Parallel Processing and Concurrency in the .NET Framework

TAP uses a single method to represent the initiation and completion of an asynchronous operation. This is in contrast to the Asynchronous Programming Model (APM orIAsyncResult) pattern, which requires Begin and End methods, and in contrast to the Event-based Asynchronous Pattern (EAP), which requires a method that has the Asyncsuffix and also requires one or more events, event handler delegate types, and EventArg-derived types. Asynchronous methods in TAP include the Async suffix after the operation name; for example, GetAsync for a get operation. If you‘re adding a TAP method to a class that already contains that method name with the Async suffix, use the suffix TaskAsync instead. For example, if the class already has a GetAsync method, use the name GetTaskAsync.

The TAP method returns either a Task or a Task<TResult>, based on whether the corresponding synchronous method returns void or a type TResult.

The parameters of a TAP method should match the parameters of its synchronous counterpart, and should be provided in the same order. However, out and ref parameters are exempt from this rule and should be avoided entirely. Any data that would have been returned through an out or ref parameter should instead be returned as part of theTResult returned by Task<TResult>, and should use a tuple or a custom data structure to accommodate multiple values. Methods that are devoted exclusively to the creation, manipulation, or combination of tasks (where the asynchronous intent of the method is clear in the method name or in the name of the type to which the method belongs) need not follow this naming pattern; such methods are often referred to as combinators. Examples of combinators include WhenAll and WhenAny, and are discussed in the Using the Built-in Task-based Combinators section of the article Consuming the Task-based Asynchronous Pattern.

For examples of how the TAP syntax differs from the syntax used in legacy asynchronous programming patterns such as the Asynchronous Programming Model (APM) and the Event-based Asynchronous Pattern (EAP), see Asynchronous Programming Patterns.

An asynchronous method that is based on TAP can do a small amount of work synchronously, such as validating arguments and initiating the asynchronous operation, before it returns the resulting task. Synchronous work should be kept to the minimum so the asynchronous method can return quickly. Reasons for a quick return include the following:

• Asynchronous methods may be invoked from user interface (UI) threads, and any long-running synchronous work could harm the responsiveness of the application.

• Multiple asynchronous methods may be launched concurrently. Therefore, any long-running work in the synchronous portion of an asynchronous method could delay the initiation of other asynchronous operations, thereby decreasing the benefits of concurrency.

In some cases, the amount of work required to complete the operation is less than the amount of work required to launch the operation asynchronously. Reading from a stream where the read operation can be satisfied by data that is already buffered in memory is an example of such a scenario. In such cases, the operation may complete synchronously, and may return a task that has already been completed.

An asynchronous method should raise an exception to be thrown out of the asynchronous method call only in response to a usage error. Usage errors should never occur in production code. For example, if passing a null reference (Nothing in Visual Basic) as one of the method’s arguments causes an error state (usually represented by anArgumentNullException exception), you can modify the calling code to ensure that a null reference is never passed. For all other errors, exceptions that occur when an asynchronous method is running should be assigned to the returned task, even if the asynchronous method happens to complete synchronously before the task is returned. Typically, a task contains at most one exception. However, if the task represents multiple operations (for example, WhenAll), multiple exceptions may be associated with a single task.

When you implement a TAP method, you can determine where asynchronous execution occurs. You may choose to execute the workload on the thread pool, implement it by using asynchronous I/O (without being bound to a thread for the majority of the operation’s execution), run it on a specific thread (such as the UI thread), or use any number of potential contexts. A TAP method may even have nothing to execute, and may just return a Task that represents the occurrence of a condition elsewhere in the system (for example, a task that represents data arriving at a queued data structure).The caller of the TAP method may block waiting for the TAP method to complete by synchronously waiting on the resulting task, or may run additional (continuation) code when the asynchronous operation completes. The creator of the continuation code has control over where that code executes. You may create the continuation code either explicitly, through methods on the Task class (for example, ContinueWith) or implicitly, by using language support built on top of continuations (for example, await in C#, Await in Visual Basic, AwaitValue in F#).

