Dela via


Best Practices for Implementing the Event-based Asynchronous Pattern

The Event-based Asynchronous Pattern provides you with an effective way to expose asynchronous behavior in classes, with familiar event and delegate semantics. To implement Event-based Asynchronous Pattern, you need to follow some specific behavioral requirements. The following sections describe requirements and guidelines you should consider when you implement a class that follows the Event-based Asynchronous Pattern.

For an overview, see Implementing the Event-based Asynchronous Pattern.

The following list shows the best practices discussed in this topic:

  • Required Behavioral Guarantees

  • Completion

  • Completed Event and EventArgs

  • Simultaneously Executing Operations

  • Accessing Results

  • Progress Reporting

  • IsBusy Implementation

  • Cancellation

  • Errors and Exceptions

  • Threading and Contexts

  • Guidelines

Required Behavioral Guarantees

If you implement the Event-based Asynchronous Pattern, you must provide a number of guarantees to ensure that your class will behave properly and clients of your class can rely on such behavior.

Completion

Always invoke the MethodNameCompleted event handler when you have successful completion, an error, or a cancellation. Applications should never encounter a situation where they remain idle and completion never occurs. One exception to this rule is if the asynchronous operation itself it designed so that it never completes.

Completed Event and EventArgs

For each separate MethodNameAsync method, apply the following design requirements:

  • Define a MethodNameCompleted event on the same class as the method.

  • Define an EventArgs class and accompanying delegate for the MethodNameCompleted event that derives from the AsyncCompletedEventArgs class. The default class name should be of the form MethodNameCompletedEventArgs.

  • Ensure that the EventArgs class is specific to the return values of the MethodName method. When you use the EventArgs class, you should never require developers to cast the result.

    The following code example shows good and bad implementation of this design requirement respectively.

[C#]

// Good design
private void Form1_MethodNameCompleted(object sender, xxxCompletedEventArgs e) 
{ 
    DemoType result = e.Result;
}

// Bad design
private void Form1_MethodNameCompleted(object sender, MethodNameCompletedEventArgs e) 
{ 
    DemoType result = (DemoType)(e.Result);
}
  • Do not define an EventArgs class for returning methods that return void. Instead, use an instance of the AsyncCompletedEventArgs class.

  • Ensure that you always raise the MethodNameCompleted event. This event should be raised on successful completion, on an error, or on cancellation. Applications should never encounter a situation where they remain idle and completion never occurs.

  • Ensure that you catch any exceptions that occur in the asynchronous operation and assign the caught exception to the Error property.

  • If there was an error completing the task, the results should not be accessible. When the Error property is not null, ensure that accessing any property in the EventArgs structure raises an exception. Use the RaiseExceptionIfNecessary method to perform this verification.

  • Model a time out as an error. When a time out occurs, raise the MethodNameCompleted event and assign a TimeoutException to the Error property.

  • If your class supports multiple concurrent invocations, ensure that the MethodNameCompleted event contains the appropriate userSuppliedState object.

  • Ensure that the MethodNameCompleted event is raised on the appropriate thread and at the appropriate time in the application lifecycle. For more information, see the Threading and Contexts section.

Simultaneously Executing Operations

  • If your class supports multiple concurrent invocations, enable the developer to track each invocation separately by defining the MethodNameAsync overload that takes an object-valued state parameter, or task ID, called userSuppliedState. This parameter should always be the last parameter in the MethodNameAsync method's signature.

  • If your class defines the MethodNameAsync overload that takes an object-valued state parameter, or task ID, be sure to track the lifetime of the operation with that task ID, and be sure to provide it back into the completion handler. There are helper classes available to assist. For more information on concurrency management, see Walkthrough: Implementing a Component That Supports the Event-based Asynchronous Pattern.

  • If your class defines the MethodNameAsync method without the state parameter, and it does not support multiple concurrent invocations, ensure that any attempt to invoke MethodNameAsync before the prior MethodNameAsync invocation has completed raises an InvalidOperationException.

  • In general, do not raise an exception if the MethodNameAsync method without the userSuppliedState parameter is invoked multiple times so that there are multiple outstanding operations. You can raise an exception when your class explicitly cannot handle that situation, but assume that developers can handle these multiple indistinguishable callbacks

Accessing Results

Progress Reporting

  • Support progress reporting, if possible. This enables developers to provide a better application user experience when they use your class.

  • If you implement a ProgressChanged/MethodNameProgressChanged event, ensure that there are no such events raised for a particular asynchronous operation after that operation's MethodNameCompleted event has been raised.

  • If the standard ProgressChangedEventArgs is being populated, ensure that the ProgressPercentage can always be interpreted as a percentage. The percentage does not need to be accurate, but it should represent a percentage. If your progress reporting metric must be something other than a percentage, derive a class from the ProgressChangedEventArgs class and leave ProgressPercentage at 0. Avoid using a reporting metric other than a percentage.

