Windows Workflow Foundation Engine Architecture
Given the breadth of scenarios requiring workflow and the key goal of providing a singular technology layer to support all these scenarios, the workflow technology is a well-factored framework with numerous pluggable interfaces and extensibility as a core design principle. The architecture for Windows Workflow Foundation is depicted in Figure 1.1.
Figure 1.1 Windows Workflow Foundation engine architecture.
At the bottom of Figure 1.1 is the host process. Windows Workflow Foundation has no inherent execution process. Instead, Windows Workflow Foundation is an in-process engine that runs inside a host process. The host process is responsible for providing a set of services to Windows Workflow Foundation. A wide variety of host processes are available on the Windows platform including console applications, WinForms applications, web applications, web services applications, SharePoint Server, and NT Service applications. Effectively, any executable process can host the Windows Workflow Foundation runtime engine.
This also presents some interesting challenges because the capabilities of each host are often different to another host. SharePoint is a dramatically different environment than a console application. For this reason the Windows Workflow Foundation architecture hosting layer provides a set of pluggable interfaces from Windows Workflow Foundation to the host.
Sitting on top of the hosting layer in the Windows Workflow Foundation architecture is the runtime layer. The runtime layer is the core of the engine providing both workflow execution and workflow lifecycle management capabilities.
Finally, the workflow model layer is where most developers will interact with Windows Workflow Foundation. The workflow model layer includes the various workflow models, APIs, and the activities. The following sections provide more details on the hosting, runtime, and workflow model layers.
The hosting layer provides interfaces between Windows Workflow Foundation and a particular host for the following key services: Communication, Persistence, Tracking, Timer, Threading, and Transaction. The implementations of the former three services that ship with Windows Workflow Foundation are durable while the latter two services are stateless. However, none of the services are necessarily durable if you write your own. By abstracting each of these services Windows Workflow Foundation can take advantage of specific capabilities available in specific hosts. The following sections describe the functions performed by each of these services:
Persistence: Although some workflows may execute for a short period of time, workflow is inherently asynchronous and a particular workflow, such as the Ph.D. thesis approval process, may take many days or months. A workflow engine that retained its state in memory for that period would not scale as each instance of the workflow would consume memory for the duration and eventually the system memory would be exhausted. Instead, a persistent architecture is used where workflow is executed in memory, and should it be required, the workflow state will persist to a store while it waits for a response that might take some time such as the "Ph.D. approved" step in the workflow. Each host application has a specific set of persistence requirements. For example, to persist state, ASP.NET uses a set of Session State objects that has state client providers for in-memory persistence and SQL Server persistence. In contrast, SharePoint persists state in a SharePoint-specific set of tables in SQL Server and your console application may choose to persist state to the file system as XML. With such a large variety of host-specific persistence capabilities, it would not be sensible for a broadly applicable technology such as Windows Workflow Foundation to specify a single persistence provider. The Windows Workflow Foundation hosting layer persistence interface enables Windows Workflow Foundation to work across the full gamut of host persistence architectures.
Timer: Workflows often need to wait for an event to continue. The timer is the supplied clock that is used to manage these delays. For example, an approval workflow may delay and unload from memory until a particular approver completes the necessary approval. The timer implementation in this case might be a durable timer that survives a potential system restart while waiting for approval.
Tracking: A key reason to implement workflow is because the workflow model provides a greater degree of system transparency at runtime than amorphous code. Indeed, all workflows are automatically instrumented without any programming. The tracking instrumentation is consistent across both the event content and the tracking interface. Depending on the host, the target tracking infrastructure is often different. For example, a LOB application often persists workflow tracking information within the LOB database whereas a console application may persist tracking information to an XML file. The tracking interface receives tracking events from the Windows Workflow Foundation runtime and passes them on to the host application.
Communications: Workflows send and receive events or messages from their host. These events trigger workflows, and move the workflow to the next step in the overall flow. There are a wide variety of communications infrastructures available on the Windows platform including web services, .NET calls, loosely coupled messaging, and so on. For this reason, Windows Workflow Foundation does not provide its own proprietary communications layer. Instead, Windows Workflow Foundation provides pluggable interfaces that can be implemented for any communications layer. Of course, there are easy-to-use, prebuilt communication interfaces to and from common targets such as web services and for passing data objects in and out of a workflow—perhaps from a form.
