C# – HttpClient.GetAsync(…) never returns when using await/async

async-awaitasynchronouscdotnet-httpclientnet

Edit: This question looks like it might be the same problem, but has no responses…

Edit: In test case 5 the task appears to be stuck in WaitingForActivation state.

I've encountered some odd behaviour using the System.Net.Http.HttpClient in .NET 4.5 – where "awaiting" the result of a call to (e.g.) httpClient.GetAsync(...) will never return.

This only occurs in certain circumstances when using the new async/await language functionality and Tasks API – the code always seems to work when using only continuations.

Here's some code which reproduces the problem – drop this into a new "MVC 4 WebApi project" in Visual Studio 11 to expose the following GET endpoints:

/api/test1
/api/test2
/api/test3
/api/test4
/api/test5 <--- never completes
/api/test6

Each of the endpoints here return the same data (the response headers from stackoverflow.com) except for /api/test5 which never completes.

Have I encountered a bug in the HttpClient class, or am I misusing the API in some way?

Code to reproduce:

public class BaseApiController : ApiController
{
    /// <summary>
    /// Retrieves data using continuations
    /// </summary>
    protected Task<string> Continuations_GetSomeDataAsync()
    {
        var httpClient = new HttpClient();

        var t = httpClient.GetAsync("http://stackoverflow.com", HttpCompletionOption.ResponseHeadersRead);

        return t.ContinueWith(t1 => t1.Result.Content.Headers.ToString());
    }

    /// <summary>
    /// Retrieves data using async/await
    /// </summary>
    protected async Task<string> AsyncAwait_GetSomeDataAsync()
    {
        var httpClient = new HttpClient();

        var result = await httpClient.GetAsync("http://stackoverflow.com", HttpCompletionOption.ResponseHeadersRead);

        return result.Content.Headers.ToString();
    }
}

public class Test1Controller : BaseApiController
{
    /// <summary>
    /// Handles task using Async/Await
    /// </summary>
    public async Task<string> Get()
    {
        var data = await Continuations_GetSomeDataAsync();

        return data;
    }
}

public class Test2Controller : BaseApiController
{
    /// <summary>
    /// Handles task by blocking the thread until the task completes
    /// </summary>
    public string Get()
    {
        var task = Continuations_GetSomeDataAsync();

        var data = task.GetAwaiter().GetResult();

        return data;
    }
}

public class Test3Controller : BaseApiController
{
    /// <summary>
    /// Passes the task back to the controller host
    /// </summary>
    public Task<string> Get()
    {
        return Continuations_GetSomeDataAsync();
    }
}

public class Test4Controller : BaseApiController
{
    /// <summary>
    /// Handles task using Async/Await
    /// </summary>
    public async Task<string> Get()
    {
        var data = await AsyncAwait_GetSomeDataAsync();

        return data;
    }
}

public class Test5Controller : BaseApiController
{
    /// <summary>
    /// Handles task by blocking the thread until the task completes
    /// </summary>
    public string Get()
    {
        var task = AsyncAwait_GetSomeDataAsync();

        var data = task.GetAwaiter().GetResult();

        return data;
    }
}

public class Test6Controller : BaseApiController
{
    /// <summary>
    /// Passes the task back to the controller host
    /// </summary>
    public Task<string> Get()
    {
        return AsyncAwait_GetSomeDataAsync();
    }
}

Best Answer

You are misusing the API.

Here's the situation: in ASP.NET, only one thread can handle a request at a time. You can do some parallel processing if necessary (borrowing additional threads from the thread pool), but only one thread would have the request context (the additional threads do not have the request context).

This is managed by the ASP.NET SynchronizationContext.

By default, when you await a Task, the method resumes on a captured SynchronizationContext (or a captured TaskScheduler, if there is no SynchronizationContext). Normally, this is just what you want: an asynchronous controller action will await something, and when it resumes, it resumes with the request context.

So, here's why test5 fails:

Test5Controller.Get executes AsyncAwait_GetSomeDataAsync (within the ASP.NET request context).
AsyncAwait_GetSomeDataAsync executes HttpClient.GetAsync (within the ASP.NET request context).
The HTTP request is sent out, and HttpClient.GetAsync returns an uncompleted Task.
AsyncAwait_GetSomeDataAsync awaits the Task; since it is not complete, AsyncAwait_GetSomeDataAsync returns an uncompleted Task.
Test5Controller.Get blocks the current thread until that Task completes.
The HTTP response comes in, and the Task returned by HttpClient.GetAsync is completed.
AsyncAwait_GetSomeDataAsync attempts to resume within the ASP.NET request context. However, there is already a thread in that context: the thread blocked in Test5Controller.Get.
Deadlock.

