Connect the enterprise with the JCA, Part 1

A look at the J2EE Connector Architecture

The EAI (enterprise application integration) product category has grown significantly over the last 10 years. EAI eases the integration of disparate enterprise information systems (EIS). Although products such as Tibco and Vitria targeting the EAI market have succeeded, they have yet to achieve widespread adoption. As one of its missions, the new JCA (J2EE Connector Architecture) strives to bring EAI into mainstream use.

Read the whole series on the JCA:

The emerging JCA standard provides a mechanism to store and retrieve enterprise data in J2EE (Java 2 Platform, Enterprise Edition). The latest versions of many application servers, including BEA's WebLogic and IBM's WebSphere, support JCA adapters for enterprise connectivity. Using JCA to access an EIS is akin to using JDBC (Java Database Connectivity) to access a database.

Before JCA, each EAI vendor created a proprietary resource adapter interface for its own EAI product, requiring a resource adapter to be developed for each EAI vendor and EIS combination (for instance you'd need a SAP resource adapter for Vitria and a SAP resource adapter for Tibco). To solve that problem, as one of its main thrusts, JCA attempts to standardize the resource adapter interfaces.

In this article, I first deliver a high-level introduction into the JCA. Then I discuss how JCA fits into an integration strategy. After that I compare JCA to EAI vendors' products. Finally, I discuss the limitations of the current JCA platform, followed by what the future may hold.

How JCA and J2EE compare to EAI products

With that background in mind, let's consider how the current version of the JCA specification -- as well as J2EE in general -- measure up to some of the features found in EAI vendors' products.

Many EAI vendors, Vitria and Tibco for example, have either announced JCA support, or are in the process of releasing products that incorporate JCA-based adapters. Because the JCA 1.0 specification was finalized in July 2001, don't expect JCA in its initial release to match feature for feature to an EAI vendor's product, nor is that the aim. (Many features of the J2EE platform also compare to features in many EAI products.)

In light of this, and before we can discuss how JCA fits into the EAI picture, it's important to first understand some basic EAI features:

  • Resource adapters
  • Data mapping
  • Messaging brokers
  • Workflow

Let's look at each.

Resource adapters

Most EAI vendors include proprietary adapters built to work with their products. Most proprietary adapters allow for synchronous and asynchronous communication to an EIS. JCA adapters closely resemble those adapters, except JCA adapters include only a synchronous communication channel. Resource adapters represent the EAI feature JCA most directly matches, although most EAI vendors' adapters offer a larger feature set (for instance asynchronous capability) than JCA adapters.

Data mapping

With the next EAI feature -- data mapping -- data acquired in one format (for instance in the EIS's native format) by the resource adapter must be transformed into the format required for the business object. Mapping data from one system to another often proves the most time consuming aspect of system integration because you must map each business object in both systems. In response, most EAI vendors provide visual tools to enable a developer to set up such mapping.

While JCA does not offer an equivalent data-mapping facility, EJB's (Enterprise Java Beans) Container Managed Persistence (CMP) facility provides a similar functionality. However, currently not all EJB containers can use EJB CMP with JCA (using JCA as a data resource in place of JDBC). Presumably this will change as JCA becomes more widely adopted.

Message broker

Messaging brokers, another feature common to many EAI products, usually enable both point-to-point and publish/subscribe messaging. EAI products often employ messaging as the connectivity layer to tie together disparate systems.

Currently JCA does not address connectivity to an EIS in a message-oriented manner. It is possible, however, to implement some of a message broker's feature set in an EAI product by using JMS (Java Messaging Service), which is part of J2EE.

Workflow

Workflow is the management of business processes. Think of workflow as a coordinator. In and of itself, workflow lacks the capability to do anything, but rather relies on business objects, messages, and other external entities to perform functionality (for example creating a user object in a database). Workflow coordinates the use of business objects, messages, and so on to perform business processes.

JCA does not address workflow. However, look for something in the J2EE arena to address workflow, as it serves as an important component when developing a complex system.

Having discussed how JCA (and J2EE) compare with EAI tools, it's now important to look at how JCA fits into an overall integration strategy.

JCA and a general integration strategy

These days, many systems must integrate with other systems. But, what does that mean? In this section, I describe various integration types and where JCA fits into them.