The physical job of development of these interfaces for specific hosts is relatively challenging compared to other aspects of workflow development described shortly. For this reason, ISVs will typically build host layer providers into their host applications so that end-user developers can simply reuse these services. In addition, Windows Workflow Foundation ships prebuilt support for ASP.NET 2.0 and the interfaces shown in Table 1.1.
Table 1.1 Prebuilt Host Layer Service Implementations
SQL Server state persistence
Both an in-memory and SQL Server–based timer
.NET thread pool ASP.NET thread pool
Server Tracking Persistence and Event Log recording for termination
.NET components and web services
Sitting on top of the hosting layer is the runtime layer.
The runtime layer is core to Windows Workflow Foundation. In direct comparison to the other layers in the architecture, the runtime layer is not pluggable as it contains the mission-critical services required for workflow. These services include the following:
Execution: The execution service schedules activities and supports common behaviors such as event handling, exceptions, tracking, and transactions.
Tracking: The tracking service creates the tracking events that are serialized through the tracking interface.
State management: The state management service manages states that may be persisted through the persistence interface.
Scheduler: The scheduler service schedules execution of activities.
Rules: The rules service provides policy execution functionality and CodeDOM condition evaluation.
Workflow Model Layer
The workflow model layer is where most application developers will spend the majority of their time writing code for Windows Workflow Foundation. This layer includes support for multiple workflow model types, activities, and the main programming APIs use by most developers.
Windows Workflow Foundation supports two models out of the box:
Sequential workflow model: This model is primarily a structured workflow where a step in the workflow leads to another step in a manner that may be determined often at design-time and can be represented in a flow diagram such as that shown in Figure 1.2.
State machine workflow model: This model uses the paradigm of states and transitions between states to represent the workflow. There is no deterministic path between the steps from a design perspective because the workflow does not execute in a sequential nature. Rather, the workflow is a set of events that are handled in a highly variable order with one event completing and triggering a state that another event may itself trigger from. The state machine model can literally jump from any stage in the workflow to any other stages, and often will do so multiple times before reaching a completed state. The order workflow is a state machine where various messages are received and trigger the workflow to progress to a particular state. Each iteration of this workflow may result in a different path through the model as depicted in Figure 1.3.
Sequential workflows are most often used to represent structured workflows such as system-to-system workflow. These transformational workflows are self-driven once they are initiated, have a highly predictable path through the events, and are often literally sequential in nature.
Figure 1.2 Sequential workflow.
Figure 1.3 State machine workflow model.
State Machine Workflows are an effective way of representing highly people-centric workflows where the workflow thread of execution is not easily represented in a flow. This workflow model is also very useful for scenarios where a high priority event must be processed even though work is already in process or when a large number of events may occur at a particular stage in the workflow. A perfect example is a stage in the order workflow where an "order cancelled," "order updated," and "order completed" event that may be received at any time and should immediately cancel the entire process.
Although Windows Workflow includes these two models, customers can inherit from them to create their own specific models or create new models.
Regardless of model and sequencing behavior the basic element of execution and reuse is called an activity. An example of an activity is "send goods" in the previous sequential workflow example.
There are two types of activities—the simple activity and the composite activities. What makes Windows Workflow Foundation very different from traditional workflow engines is that the engine has no fixed underlying language or grammar. Instead the engine chains a set of activities that are supplied by Microsoft and created by you the developer. Microsoft-supplied constructs include "If," "Code" blocks, activities for web services, and many more. In addition, you can create your own control flow activities such as "do until" but more likely you will be creating higher level activities such as the "Receive Order" activity described previously. By using an activity execution methodology rather than a language, Windows Workflow Foundation is able to support a broad range of scenarios and you can reuse your activities across multiple workflows. Indeed the flow and the state workflow models share a majority of activities with some specific inclusions and exclusions to each model. Activities become the unit of encapsulation in much the same way that ActiveX controls were for Visual Basic 6 applications. It is expected that customers and partners will share activities in the community and generate business from activity creation.