The Task class provides a life cycle for asynchronous operations, and that cycle is represented by the TaskStatus enumeration. To support corner cases of types that derive from Task and Task<TResult>, and to support the separation of construction from scheduling, the Task class exposes a Start method. Tasks that are created by the publicTask constructors are referred to as cold tasks, because they begin their life cycle in the non-scheduled Created state and are scheduled only when Start is called on these instances. All other tasks begin their life cycle in a hot state, which means that the asynchronous operations they represent have already been initiated and their task status is an enumeration value other than TaskStatus.Created. All tasks that are returned from TAP methods must be activated. If a TAP method internally uses a task’s constructor to instantiate the task to be returned, the TAP method must call Start on the Task object before returning it. Consumers of a TAP method may safely assume that the returned task is active and should not try to call Start on any Task that is returned from a TAP method. Calling Start on an active task results in an InvalidOperationException exception.

In TAP, cancellation is optional for both asynchronous method implementers and asynchronous method consumers. If an operation allows cancellation, it exposes an overload of the asynchronous method that accepts a cancellation token (CancellationToken instance). By convention, the parameter is named cancellationToken.

The asynchronous operation monitors this token for cancellation requests. If it receives a cancellation request, it may choose to honor that request and cancel the operation. If the cancellation request results in work being ended prematurely, the TAP method returns a task that ends in the Canceled state; there is no available result and no exception is thrown. The Canceled state is considered to be a final (completed) state for a task, along with the Faulted and RanToCompletion states. Therefore, if a task is in the Canceled state, its IsCompleted property returns true. When a task completes in the Canceled state, any continuations registered with the task are scheduled or executed, unless a continuation option such as NotOnCanceled was specified to opt out of continuation. Any code that is asynchronously waiting for a canceled task through use of language features continues to run but receives an OperationCanceledException or an exception derived from it. Code that is blocked synchronously waiting on the task through methods such as Wait and WaitAll also continue to run with an exception.

If a cancellation token has requested cancellation before the TAP method that accepts that token is called, the TAP method should return a Canceled task. However, if cancellation is requested while the asynchronous operation is running, the asynchronous operation need not accept the cancellation request. The returned task should end in the Canceled state only if the operation ends as a result of the cancellation request. If cancellation is requested but a result or an exception is still produced, the task should end in the RanToCompletion or Faulted state. For asynchronous methods used by a developer who wants cancellation first and foremost, you don‘t have to provide an overload that doesn’t accept a cancellation token. For methods that cannot be canceled, do not provide overloads that accept a cancellation token; this helps indicate to the caller whether the target method is actually cancelable. Consumer code that does not desire cancellation may call a method that accepts a CancellationToken and provideNone as the argument value. None is functionally equivalent to the default CancellationToken.

Some asynchronous operations benefit from providing progress notifications; these are typically used to update a user interface with information about the progress of the asynchronous operation. In TAP, progress is handled through an IProgress<T> interface, which is passed to the asynchronous method as a parameter that is usually namedprogress. Providing the progress interface when the asynchronous method is called helps eliminate race conditions that result from incorrect usage (that is, when event handlers that are incorrectly registered after the operation starts may miss updates). More importantly, the progress interface supports varying implementations of progress, as determined by the consuming code. For example, the consuming code may only care about the latest progress update, or may want to buffer all updates, or may want to invoke an action for each update, or may want to control whether the invocation is marshaled to a particular thread. All these options may be achieved by using a different implementation of the interface, customized to the particular consumer’s needs. As with cancellation, TAP implementations should provide an IProgress<T>parameter only if the API supports progress notifications. For example, if the ReadAsync method discussed earlier in this article is able to report intermediate progress in the form of the number of bytes read thus far, the progress callback could be an IProgress<T> interface: