Does calling asynchronous Task based WCF method ut

问题:

I have the following WCF contract:

[ServiceContract(Namespace = "http://abc/Services/AdminService")]
public interface IAdminService
{
    [OperationContract]
    string GetServiceVersion();

    // More methods here
}

The GetServiceVersion is a simple method returning some string. It is used as a ping to check whether the service is reachable.

Now I would like to call it asynchronously, thinking it would be more efficient than using .NET threads to call it in background.

So, I have come up with the following interface just for that purpose:

[ServiceContract(Namespace = "http://abc/Services/AdminService")]
public interface IMiniAdminService
{
    [OperationContract(Action = "http://abc/Services/AdminService/IAdminService/GetServiceVersion", ReplyAction = "http://abc/Services/AdminService/IAdminService/GetServiceVersionResponse")]
    Task<string> GetServiceVersionAsync();
}

This makes it possible to invoke the GetServiceVersion API asynchronously:

var tmp = new ChannelFactory<IAdminService>("AdminServiceClientEndpoint");
var channelFactory = new ChannelFactory<IMiniAdminService>(tmp.Endpoint.Binding, tmp.Endpoint.Address);
var miniAdminService = channelFactory.CreateChannel();
return miniAdminService.GetServiceVersionAsync().ContinueWith(t =>
{
    if (t.Exception != null)
    {
        // The Admin Service seems to be unavailable
    }
    else
    {
        // The Admin Service is available
    }
});

The code works.

My question is this - does it utilize the IOCP to invoke the continuation?

In general, is there a way to know whether a continuation is invoked through IOCP or not (in the debugger, if needed) ?

P.S.

Here is the stack trace of my async WCF method continuation:

>   *** My Code *** Line 195    C#
    mscorlib.dll!System.Threading.Tasks.ContinuationTaskFromResultTask<string>.InnerInvoke() + 0x111 bytes  
    mscorlib.dll!System.Threading.Tasks.Task.Execute() + 0x69 bytes 
    mscorlib.dll!System.Threading.Tasks.Task.ExecutionContextCallback(object obj) + 0x4f bytes  
    mscorlib.dll!System.Threading.ExecutionContext.RunInternal(System.Threading.ExecutionContext executionContext, System.Threading.ContextCallback callback, object state, bool preserveSyncCtx) + 0x28d bytes 
    mscorlib.dll!System.Threading.ExecutionContext.Run(System.Threading.ExecutionContext executionContext, System.Threading.ContextCallback callback, object state, bool preserveSyncCtx) + 0x47 bytes  
    mscorlib.dll!System.Threading.Tasks.Task.ExecuteWithThreadLocal(ref System.Threading.Tasks.Task currentTaskSlot) + 0x3b5 bytes  
    mscorlib.dll!System.Threading.Tasks.Task.ExecuteEntry(bool bPreventDoubleExecution) + 0x104 bytes   
    mscorlib.dll!System.Threading.Tasks.Task.System.Threading.IThreadPoolWorkItem.ExecuteWorkItem() + 0x2a bytes    
    mscorlib.dll!System.Threading.ThreadPoolWorkQueue.Dispatch() + 0x249 bytes  
    mscorlib.dll!System.Threading._ThreadPoolWaitCallback.PerformWaitCallback() + 0x1e bytes    
    [Native to Managed Transition]  

Now, this stack trace looks very similar to the one I get for a method called from Task.Factory.StartNew, which is indeed Thread Pool based:

