TPL DataFlow处理异常的正确方法

Question

我在使用TPL DataFlow管理队列（数据库）并将工作重定向到网格计算服务的Windows服务中遇到问题。有一次BufferBlock停止释放任务，我不确定为什么。我认为这是因为在执行某些任务期间发生了一些异常，但是它们被抑制了，很难理解BufferBlock在何时停止接受新任务。

我试图在下面的工作示例中简化它。它没有任何异常处理，我想知道如何在TPL中正确处理异常。我在这里TPL Dataflow, guarantee completion only when ALL source data blocks completed找到了类似的内容。在此示例中，我有100个请求，并与10个请求一起批量处理数据。模拟某些异常，如果ID％9 == 0，则会发生如果我没有捕获到此异常，它将工作一些，然后停止接受新请求。我相信，如果我处理并返回Result.Failure可以正常工作，但是我不确定在生产环境中是否有适当的方法。

我是TPL的新手，如果我不能更清楚地说明我的问题，请忘记我。GitHub Project

Image Empty Slots

using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Net;
using System.Text;
using System.Threading;
using System.Threading.Tasks;
using System.Threading.Tasks.Dataflow;
using System.Timers;
using CSharpFunctionalExtensions;

namespace TestTPL
{
    public class ServicePipeline
    {
        public const int batches = 100;
        private int currentBatch = 0;

        public ServicePipeline(int maxRequestsInParallel)
        {
            MaxRequestsInParallel = maxRequestsInParallel;
        }

        public int MaxRequestsInParallel { get; }
        public BufferBlock<MyData> QueueBlock { get; private set; }
        public List<TransformBlock<MyData, Result>> ExecutionBlocks
            { get; private set; }
        public ActionBlock<Result> ResultBlock { get; private set; }

        private void Init()
        {
            QueueBlock = new BufferBlock<MyData>(new DataflowBlockOptions()
                { BoundedCapacity = MaxRequestsInParallel });
            ExecutionBlocks = new List<TransformBlock<MyData, Result>>();
            ResultBlock = new ActionBlock<Result>(_ => _.OnFailure(
                () => Console.WriteLine($"Error: {_.Error}")));

            for (int blockIndex = 0; blockIndex < MaxRequestsInParallel;
                blockIndex++)
            {
                var executionBlock = new TransformBlock<MyData, Result>((d) =>
                {
                    return ExecuteAsync(d);
                }, new ExecutionDataflowBlockOptions() { BoundedCapacity = 1 });
                executionBlock.LinkTo(ResultBlock, new DataflowLinkOptions()
                    { PropagateCompletion = true });
                QueueBlock.LinkTo(executionBlock, new DataflowLinkOptions()
                    { PropagateCompletion = true });
                ExecutionBlocks.Add(executionBlock);
            }
        }

        public static Result ExecuteAsync(MyData myData)
        {
            //try
            //{
            WebClient web = new WebClient();
            TaskCompletionSource<Result> res = new TaskCompletionSource<Result>();
            Task task = Task<Result>.Run(() => web.DownloadStringAsync(
                new Uri("http://localhost:49182/Slow.ashx")));
            task.Wait();
            Console.WriteLine($"Data = {myData}");
            if (myData != null && myData.Id % 9 == 0)
                throw new Exception("Test");
            return Result.Ok();
            //}
            //catch (Exception ex)
            //{
            //    return Result.Failure($"Exception: {ex.Message}");
            //}
        }

        public async void Start()
        {
            Init();
            while (currentBatch < batches)
            {
                Thread.Sleep(1000);
                await SubmitNextRequests();
            }
            Console.WriteLine($"Completed: {batches}");
        }

        private async Task<int> SubmitNextRequests()
        {
            var emptySlots = MaxRequestsInParallel - QueueBlock.Count;
            Console.WriteLine($"Empty slots: {emptySlots}" +
                $", left = {batches - currentBatch}");
            if (emptySlots > 0)
            {
                var dataRequests = await GetNextRequests(emptySlots);
                foreach (var data in dataRequests)
                {
                    await QueueBlock.SendAsync(data);
                }
            }
            return emptySlots;
        }

        private async Task<List<MyData>> GetNextRequests(int request)
        {
            MyData[] myDatas = new MyData[request];
            Task<List<MyData>> task = Task<List<MyData>>.Run(() =>
            {
                for (int i = 0; i < request; i++)
                {
                    myDatas[i++] = new MyData(currentBatch);
                    currentBatch++;
                }
                return new List<MyData>(myDatas);
            });
            return await task;
        }
    }

    public class MyData
    {
        public int Id { get; set; }
        public MyData(int id) => Id = id;
        public override string ToString() { return Id.ToString(); }
    }
}

编辑：10/30/2019在处理并显式调用异常时，它可以按预期方式工作Result.Failure（$“ Exception：{ex.Message}”）;

    public static Result ExecuteAsync(MyData myData)
    {
        try
        {
            WebClient web = new WebClient();
            TaskCompletionSource<Result> res = new TaskCompletionSource<Result>();
            Task task = Task<Result>.Run(() => Thread.Sleep(2000));
            task.Wait();
            Console.WriteLine($"Data = {myData}");
            if (myData != null && myData.Id % 9 == 0)
                throw new Exception("Test");
            return Result.Ok();
        }
        catch (Exception ex)
        {
            return Result.Failure($"Exception: {ex.Message}");
        }
    }

Answer 1

链接两个块时，有一个选项可以向前传播完成，但不能向后传播。当使用BoundedCapacity选项时，这将成为一个问题，并且会发生错误，因为它会阻塞管道的进纸器并导致死锁。但是，手动传播完成非常容易。这是您可以使用的方法。

async void OnErrorComplete(IDataflowBlock block1, IDataflowBlock block2)
{
    await Task.WhenAny(block1.Completion); // Safe awaiting
    if (block1.Completion.IsFaulted) block2.Complete();
}

它异步等待block1完成，如果失败，则立即完成block2。通常，完成上游块就足够了，但是如果需要，您还可以传播特定的异常：

async void OnErrorPropagate(IDataflowBlock block1, IDataflowBlock block2)
{
    await Task.WhenAny(block1.Completion); // Safe awaiting
    if (block1.Completion.IsFaulted)
        block2.Fault(block1.Completion.Exception.InnerException);
}