Integration falls into two main buckets:

  • Inbound integration: outside systems initiate data requests to your system
  • Outbound integration: your system initiates data requests to other systems

All of the following integration types can be applied in both an inbound and an outbound manner.

User interface integration

User interface (UI) integration represents the most coarse-grain type of integration. UI-level integration implies that the data passed between systems will exist in the form of a UI representation. An outbound integration at the UI level entails requesting the UI (most likely a Webpage) from a remote system, then possibly manipulating it before displaying it as if it were part of your system's UI. An inbound integration at the UI level entails allowing an outside system to request UI pages on your system for inclusion on a remote system.

Prefer UI integration over other options when it is unimportant to distinguish the data type being retrieved. UI integration often requires the least effort to implement.

Message integration

Message-level integration, all the rage with the advent of Web services, implies that the data passed between systems will be in the form of a message (a defined, data-driven text format). An outbound message integration involves requesting data from a remote system in a message form (most likely a SOAP (Simple Object Access Protocol) message). With an inbound integration, your system receives a request for data via a message and responds with a message.

Message-oriented integration lends itself to loose coupling between systems because the systems remain unaware of the object types that exist on the remote system. That type of loose coupling works well with applications that wish to communicate over the Internet.

Object/RPC integration

Object/RPC (remote procedure call) integration implies integrating systems using distributed objects (that is, using EJB calls to integrate). With object-level integration, data passes between systems as parameters to method calls. In an outbound object-level integration, your system invokes objects on remote systems, while in an inbound object-level integration, a remote system calls objects on your system to retrieve data.

One of an object-level integration's main advantages is that you can call detailed APIs with full type safety and easily propagate the error codes and exceptions between systems.

Data integration

Lastly, data-level integration implies that the data passed between systems will be in a data/record-oriented manner. In an outbound data-level integration, your system requests data in a record-oriented fashion from other systems. With an inbound data-level integration, a remote system requests data from your system in a record-oriented manner.

The advantage of a data level integration: it lends itself to data mapping from one system onto the business objects in another system. JCA falls into the data-level integration category and therefore has the strengths and weaknesses of this integration type.

Now that we've covered where JCA fits into the overall integration puzzle, we are ready to discuss the JCA's structure.

JCA's structure

Let's begin with a JCA overview. Its main components include the resource adapter, system contracts, and the Common Client Interface (CCI), which together give JCA the power to access data in enterprise systems.

Resource adapter

In order to use JCA in a J2EE container, you first must have a JCA resource adapter, which resembles a JDBC driver. A JCA adapter is specific to an EIS (for example SAP or PeopleSoft) and is contained in a Resource Adapter Archive (RAR) file composed of the jar files and native libraries necessary to deploy the resource adapter on a J2EE container.

A JCA adapter interacts with a J2EE server with system contracts. They enable the J2EE server to propagate the context in which a JCA adapter is being called. You'll find three types of system contracts:

  • Connection management
  • Transaction management
  • Security

Connection management contracts

The connection management contract describes the understanding a J2EE container has with the adapter regarding establishing, pooling, and tearing down connections. The connection management contract also allows listeners created on a connection to respond to events (for instance if the connection becomes lost or experiences an error). Also note that the underlying protocol an adapter uses to connect to an EIS is outside the scope of the JCA specification.

All JCA resource adapters must supply two implementations with the adapter. First, a ConnectionFactor provides a vehicle for creating connections. Second, the Connection class represents this particular resource adapter's underlying connection.

Transaction management contract

The transaction management contract controls transactions in two different ways. First, it allows distributed transactions, which provide a mechanism to propagate transactions that originate from inside an application server to an EIS system. For example, in an EJB, a transaction may be created. If this EJB then employs a JCA resource adapter, the transaction management contract enables the transaction to propagate to the EIS (via the application server invoking the resource adapter's X/Open XA interfaces). In that situation, the transaction manager on the application server would control multiple resources to conduct distributed transaction coordination (i.e., two-phase commit).

Second, the transaction management contract can control transactions by creating local transactions. Local transactions are local in the sense that they exist only on a particular EIS resource. The transaction contract allows these transactions to be controlled, but they are related to any transaction that exists on the application server where the JCA resource adapter is running.

Also note that the resource adapter need not implement the the transaction management contract. Making this optional allows for resource adapters in nontransaction resources.