The set of flow control activities that ship with Windows Workflow Foundation include the following:
Control flow activities: Sequence, Parallel, While, IfElse, Listen, EventDriven, ConditionedActivityGroup, Replicator, Delay
Transaction and exception activities: ExceptionHandler, Throw, Compensate, Suspend, and Terminate
Data/form-centric activities: UpdateData, SelectData, WaitForData, WaitForQuery
Communication activities: InvokeWebService, WebServiceReceive, WebServiceResponse, InvokeMethod, and EventSink
The code activity: Code
There are three additional activities that are specific for the state machine workflow model: StateInitalization, State, and SetState.
Each activity is derived from the Activity base class and includes the code that will execute when the activity and a set of design-time properties for use in the designer is called. Later chapters in this book will go into detail on each of these activities; however, a few activities are worth mentioning here. The data- and form-centric activities enable you to bind data from forms and easily surface that information into a workflow; the web services activities give you the capability to consume and expose web services; more advanced activities such as the conditioned activity group enable policy- or rules-based condition evaluation.
Over time, many activities will become available through the broader activity ecosystem through blogs and shared source as well as through Microsoft partners.
Now that we have addressed activities in detail it is important to point out that the model itself is nothing more than a "root" level activity in the Windows Workflow Foundation infrastructure.
One of the significant features of Windows Workflow Foundation is that it offers you the ability to dynamically create workflow at runtime and dynamically update an existing workflow. This is discussed in more detail later in this book.
Now that you have completed your tour of the Windows Workflow Foundation architecture, it’s time to examine the coding and design-time support for you to interact with the engine.
Typically, workflow technologies have provided graphical designers that give users the capability to drag and drop shapes to build a workflow. Some tools are targeted at business users while other tools are targeted at developers. Windows Workflow Foundation provides a set of graphical tools inside Visual Studio .NET targeted at developers. The goal of the Visual Studio .NET design experience is to be as familiar as possible to existing .NET developers writing C# and VB.NET applications and building Longhorn applications using Windows Presentation Foundation and Windows Communication Foundation.
After Windows Workflow Foundation is installed, an additional category appears on the Visual Studio .NET File New dialog box called "Workflow," as shown in Figure 1.4.
Most interesting in this introductory chapter are the Sequential Workflow Library and State Machine Workflow Library templates. The Workflow Activity Library is a template for creating custom activities and creates a project with an activity-specific designer, making it easier for you to create activities. The workflow console application and state machine console application are simple host applications for Workflow, and finally, the empty workflow project is an unconfigured workflow project.
When the Sequential Workflow template is chosen, a new project is created and System.Workflow, System.WorkflowActivities, and System.WorkflowComponentModel are all added to the project references. In addition workflow1.cs and workflow1.designer.cs are added to the project. By default, selecting workflow1.designer.cs with the mouse will graphically render the workflow directly inside the Visual Studio designer as shown in Figure 1.5.
Figure 1.4 File New Visual Basic workflow project.
Figure 1.5 Visual Studio .NET designer for sequential workflow.
The workflow is described by dragging and dropping shapes from the toolbox. Just like traditional WinForms or ASP.NET development, when a shape has code associated with it, clicking on the shape will open the code-behind where custom code may be entered directly in VB or C#. Any changes to the code-behind file that are relevant to display in the workflow are automatically reflected back in the designer.
In addition you can add a new item to the project that is a workflow with xoml. The xoml file is actually an XML file containing an XML definition of the workflow. That XML is fully accessible to the developer by choosing File, Open With and selecting the XML editor inside Visual Studio .NET. This XML and Visual representation is consistent with the Windows Presentation Framework approach in WinFX as shown in Figure 1.5.
The state machine workflow model is depicted in Figure 1.6.
Figure 1.6 Visual Studio .NET designer for state machine workflow.
