TPL.Dataflow - 防止在 ActionBlock 中发生未处理的异常时挂起<T>

Question

刚从 System.Threading.Tasks.Dataflow 开始，不确定我是否理解 ActionBlock 中未处理异常的正确错误处理技术。

我现在有什么导致挂起： - ActionBlock 有未处理的异常，不再处理 - 生产者无法完成，因为它已经结束 BoundedCapacity

这是我的代码（已简化以显示一位消费者）。

internal class Program
{
    private static int _processCounter = 0;

    internal class MyClass
    {
        public MyClass(int id)
        {
            this.Id = id;
        }

        internal int Id { get; set; }
    }

    private static void Main(string[] args)
    {
        BufferBlock<MyClass> queue = new BufferBlock<MyClass>(new DataflowBlockOptions {BoundedCapacity = 10,});

        ActionBlock<MyClass> consumer =
            new ActionBlock<MyClass>(record => Process(record),
                new ExecutionDataflowBlockOptions {BoundedCapacity = 1,});

        queue.LinkTo(consumer, new DataflowLinkOptions {PropagateCompletion = true,});

        Task producer = Produce(queue);

        Trace.TraceInformation("Starting to wait on producer and consumer...");

        Task.WhenAll(producer, consumer.Completion).Wait(); // <-- this will hang. consumer.Completion is faulted, but producer is still "running".

    }

    private static async Task Produce(BufferBlock<MyClass> queue)
    {
        for (int i = 0; i < 20; i++)
        {
            await queue.SendAsync(new MyClass(i));
            Trace.TraceInformation("Sending object number {0}", i);
            await Task.Delay(1);
        }
        Trace.TraceInformation("Completing the producer");
        queue.Complete();
            // <-- we never get here because one of the SendAsync will be waiting to not excede BoundedCapacity = 10
    }

    private static void Process(MyClass myClass)
    {
        int counter = Interlocked.Increment(ref _processCounter);
        Trace.TraceInformation("Processing object number {0}", myClass.Id);
        if (counter > 4)
        {
            Trace.TraceInformation("About to throw exception for object {0}", myClass.Id);
            throw new ArgumentException("Something bad happened");
        }
    }
}

输出：

ConsoleApplication5.vshost.exe Information: 0 : Sending object number 0
ConsoleApplication5.vshost.exe Information: 0 : Starting to wait on producer and consumer...
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 1
ConsoleApplication5.vshost.exe Information: 0 : Processing object number 0
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 2
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 3
ConsoleApplication5.vshost.exe Information: 0 : Processing object number 1
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 4
ConsoleApplication5.vshost.exe Information: 0 : Processing object number 2
ConsoleApplication5.vshost.exe Information: 0 : Processing object number 3
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 5
ConsoleApplication5.vshost.exe Information: 0 : Processing object number 4
ConsoleApplication5.vshost.exe Information: 0 : About to throw exception for object 4
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 6
A first chance exception of type 'System.ArgumentException' occurred in ConsoleApplication5.exe
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 7
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 8
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 9
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 10
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 11
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 12
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 13
ConsoleApplication5.vshost.exe Information: 0 : Sending object number 14
<never finishes>

问题是，等待此类执行以确保其完成或传播异常的正确方法是什么。谢谢！

Answer 1

您可以做很多事情，因为这与您如何构建代码有关。最简单的可能是对生产者使用CancellationToken并先等待消费者：

private static void Main(string[] args)
{
    // ...

    var cts = new CancellationTokenSource();
    Task producer = Produce(queue, cts.Token);

    Trace.TraceInformation("Starting to wait on producer and consumer...");
    try
    {
        await consumer.Completion;
    }
    catch
    {
        cts.Cancel();
        // handle
    }

    try
    {
        await producer
    }
    catch
    {
        // handle
    }
}

private static async Task Produce(BufferBlock<MyClass> queue, CancellationToken token)
{
    for (int i = 0; i < 20; i++)
    {
        await queue.SendAsync(new MyClass(i), token);
        Trace.TraceInformation("Sending object number {0}", i);
        await Task.Delay(1);
    }
    Trace.TraceInformation("Completing the producer");
    queue.Complete();
}

TPL.Dataflow - 防止在 ActionBlock 中发生未处理的异常时挂起<T>

TPL.Dataflow - preventing hangs when unhanded exception occurs in ActionBlock<T>

.net

c#

tpl-dataflow