name arrangements and return an Arrangement class object. The most important considerations in designing Business Rule Objects is that they should be both simple to modify (as this modifies the behavior of an entire application or set of applications) and self-contained, in order to be reusable in all tiers of an applications and between different applications.
For example, a Validator interface can provide a single method called validate, which returns true or false, while the class that implements the Validator interface in turn contains a constructor that accepts one or more Transfer Objects and/or Supporting Objects. Complex business rules can then be composed of a number of Validators combined in a logical expression.
For example, assuming Validators A, B, and C, these could be combined as follows:
boolean d = A.validate() & (B.validate() | C.validate());
Now consider the business event “Withdraw an amount from an account” and let A, B, and C be as follows:
-
A.validate: Return true if the customer is the owner of the account
-
B.validate: Return true if the balance of the account is equal to or greater than the amount
-
C.validate: Return true if the customer has a credit tied to the account that is equal to or greater than the amount
Then define the composite rule this way: True if the customer is the owner of the account AND if the balance of the account is equal to or greater than the amount OR if the customer has a credit tied to the account that is equal to or greater than the amount.
The Transfer Objects involved in this rule could be, for example:
-
Rule A: Customer and Account
-
Rule B: Account
-
Rule C: Product and Account
This results in the object seen in Figure 2.10:
)
Figure 2.10. A CustomerAccountValidator object
Note that A, B, and C are self-contained and reusable in any tier and in any application that deals with Customers, Accounts, and Products. By changing the implementation of either one of the Validators, the behavior of the application can be changed without any major recoding.
Business Rule Objects can be implemented through the use of a commercial rule engine or by implementing them as normal Java objects.
Managing the Impedance Mismatch
As object-oriented applications are developed on top of legacy systems and data sources that are not object-oriented, the impedance mismatch between the object-oriented structuring of enterprise data (entities) and the corresponding legacy or relational structuring becomes apparent. Most enterprises have an increasing need to become less dependent on their legacy systems. Therefore, domain models created for new applications without taking legacy systems into consideration can deviate dramatically from the legacy systems' structuring of enterprise data. Unless this object-oriented structuring is adhered to, however, J2EE applications risk becoming the legacy systems of the 21st century.
The Enterprise JavaBeans (EJB™) specification provides entity beans in combination with container-managed persistence as the primary method for encapsulating data sources. But there are situations when this approach cannot be taken. As discussed earlier, entity beans introduce a potential scalability and cache inconsistency problem; for some types of applications, therefore, entity beans are not feasible. When it comes to container-managed persistence, there are a number of situations in which straightforward object-to-persistent data mappings are not possible. The following section discusses this in more detail.
When Container-Managed Persistence Cannot Be Used
With container-managed persistence (CMP), the EJB container manages data source access and mappings between transient entities (represented as entity beans) and persistent entities (represented as persistent records). In order for an application to use CMP, the following conditions must be met:
-
CMP is used only with entity beans.
-
The EJB Container Provider provides a tool for mapping entity beans to persistent records.
As noted, there are a number of situations in which CMP cannot be used and they are explained in this list:
-
More than one entity is mapped to a single legacy system service. (In this case, a CMP implementation might call the same service more than once.)
-
An attribute value might need to be converted. (For example, a value returned might need to be converted from uppercase to lowercase or be internationalized.)An attribute of an entity is mapped to a part of a field in a record. (This means that the field must be parsed and that business rules might need to be applied.)
-
An entity is mapped to multiple legacy system services. (For example, a list service might be followed by a read service based on the values of one of the rows returned in the list.)
-
An entity is mapped to more than one data source, and it is necessary to pick attribute values from different sources and combine them into a single Transfer Object.
-
The data source does not provide support for distributed transactions.
The Data Access Object pattern, in combination with Bean-Managed Persistence (BMP), provides a solution for all of these situations. The next chapter describes this solution in detail.
Pitfalls of Caching Persistent Data
Because data access is a potential bottleneck in any distributed system, it is tempting to use application data caching to reduce the number of transactions with the resource tier. In a thousands-of-transactions-per-second application, a cache can help significantly by avoiding the trip to the resource tier whenever possible. It is important, however, to be aware of two potential pitfalls in the use of caches in distributed systems:
-
An application data cache can introduce a single point of failure. Where the application's data caches cannot be replicated in a transactional way without using a database, a particular application cache is useful only if all requests arrive at the same machine. That cache thus becomes a single point of failure. Transactions to synchronize distributed caches are needed; without them, a cache replica might be updated by a client before being resynchronized with the other caches.
-
An application cache will become inconsistent with the persistent store as soon as the application is no longer the only way to access the persistent store. For example, if the persistent store is part of a legacy system, there will be numerous other applications accessing the persistent store through other channels. Hence the cache will constantly become invalid, requiring frequent refresh calls to the legacy system.
Applying this knowledge to an application can guide us in choosing the right granularity for data access, and can also be helpful in choosing between session beans and entity beans.
Entity beans are meant to serve as a cache for persistent data that is accessible by many simultaneous requests, existing only in a single active instance for a particular business entity instance. If they existed in more than one instance, as noted above, the application could have a cache consistency problem.