This simple state machine workflow looks similar to a sequential workflow. However, in this case each box represents a state and in most examples the lines in the diagram will not just go in a forward direction as depicted but also backwards—for example an event from the orderprocessedstate may send the state machine back to the orderopenstate. Also notice the state machine–specific activities, such as StateInitialization, in the activity toolbox.
Visual Studio .NET provides "smart tags" to alert the user that properties on specific activities are not properly configured. When the workflow is complete the developer builds the project and the two partial classes that represent the code-beside and the workflow model are compiled and stored in a managed assembly.
Windows Workflow Foundation provides complete support for the familiar F5 debugging experience. Breakpoints can be set on the models at design-time using the mouse or F9 key and when a break-point is triggered, a complete call-stack is available and the debugger allows stepping into and out of the code for the workflow in debug mode.
There is no business user design surface primarily because these surfaces tend to be scenario-specific and Windows Workflow Foundation needs to support a variety of scenarios. The good news is that you can build your own business user design surface on top of Windows Workflow Foundation. The workflow designer can be completely rehosted outside of Visual Studio .NET in your custom application. Further, every aspect of the designer can be reskinned so that the look and feel of the designer control, including the size, color, and geometry of the shapes, matches your application style. By rehosting the designer and providing your customers with specialized activity packages, you can give your end-users the ability to create and modify workflow. If rehosting the designer does not provide enough flexibility then ISVs and developers can create the XML representation of the workflow from a custom tool directly, or better still, build an activity tree in code.
A sample XML representation of a workflow is depicted here:
<?Mapping XmlNamespace="Activities" ClrNamespace="System.Workflow.Activities" Assembly="System.Workflow.Activities" ?> <SequentialWorkflow x:Class="MyWorkflow" xmlns="Activities" xmlns:x="Definition"> ... </SequentialWorkflow>
There is another option for generating workflow that is a significant innovation in Windows Workflow Foundation. Unlike many workflow technologies, Windows Workflow Foundation supports a complete coding experience enabling you to create and modify workflows in code. You can choose to use a tool as simple as Notepad and the command-line compiler wfc.exe to author workflow. Workflow is simply a class. A sample sequential workflow that executes a single activity called SendEmail is shown here:
Public Class Workflow1 Inherits SequentialWorkflow Public Sub New() MyBase.New() InitializeComponent() End Sub Private Sub InitializeComponent() Me.s = New SendEmail Me.s.MailFrom = "Paul" Me.s.MailTo = "James" AddHandler Me.s.OnBeforeSend, AddressOf Me.OnBeforeSend Me.Activities.Add(Me.s) Me.DynamicUpdateCondition = Nothing Me.ID = "Workflow1" End Sub Private WithEvents s As SendEmail Private Sub OnBeforeSend(ByVal sender As System.Object, ByVal e As _ System.EventArgs) Me.s.MailSubject = "Hello at " & System.DateTime.Now.ToString() End Sub End Class
There is a designer for activities inside Visual Studio .NET. However, just like workflows, activities are also simply classes derived from the Activity base class and can be completely created in code. The SendEmail activity skeleton fragment is shown here:
<ToolboxItem (GetType(ActivityToolboxItem))> _ Partial Public Class SendEmail Inherits System.Workflow.ComponentModel.Activity Public Sub New() MyBase.New() InitializeComponent() End Sub Public Shared MailToProperty As DependencyProperty = _ DependencyProperty.Register("MailTo", GetType(System.String), _ GetType(SendEmail)) Public Shared MailFromProperty As DependencyProperty = _ DependencyProperty.Register("MailFrom", GetType(System.String), _ GetType(SendEmail)) Public Shared MailSubjectProperty As DependencyProperty = _ DependencyProperty.Register("MailSubject", GetType(System.String), _ GetType(SendEmail)) <DesignerSerializationVisibility(DesignerSerializationVisibility.Visible)> _ <ValidationVisibility(ValidationVisibility.Optional)> _ <Browsable(True)> _ Public Property MailTo() As System.String Get Return CType(MyBase.GetValue(SendEmail.MailToProperty), String) End Get Set(ByVal value As System.String) MyBase.SetValue(SendEmail.MailToProperty, value) End Set End Property <DesignerSerializationVisibility (DesignerSerializationVisibility.Visible)> _ <ValidationVisibility (ValidationVisibility.Optional)> _ <Browsable (True)> _ Public Property MailFrom() As System.String Get Return CType(MyBase.GetValue(SendEmail.MailFromProperty), String) End Get Set(ByVal value As System.String) MyBase.SetValue(SendEmail.MailFromProperty, value) End Set End Property <DesignerSerializationVisibility(DesignerSerializationVisibility.Visible)> _ <ValidationVisibility(ValidationVisibility.Optional)> _ <Browsable(True)> _ Public Property MailSubject() As System.String Get Return CType(MyBase.GetValue(SendEmail.MailSubjectProperty), String) End Get Set(ByVal value As System.String) MyBase.SetValue(SendEmail.MailSubjectProperty, value) End Set End Property Public Event OnBeforeSend As eventhandler End Class
In addition to design-time API workflow generation, dynamic update makes it possible to update running workflows on the fly. The dynamic update functionality opens the door to many interesting scenarios that were not previously possible with traditional workflow engines such as examining the state of the model at runtime and making behavioral changes as a result of variables within the flow.