  • Ensure that the ProgressChanged event is raised on the appropriate thread and at the appropriate time in the application lifecycle. For more information, see the Threading and Contexts section.

IsBusy Implementation

  • Do not expose an IsBusy property if your class supports multiple concurrent invocations. For example, XML Web service proxies do not expose an IsBusy property because they support multiple concurrent invocations of asynchronous methods.

  • The IsBusy property should return true after the MethodNameAsync method has been called and before the MethodNameCompleted event has been raised. Otherwise it should return false. The BackgroundWorker and WebClient components are examples of classes that expose an IsBusy property.

Cancellation

  • Support cancellation, if possible. This enables developers to provide a better application user experience when they use your class.

  • In the case of cancellation, set the Cancelled flag in the AsyncCompletedEventArgs object.

  • Ensure that any attempt to access the result raises an InvalidOperationException stating that the operation was canceled. Use the AsyncCompletedEventArgs.RaiseExceptionIfNecessary method to perform this verification.

  • Ensure that calls to a cancellation method always return successfully, and never raise an exception. In general, a client is not notified as to whether an operation is truly cancelable at any given time, and is not notified as to whether a previously issued cancellation has succeeded. However, the application will always be given notification when a cancellation succeeded, because the application takes part in the completion status.

  • Raise the MethodNameCompleted event when the operation is canceled.

Errors and Exceptions

  • Catch any exceptions that occur in the asynchronous operation and set the value of the AsyncCompletedEventArgs.Error property to that exception.

Threading and Contexts

For correct operation of your class, it is critical that the client's event handlers are invoked on the proper thread or context for the given application model, including ASP.NET and Windows Forms applications. Two important helper classes are provided to ensure that your asynchronous class behaves correctly under any application model: AsyncOperation and AsyncOperationManager.

AsyncOperationManager provides one method, CreateOperation, which returns an AsyncOperation. Your MethodNameAsync method calls CreateOperation and your class uses the returned AsyncOperation to track the lifetime of the asynchronous task.

To report progress, incremental results, and completion to the client, call the Post and OperationCompleted methods on the AsyncOperation. AsyncOperation is responsible for marshaling calls to the client's event handlers to the proper thread or context.

Note

You can circumvent these rules if you explicitly want to go against the policy of the application model, but still benefit from the other advantages of using the Event-based Asynchronous Pattern. For example, you may want a class operating in Windows Forms to be free threaded. You can create a free threaded class, as long as developers understand the implied restrictions. Console applications do not synchronize the execution of Post calls. This can cause ProgressChanged events to be raised out of order. If you wish to have serialized execution of Post calls, implement and install a System.Threading.SynchronizationContext class.

For more information about using AsyncOperation and AsyncOperationManager to enable your asynchronous operations, see Walkthrough: Implementing a Component That Supports the Event-based Asynchronous Pattern.

Guidelines

  • Ideally, each method invocation should be independent of others. You should avoid coupling invocations with shared resources. If resources are to be shared among invocations, you will need to provide a proper synchronization mechanism in your implementation.

  • Designs that require the client to implement synchronization are discouraged. For example, you could have an asynchronous method that receives a global static object as a parameter; multiple concurrent invocations of such a method could result in data corruption or deadlocks.

  • If you implement a method with the multiple-invocation overload (userState in the signature), your class will need to manage a collection of user states, or task IDs, and their corresponding pending operations. This collection should be protected with lock regions, because the various invocations add and remove userState objects in the collection.

  • Consider reusing CompletedEventArgs classes where feasible and appropriate. In this case, the naming is not consistent with the method name, because a given delegate and EventArgs type are not tied to a single method. However, forcing developers to cast the value retrieved from a property on the EventArgs is never acceptable.

  • If you are authoring a class that derives from Component, do not implement and install your own SynchronizationContext class. Application models, not components, control the SynchronizationContext that is used.

  • When you use multithreading of any sort, you potentially expose yourself to very serious and complex bugs. Before implementing any solution that uses multithreading, see Managed Threading Best Practices.

See Also

Tasks

How to: Use Components That Support the Event-based Asynchronous Pattern

Walkthrough: Implementing a Component That Supports the Event-based Asynchronous Pattern

Reference

AsyncOperation

AsyncOperationManager

AsyncCompletedEventArgs

ProgressChangedEventArgs

BackgroundWorker

Concepts

Implementing the Event-based Asynchronous Pattern

Deciding When to Implement the Event-based Asynchronous Pattern

Best Practices for Implementing the Event-based Asynchronous Pattern

Other Resources

Multithreaded Programming with the Event-based Asynchronous Pattern