Here's why the other ones work:

(test1, test2, and test3): Continuations_GetSomeDataAsync schedules the continuation to the thread pool, outside the ASP.NET request context. This allows the Task returned by Continuations_GetSomeDataAsync to complete without having to re-enter the request context.
(test4 and test6): Since the Task is awaited, the ASP.NET request thread is not blocked. This allows AsyncAwait_GetSomeDataAsync to use the ASP.NET request context when it is ready to continue.

And here's the best practices:

In your "library" async methods, use ConfigureAwait(false) whenever possible. In your case, this would change AsyncAwait_GetSomeDataAsync to be var result = await httpClient.GetAsync("http://stackoverflow.com", HttpCompletionOption.ResponseHeadersRead).ConfigureAwait(false);
Don't block on Tasks; it's async all the way down. In other words, use await instead of GetResult (Task.Result and Task.Wait should also be replaced with await).

That way, you get both benefits: the continuation (the remainder of the AsyncAwait_GetSomeDataAsync method) is run on a basic thread pool thread that doesn't have to enter the ASP.NET request context; and the controller itself is async (which doesn't block a request thread).

More information:

My async/await intro post, which includes a brief description of how Task awaiters use SynchronizationContext.
The Async/Await FAQ, which goes into more detail on the contexts. Also see Await, and UI, and deadlocks! Oh, my! which does apply here even though you're in ASP.NET rather than a UI, because the ASP.NET SynchronizationContext restricts the request context to just one thread at a time.
This MSDN forum post.
Stephen Toub demos this deadlock (using a UI), and so does Lucian Wischik.

Update 2012-07-13: Incorporated this answer into a blog post.

Related Solutions

C# – Async call with await in HttpClient never returns

Check out this answer to my question which seems to be very similar.

Something to try: call ConfigureAwait(false) on the Task returned by GetStreamAsync(). E.g.

var result = await httpClient.GetStreamAsync("weeklyplan")
                             .ConfigureAwait(continueOnCapturedContext:false);

Whether or not this is useful depends on how your code above is being called - in my case calling the async method using Task.GetAwaiter().GetResult() caused the code to hang.

This is because GetResult() blocks the current thread until the Task completes. When the task does complete it attempts to re-enter the thread context in which it was started but cannot because there is already a thread in that context, which is blocked by the call to GetResult()... deadlock!

This MSDN post goes into a bit of detail on how .NET synchronizes parallel threads - and the answer given to my own question gives some best practices.

C# – async/await – when to return a Task vs void

Normally, you would want to return a Task. The main exception should be when you need to have a void return type (for events). If there's no reason to disallow having the caller await your task, why disallow it?
async methods that return void are special in another aspect: they represent top-level async operations, and have additional rules that come into play when your task returns an exception. The easiest way is to show the difference is with an example:

static async void f()
{
    await h();
}

static async Task g()
{
    await h();
}

static async Task h()
{
    throw new NotImplementedException();
}

private void button1_Click(object sender, EventArgs e)
{
    f();
}

private void button2_Click(object sender, EventArgs e)
{
    g();
}

private void button3_Click(object sender, EventArgs e)
{
    GC.Collect();
}

f's exception is always "observed". An exception that leaves a top-level asynchronous method is simply treated like any other unhandled exception. g's exception is never observed. When the garbage collector comes to clean up the task, it sees that the task resulted in an exception, and nobody handled the exception. When that happens, the TaskScheduler.UnobservedTaskException handler runs. You should never let this happen. To use your example,

public static async void AsyncMethod2(int num)
{
    await Task.Factory.StartNew(() => Thread.Sleep(num));
}

Yes, use async and await here, they make sure your method still works correctly if an exception is thrown.

For more information see: https://docs.microsoft.com/en-us/archive/msdn-magazine/2013/march/async-await-best-practices-in-asynchronous-programming

Have I encountered a bug in the HttpClient class, or am I misusing the API in some way?

Best Answer

Related Solutions

C# – Async call with await in HttpClient never returns

C# – async/await – when to return a Task vs void

Related Topic