This section has just touched the surface of activities; you will learn much more about activities later in this book.
In summary, Windows Workflow Foundation provides a complete authoring experience inside Visual Studio .NET with custom designers and debugging support but also provides the flexibility for you to create workflow using code directly or through XML. Now take a quick look at how a sample application uses Windows Workflow Foundation.
Office 12 Workflow
Office is the most popular desktop productivity program on the planet. One of the key feature requests from Office customers is the capability to collaborate on documents using workflow. Office 12 uses the Windows Workflow Foundation engine embedded inside the SharePoint host for workflow. The office client applications, such as Word, can kick off workflow through web services integration with Office. That workflow is executed on SharePoint and as a result a document may be sent through email to a user in Outlook to perform an action. Once the action is performed the workflow continues.
Office also provides a simple design experience in FrontPage and for more complex design, Office has its own specific activity package containing more than 30 custom activities and customers can use this directly in Visual Studio .NET. In addition, the InfoPath designer in Office 12 supports data binding form elements to a workflow. Several prebuilt workflows ship with Office 12 including Review, Approval, and Document Expiration and Office 12 supports custom workflows. Figure 1.7 shows the Review workflow in Microsoft Word.
Figure 1.7 Microsoft Word showing a Review workflow.
Developers looking to create people-based workflows should first look into customizing Windows Workflow Foundation in the context of the SharePoint host as this strategy will get you access to the Office applications immediately. There is no doubt workflow is a key feature of Office 12 and the Windows Workflow Foundation technology will be used immediately by many information workers around the world.
You have read about the Windows Workflow architecture, design tools, and a brief description of the Office 12 functionality; however, you might be wondering when and why you should use Workflow. Now that you have briefly learned about the capabilities that Windows Workflow Foundation provides, the final section of this chapter addresses the core tenets of workflow. You can think of these as the call to action because these four tenets address key aspects of application design that are fulfilled by workflow.
Call to Action: Core Workflow Tenets
Before we get to the specific tenets, it is appropriate to revise the tenets for service-oriented architecture because often workflow will be coupled in application architecture with web services. As part of the next generation web services effort, several key tenets have been identified. Although these are described in detail elsewhere on the web they have been summarized as follows:
Boundaries are explicit: When compared to objects or remote objects, services have clear boundaries. Calling services is relatively expensive compared to calling code within a service. Choosing where these boundaries exist is an interesting challenge that leads to the second tenet.
Services are autonomous: The key point here is that services, unlike objects, can execute autonomously from other services. The key word here is can because of course in many cases one service will call another.
Share schema and contract and not class: In short, the notion here is that loose coupling is good and sharing a schema between services enables a greater degree of loose coupling than sharing explicit class types.
Service compatibility is based on policy: This tenet emphasizes the separation of semantic and structural concerns. Structural aspects of the services are stored in policy files and service compatibility is driven from these, in some cases without modifying the specific business logic of the service.
