Example: Sales Order Service
We will use the example of a Sales Order Service supporting an Order-to-Delivery business process to illustrate the various facets of observable dependencies and behaviors. An overview of the Order-to-Delivery business process is depicted in Figure 5-1. This figure presents a usage scenario for the Sales Order Service, which is one of the participants in the process. The Sales Order Service manages the full life cycle of an order. Let’s examine how it participates in the process and note the relevant observable facets that come into play.
Figure 5-1: Sales Order Service in the Order-to-Delivery Business Process
Placing the Order
The Sales Order Service accepts an order from the Web Site via the placeOrder() operation of the Sales Order Service Interface (Figure 5-2). The invocation of this operation constitutes a trigger, and the fact that it is invoked by the Web Site indicates a dependency on the interface.
Figure 5-2: Sales Order Service Interface—placeOrder() Operation
The invocation of the placeOrder() operation triggers an associated behavior of the service. This behavior validates the order and obtains payment, in the process deciding whether or not to accept the order. These activities involve interactions with other components, components that are not part of the service and are thus dependencies. Let’s look at the interfaces involved, and later we’ll identify the components that provide those interfaces.
The validity of each itemID is established by calling the validate-ProductID() operation of the Product Query Interface (Figure 5-3).
Figure 5-3: Product Query Interface—validateProductID() Operation
The validity of the customerID is established by calling the get-Customer() operation of the Customer Query Interface (Figure 5-4).
Figure 5-4: Customer Query Interface—getCustomer() Operation
Payment is obtained by calling the obtainPayment() operation on the Credit Interface (Figure 5-5).
Figure 5-5: Credit Interface—obtainPayment() Operation
At this point the service returns the Place Order Response to the waiting caller of placeOrder(). This data structure indicates whether the order was accepted and, if not accepted, the reason why. The data structure exposes the fact that the service is validating the order and obtaining payment, thus making this portion of the behavior observable to the caller of placeOrder().
The observable behavior does not end with the return of the Place Order Response data structure. If the order is accepted, the service then sends the order to Order Fulfillment using the fillOrder() operation of the Order Fulfillment Interface to accomplish this (Figure 5-6). This is yet another dependency. Note that, as designed, this is an In-Only operation and does not return a response.
Figure 5-6: Order Fulfillment Interface—fillOrder() Operation
At this point the behavior that began with the invocation of the placeOrder() operation comes to a conclusion. Putting all the pieces together, the behavior triggered by this invocation is that shown in Figure 5-7. This diagram also indicates components for which there are observable dependencies: Product Service, Customer Service, Credit Service, and Order Fulfillment Service.
Figure 5-7: Triggered Behavior for placeOrder()
When Order Fulfillment ships the items, it sends a copy of the shipment notice to the Sales Order Service. It does this by calling the orderShipped() operation of the Sales Order Status Interface (Figure 5-8). This is another dependency: Order Fulfillment depends on this interface. As designed, this is an In-Only operation.
Figure 5-8: Sales Order Status Interface—orderShipped() Operation
The triggered behavior related to this interaction, the update of order status, is simple, although somewhat obscure from an observability perspective (Figure 5-9). The obscurity arises because the update of the order status cannot be inferred from this interaction alone. It is only the fact that the order status can be retrieved (via other operations), coupled with the fact that this status indicates whether or not the order has shipped, that reveals the fact that the order status exists and has been changed. We have identified an element of observable state.
Figure 5-9: orderShipped() Triggered Behavior
When the Carrier reports that the shipment has been delivered, Order Fulfillment forwards the delivery notice to the Sales Order Service by calling the orderDelivered() operation of the Sales Order Status Interface (Figure 5-10). This is another dependency. Note that this is an In-Only operation and does not return a response.
Figure 5-10: Sales Order Status Interface—orderDelivered() Operation
Figure 5-11 shows the triggered behavior resulting from the invocation of the orderDelivered() operation. Once again, the existence of the order status update activity is inferred from information visible through other interface operations.
Figure 5-11: orderDelivered() Triggered Behavior
Observable State Information
Some component operations can reveal that information has been retained within a component. Consider the getOrder() operation of the Sales Order Service Interface shown in Figure 5-12. It returns information about the sales order and its line items. In order for the component to be able to return this information, it must retain it as part of its state. This makes that portion of the state observable.
Figure 5-12: Sales Order Service Interface—getOrder() Operation
There are two types of information typically observable through such interfaces: operational information and milestone-level status.
Figure 5-13 shows, at a conceptual level, the operational information involved in the ordering and shipping of goods. It also indicates which components are responsible for managing individual information elements. The Sales Order Service manages operational information related to the sales order, including the Sales Order, the Sales Order Line Items, the billing information, and the shipping and billing addresses. The service is stateful since it retains this information. The fact that the service makes this information (and changes to it) visible at its interfaces makes this state information observable.