Ideally, entity bean granularity would match the domain model entity granularity; as a cache, however, entity beans suffer from the problems described earlier. Furthermore, entity beans cannot be treated as rows in a relational database, since entity-to-entity bean relations and the joining of multiple entity beans into a single object create extreme performance demands.
When combined, and given the single point of failure risk, these drawbacks makes entity beans inappropriate for applications that need to serve large numbers of simultaneous requests to the same entity. Furthermore, they are inappropriate for applications that need to integrate with legacy systems, due to the cache inconsistency problem caused by other applications.
In order to avoid these pitfalls, it is essential to remember that an application data cache often cannot be used for shared data using general-purpose J2EE application servers. Instead, data access must normally be performed in large chunks, fetching as much as possible each time. The use of the Data Access Object pattern in combination with coarse-grained legacy system calls provides a means of combining the fine-grained domain model with a coarse-grained legacy system representation of core business entities.
Data Access Objects
The Data Access Object pattern addresses the impedance mismatch discussed in the previous section by isolating the business tier components from the actual interfaces to the legacy systems. Since the data passed to and returned from these interfaces are normally not formatted according to the domain model used by modern applications, the most important task of the Data Access Object is to reformat the enterprise data into Transfer Objects and vice versa. Furthermore, by designing Data Access Objects and Transfer Objects based on entities of the domain model rather than on the information structure of the legacy systems, the impedance mismatch can be isolated within the integration tier, not affecting the remainder of the application.
If we view the Data Access Object as having two parts, an interface and an adapter [Design Patterns, E. Gamma et. al., Addison-Wesley 1994], the way in which the Data Access Object pattern provides a mechanism for application portability across legacy systems becomes straightforward. By keeping the interface while replacing the adapter, a business tier component can access data from legacy system A today and legacy system B tomorrow. Furthermore, standardizing the interface of the Data Access Object allows the final decoupling of the application from the information structure of the legacy system.
It is important to realize that the interface of the Data Access Object must not provide any services beyond what would be expected from any data source. Normally, these services are creating, reading, updating, deleting, and searching. By keeping strictly to this set of services, the ability of the Data Access Object to adapt to new data sources can be maximized.
Figure 2.11 shows a session bean consisting of two Data Access Objects. Each Data Access Object manages Transfer Object instances of a different class, each containing data from three different data sources. These data sources can be included in distributed transactions controlled from the session bean.
)
Figure 2.11. A Session bean, two Data Access Objects, and three data sources
All persistent stores can be encapsulated within a Data Access Object. The purpose of a Data Access Object is to provide methods for creating, reading, updating, and deleting Transfer Objects (not Composite Transfer Objects or Supporting Objects). Distributed transactions can include multiple invocations of Data Access Objects. A Data Access Object must be dedicated to one and only one Transfer Object class—Party, for example. Since Transfer Objects can consist of data from more than one data source, a Data Access Object is able to merge data from one or more data sources into a single Transfer Object.
The interface of a Data Access Object should provide the following methods:
-
create(newVo:TransferObject):ID
Create a new record in the persistent store. Return a primary key to that record.
-
read(primaryKey:ID):TransferObject
Read a record given a primary key. An exception should be raised if there is no matching record.
-
update(updatedVo:TransferObject):void
Update a record with the data contained in the provided Transfer Object. An exception should be raised if there is no record containing the primary key provided in the Transfer Object.
-
delete(primaryKey:ID):void
Delete a record given a primary key. An exception should be raised if there is no matching record.
-
find(constraint:TransferObject, relOp:Comparator): Iterator
Find Transfer Objects having attribute values matching the values provided in the constraint. An object implementing the java.util.Comparator interface is provided to compare the constraint with the Value Objects retrieved from the data source. The Comparator object can be implemented specifically for each Transfer Object to provide a method for comparing two objects for equality, greater than, or lesser than.
-
find(query:Query):Iterator
Return an Iterator object containing all records matching the query. The query must evaluate a set of Transfer Objects of the same class as managed by the Data Access Object. An iterator containing zero records is a valid result and should not raise an exception.
In cases in which some methods cannot be implemented due to limitations in the persistent store, the method can either raise an exception or provide a default implementation.
It is important to note that all Transfer Object instances passed to or returned from these methods in a particular Data Access Object must be of the same class—the Transfer Object class that is managed by the Data Access Object. However, because all Data Access Objects provide a common set of operations, it is possible to provide a common base interface for all Data Access Objects (see Figure 2.12).
)
Figure 2.12. The Data Access Object interface
As discussed in Chapter 1, since each Transfer Object class is to be managed by a corresponding Data Access Object class, a typical application will consist of a large number of different Data Access Object classes. A good approach to managing a large set of classes is through an Abstract Factory [Design Patterns, E. Gamma et. al., Addison-Wesley 1994]. The concept of a Data Access Object Factory is discussed in the next chapter.
Through the use of adapters, Data Access Objects provide application portability across legacy systems. This approach makes it possible for an enterprise to implement adapters based on the common domain model for all its legacy systems. Since these adapters adhere to the standardized Data Access Object interface, it is straightforward to implement J2EE applications based on the common domain model independently from the implementation of the adapters. This decoupling provides the means of reusing the integration tier from application to application. Furthermore, since the design model used to implement the integration tier is derived from the common domain model, a traceable, reusable software component can be utilized throughout the enterprise, with any legacy system.