Security contract

The security contract enables the application server to connect to an EIS system using security properties. The application server authenticates with the EIS system by using security properties composed of a principle (a user id) and credentials (a password, a certificate, and so on). An application server can employ two methods to authenticate to an EIS system (via a resource adapter). With the first method, container-managed sign-on, the security credentials configure when the resource adapter is deployed on the application server. You can choose from several ways to configure security properties when using container-managed sign-on. First, with Configured Identity, all resource adapter connections use the same identity when connecting to the EIS system. Second, with Principal Mapping, the principal used when connecting to the EIS system is based on a combination of the current principal in the application server and the mapping (which maps how the principal in the application server will map to a principal in the EIS system). The third is Caller Impersonation, where the principal used in the EIS system exactly matches the principal in the application server. The fourth is Credentials Mapping, which is similar to Caller Impersonation, except the type of credentials must be mapped from application server credentials to EIS credentials.

While it's easiest to configure the security properties at deployment time, such a strategy proves slightly less flexible because the security properties cannot change at runtime. As an alternative, you can configure security properties by component-managed sign-on, which allows you to pass security properties each time a connection is acquired from the resource adapter.

CCI

To retrieve and update data, you employ JCA's CCI layer, a procedure resembling using JDBC to call stored procedures. A JCA resource adapter is not required to support the CCI layer (the resource adapter creators can choose their own API set), and, even if the resource adapter does support CCI, it may also support an API specific for that particular adapter.

The CCI APIs can be divided into four sections: First, the APIs related to establishing a connection to an EIS, also referred to as the Connection Interfaces. The second area of the CCI APIs cover command execution on an EIS, referred to as the Interaction Interfaces. Third is the Record/ResultSet Interfaces, which encapsulate the query results to an EIS. The fourth area, referred to as the Metadata Interfaces, allows EIS's metadata (the type of data) to be queried.

After this brief overview of the CCI APIs, it is useful to look at an example that shows the query of an employee count from an EIS:

...
        int count;
        try {
            ConnectionSpec spec = new CciConnectionSpec(user, password);
            Connection con = cf.getConnection(spec);
            Interaction ix = con.createInteraction();
            CciInteractionSpec iSpec = new CciInteractionSpec();
            iSpec.setSchema(user);
            iSpec.setFunctionName("EMPLOYEECOUNT");
            RecordFactory rf = cf.getRecordFactory();
            IndexedRecord iRec = rf.createIndexedRecord("InputRecord");
            Record rec = ix.execute(iSpec, iRec);
            Iterator iter = ((IndexedRecord)rec).iterator();
            while(iter.hasNext()) {   
                Object obj = iter.next();
                if(obj instanceof Integer) {
                    count = ((Integer)obj).intValue();
                }
            }
            con.close();
        } 
        catch(Exception e) {
            e.printStackTrace();
        }
        System.out.println("Employee count is: " + count);
...

Limitations of the JCA 1.0 and its future

The JCA, as its single biggest weakness, lacks an asynchronous communication vehicle. As a consequence, pulling information out of an EIS proves straightforward, but having an EIS send information (for instance data updates) to your system is not currently in the JCA specification.

In another major weakness, the JCA specification lacks a common API for data access. The CCI mentioned above is optional, so no dependable mechanism exists that a developer can use to access data using JCA (only the system contracts are guaranteed to be consistent).

The good news is that most of the current JCA's limitations will be addressed in the specification's next version, JCA 2.0, currently in development as JSR (Java Specification Request) 112. Version 2.0 will address asynchronous capabilities, JMS integration with JCA, metadata for the CCI layer, and XML use in the CCI layer.

Wrap up

This article provided an overview the of the JCA 1.0 specification and how it fits into the EAI product category. While not widely deployed yet, the JCA specification should prove a critical tool for developing a large J2EE system. Part 2 in the JCA series will go into the details of JCA by developing a simple JCA adapter.

Dirk Reinshagen, an architect at Zaplet, Inc., a commercial software vendor in the San Francisco Bay Area, has more than eight years of software architecture and development experience. He holds a B.S. in computer science from Tufts University.

Learn more about this topic

  • Dirk Reinshagen's "XML Messaging" series (JavaWorld):

Join the discussion
Be the first to comment on this article. Our Commenting Policies
See more