>   *** My Code *** Line 35 C#
    mscorlib.dll!System.Threading.Tasks.Task<int>.InnerInvoke() + 0x59 bytes    
    mscorlib.dll!System.Threading.Tasks.Task.Execute() + 0x60 bytes 
    mscorlib.dll!System.Threading.Tasks.Task.ExecutionContextCallback(object obj) + 0x37 bytes  
    mscorlib.dll!System.Threading.ExecutionContext.RunInternal(System.Threading.ExecutionContext executionContext, System.Threading.ContextCallback callback, object state, bool preserveSyncCtx) + 0x1a2 bytes 
    mscorlib.dll!System.Threading.ExecutionContext.Run(System.Threading.ExecutionContext executionContext, System.Threading.ContextCallback callback, object state, bool preserveSyncCtx) + 0x33 bytes  
    mscorlib.dll!System.Threading.Tasks.Task.ExecuteWithThreadLocal(ref System.Threading.Tasks.Task currentTaskSlot) + 0x2ff bytes  
    mscorlib.dll!System.Threading.Tasks.Task.ExecuteEntry(bool bPreventDoubleExecution) + 0xd3 bytes    
    mscorlib.dll!System.Threading.Tasks.Task.System.Threading.IThreadPoolWorkItem.ExecuteWorkItem() + 0x22 bytes    
    mscorlib.dll!System.Threading.ThreadPoolWorkQueue.Dispatch() + 0x22e bytes  
    mscorlib.dll!System.Threading._ThreadPoolWaitCallback.PerformWaitCallback() + 0x18 bytes    
    [Native to Managed Transition]  

回答1:

First, you'd need to add TaskContinuationOptions.ExecuteSynchronously, to make sure the continuation callback is called on the same thread the async IO operation has been finalized:

return miniAdminService.GetServiceVersionAsync().ContinueWith(t =>
{
    if (t.Exception != null)
    {
        // The Admin Service seems to be unavailable
    }
    else
    {
        // The Admin Service is available
    }
}, TaskContinuationOptions.ExecuteSynchronously);

Apparently, there is no API in .NET to tell if the thread is a IOCP pool thread. You can only tell if the thread is a thread pool thread (Thread.CurrentThread.IsThreadPoolThread), which is true for IOCP threads too.

In Win32, an IOCP thread pool is created with CreateIoCompletionPort API, but I couldn't find a Win32 API to check if the thread belongs to such pool, either.

So, here is a bit contrived example to check this theory in practice, using HtppClient as the test vehicle. First, we make sure all non-IOCP threads have populated the ThreadStatic variable s_mark with -1. Then we initiate an IO-bound operation and check s_mark on thread where the IO-bound operation gets completed:

using System;
using System.Net.Http;
using System.Threading;
using System.Threading.Tasks;

namespace ConsoleApplication_22465346
{
    public class Program
    {
        [ThreadStatic]
        static volatile int s_mark;

        // Main
        public static void Main(string[] args)
        {
            const int THREADS = 50;

            // init the thread pool
            ThreadPool.SetMaxThreads(
                workerThreads: THREADS, completionPortThreads: THREADS);
            ThreadPool.SetMinThreads(
                workerThreads: THREADS, completionPortThreads: THREADS);

            // populate s_max for non-IOCP threads
            for (int i = 0; i < THREADS; i++)
            {
                ThreadPool.QueueUserWorkItem(_ =>
                { 
                    s_mark = -1;
                    Thread.Sleep(1000);
                });
            }
            Thread.Sleep(2000);

            // non-IOCP test
            Task.Run(() =>
            {
                // by now all non-IOCP threads have s_mark == -1
                Console.WriteLine("Task.Run, s_mark: " + s_mark);
                Console.WriteLine("IsThreadPoolThread: " + Thread.CurrentThread.IsThreadPoolThread);
            }).Wait();

            // IOCP test
            var httpClient = new HttpClient();
            httpClient.GetStringAsync("http://example.com").ContinueWith(t =>
            {
                // all IOCP threads have s_mark == 0
                Console.WriteLine("GetStringAsync.ContinueWith, s_mark: " + s_mark);
                Console.WriteLine("IsThreadPoolThread: " + Thread.CurrentThread.IsThreadPoolThread);
            }, TaskContinuationOptions.ExecuteSynchronously).Wait();

            Console.WriteLine("Enter to exit...");
            Console.ReadLine();
        }
    }
}

The output:

Task.Run, s_mark: -1
IsThreadPoolThread: True
GetStringAsync.ContinueWith, s_mark: 0
IsThreadPoolThread: True
Enter to exit...

I think this might be enough evidence to confirm the theory that an IO-bound continuation does happen on an IOCP thread.

A good read, related: "There Is No Thread" by Stephen Cleary.