These tenets focus on boundaries between components, and the point-to-point connections between these components needed to build composite applications. These four tenets in themselves are insufficient for a service-oriented architecture where distributed services are composed into composite applications. What is missing? The key missing pieces are guidance on how to compose these services together, the characteristics of such a system, and the need to involve people as a part of composite application design. To that end, the four core tenets that follow assume the initial set of tenets and expand on them to provide guidance for composite application development:
Workflows are long-running and stateful: Fundamentally, systems are created by the composition of multiple services and, following good design practices, components that perform specific roles should be created as autonomous services. The behavior between these services can be as simple as passing and mapping behaviors or more complex such as sharing transactional semantics and temporal constraints. The temporal aspects of service composition enforce asynchronous requirements on the system. A service that submits a purchase order to other services as part of a composite application waits for two hours for an acknowledgement. This asynchronous controller, or model, should be independently factored as a specialized service component. Hand in hand with the requirement to model asynchrony is the need to include state as part of the model. The service that sent out a purchase order and received an invoice requires information on the purchase order to later update another service. State may be relevant to a single call, relevant across multiple calls, or relevant across an entire asynchronous workflow.
Workflows are transparent and dynamic through their lifecycle: Services and the applications to which they compose should not be, as they are today, like concrete bunkers with no windows. Although policy provides rigor in the definition of how to talk to the service, what the service does—its behavior—is opaque beyond its method call syntax. System behavior should be transparent, enabling the developer to rapidly ascertain the behavior at design-time and make a change in that behavior. Even more importantly, system behavior should be transparent at the runtime level. If the behavior of each service in a composite application were transparent the advantages would be significant. No longer is the service a concrete bunker; rather, it is more analogous to a greenhouse. Troubleshooting, tracking, and understanding the overall composite application behavior becomes dramatically simpler because of the additional visibility into the thread of execution. Entirely new scenarios can be envisioned where the behavior of the system thus far executed can be queried at runtime and used to influence of the future behavior. With access to the system behavior metadata, the thread of execution can analyze the currently completed behavior and make changes to its behavior on the fly on the basis of new program conditions. Having asserted that system behavior should be transparent by default, there are times when the developer will want to decrease the level of transparency. For example, some service behavior may need to be obstructed for intellectual property reasons, or specific service behavior is visible within a bounded set of services but not beyond that. The dial that sets the service transparency should be set through policy and access control rather than the traditional development approach where a transparent system is not a default and developers add small windows to prebuilt concrete bunkers.
Workflows coordinate work performed by people and by software: The original tenets for services make no reference to people and yet people are a vital part of any composite application. Today services typically send messages to people through email and pagers, or receive inputs from people through user interfaces that pass their parameters to and from services. Sometimes a single person provides the information required for the composite application to continue but more often than not an entirely out-of-band interaction occurs between multiple people in a manner that has been historically challenging to model in software before a result is returned to the composite application to continue. People are important; optimizing their behavior as part of the overall system behavior may provide as high a return as optimizing the system behavior itself with some systems having 80% of their cost in exceptions managed by people. The downside of including people from a software engineering perspective is that modeling their behavior is significantly less straightforward than modeling service behavior. People tend to work in a more ad hoc manner—it is typically more challenging to be imperative with people, and the flow of information between people may not be easily drawn in a flow-based sense because they make choices that may change the workflow at runtime, but is more ad hoc and is better represented as a set of interacting exception conditions.
Workflows are based on extensible models: Every workflow comprises some number of steps, each of which performs a particular part of a complete business process. The set of actions that are used to construct a workflow can be thought of as comprising a model for that particular problem domain. Different problem domains have different actions, and so a single model isn’t appropriate for all workflows. Instead, different groups of actions—different models—can be created for specific domains. Those actions can then be used to construct workflows supporting various business processes in that domain.
With these tenets you can evaluate when to use workflow in your application. There are many, many applications that could benefit from the use of workflow, yet the lack of availability of a workflow model is highly consistent with the typical programming paradigm. With Windows Workflow Foundation for the first time you get to take advantage of this opportunity; moving forward, it will change the way you think about application development.