Figure 5-13: Operational Information in the Order-to-Delivery Process
When a component retains stateful information, component users need to know the relationship between the interface operations and this stateful information in order to use the operations correctly. Users need to know which operations modify and reveal the state, and the details about which portions of the state are modified or revealed.
There is another kind of information often visible through service interfaces: milestone-level status. This is usually an abstracted summary of the overall solution state, which includes state information originating outside the component.
For example, the status attribute of the sales order takes on one of a number of values depending upon the overall status of the order (Figure 5-14). Much of the state information being summarized resides outside the scope of the Sales Order Service. Thus, there must be interfaces (and systems implementing or invoking those interfaces) that provide the detailed state information needed to update this summary state information. In this case, these are the orderShipped() and orderDelivered() operations of the Sales Order Status Interface that are invoked by the Order Fulfillment Service.
Figure 5-14:Order Milestone Status
Observable State and Cached Information
The Sales Order Service makes use of some state information that it does not directly manage (own): customer information, product information, and information about the related shipments. Other services (components) are the systems of record for this information, but at least some of this information is cached in the Sales Order Service. This situation can raise some interesting design challenges, challenges that when resolved, can impact the observable behavior of the Sales Order Service.
A core design challenge is deciding how to maintain consistency between the cached information as viewed through the Sales Order Service and the same information viewed through its actual system of record. If it is possible (and it almost always is) for inconsistencies to arise, then the component’s observable behaviors must indicate the scenarios under which this can arise. Users need to be aware that such inconsistencies are possible and the circumstances under which they can arise.
Let’s take a look at how such a situation can arise in the Sales Order Service. If the other systems of record have separate data stores (e.g., databases), then it must be the case that the Sales Order Service retains copies of at least some of the information in its physical data store (Figure 5-15). Since this information is a copy, inconsistencies will arise if the system of record is updated but the copy is not. The more information that is copied, the more likely it is that a discrepancy will arise and be observed.
Figure 5-15: Sales Order Service Data Store
Maintaining the accuracy of cached information requires interactions with the information’s system of record. One common approach is for the system of record to provide facilities to inform interested parties of changes to its information. The system of record provides an interface for interested parties to subscribe to such notifications, and a second interface to actually notify the parties of changes. This approach is often taken when it is likely that more than one party will be interested in the changes. Note that the uses of these interfaces constitute additional observable dependencies.
In the present design, the Sales Order Service has two relationships of this type, one for product information and the other for customer information. Figure 5-16 shows the interfaces provided by the Product Service for this purpose. The Sales Order Service is a user of these interfaces.
Figure 5-16: Product Change Notification Context
Two interesting questions arise with respect to the subscription interface. The first of these relates to the granularity of the subscription: Does the interested party subscribe to changes to particular products, or to all products? The second relates to the timing of the subscription: When does subscription occur? When the component is deployed? When it is started? Does it occur at some other time?
In practice, subscriptions are often realized without implementing subscription interfaces at all. Instead, the design uses a messaging service (e.g., JMS) for the delivery of notifications. The granularity issue is addressed through the choice of the number of destinations (topics or queues) in the design and the determination of which notifications will be sent to which destinations. Subscriptions are implemented through deployment-time configuration of components as listeners to specific destinations.
The productChange() operation raises another interesting situation from the Sales Order Service’s perspective: its invocation triggers activity in the service that is not related to an operation being provided by the Sales Order Service itself (Figure 5-17). It is the arrival of the notification that triggers the service’s activity. This is commonly referred to as an event-driven interaction. As shown, the process depicted in the diagram does not indicate what action should be taken as a result of this notification. However, if the final resolution requires either a change to the observable state of the service (e.g., replacing the item with another item) or an interaction with an external component (such as sending an e-mail notification), these actions constitute changes to the observable behavior that must be documented.
Figure 5-17: Product Change Notification Process
Avoiding Caches: Nested Retrieval
An approach to minimizing inconsistencies is to minimize the amount of cached information in a component. For example, instead of caching a lot of data, the component might only cache an identifier. Then, when the service needs more information about the identified entity, it retrieves it dynamically from the system of record for that entity.
The placeOrder() operation described earlier contains two interactions of this type. First, it interacts with the Product Service to validate the productID in the order request. Second, it interacts with the Customer Service for two reasons: to validate the customerID, and to retrieve the customerName, which is a required field for the Fill Order Request (Figure 5-6) data structure used in the fillOrder() invocation. Note that only the identifiers for these two entities are retained as part of the Sales Order Service’s state.