Question

方案是将各种任务排入队列，节制它们以进行并行处理，并能够取消它们。我的问题是，由于所有任务已经挂在信号量上，取消它们实际上比任务本身需要更长的时间。

很明显，.Run将所有任务都放入.WaitAsync中，因此其状态为WaitingForActiviation。因此，赋予任务本身的令牌实际上是没有意义的：所有1000个任务已经在运行。

将令牌提供给WaitAsync似乎会使应用程序在取消时冻结。

static SemaphoreSlim batcher = new SemaphoreSlim(5);

        static void Main(string[] args)
        {            
            var tokenStore = new CancellationTokenSource();
            var tasks = Enumerable.Range(1, 1000).Select(i => 
                DoableWork(tokenStore.Token)
            ).ToList();

            do {
                if (Console.KeyAvailable && Console.ReadKey(true).Key == ConsoleKey.Escape) {
                    Console.WriteLine("Cancelling tasks");
                    tokenStore.Cancel();
                }
                Thread.Sleep(1000);
                Console.WriteLine($"Tasks: {string.Join(", ", tasks.GroupBy(t => t.Status).Select(x => $"{x.Count()} {x.Key}"))}");
            } while (tasks.Any(t => t.Status < TaskStatus.RanToCompletion));
        }

        private static Task DoableWork(CancellationToken token)
        {
            return Task.Run(async () => {
                try {
                    await batcher.WaitAsync();
                    token.ThrowIfCancellationRequested();
                    await Task.Delay(200, token); // Do stuff
                } catch (OperationCanceledException) {
                    throw;
                } catch (Exception) {
                    // Logging
                    throw;
                } finally {
                    batcher.Release();
                }
            }, token);
        }

结果：

Tasks: 25 RanToCompletion, 975 WaitingForActivation
Tasks: 50 RanToCompletion, 950 WaitingForActivation
Tasks: 75 RanToCompletion, 925 WaitingForActivation
Tasks: 100 RanToCompletion, 900 WaitingForActivation
Tasks: 120 RanToCompletion, 880 WaitingForActivation
Tasks: 145 RanToCompletion, 855 WaitingForActivation
Cancelling tasks
Tasks: 145 RanToCompletion, 16 Canceled, 839 WaitingForActivation
Tasks: 145 RanToCompletion, 33 Canceled, 822 WaitingForActivation
Tasks: 145 RanToCompletion, 51 Canceled, 804 WaitingForActivation
Tasks: 145 RanToCompletion, 66 Canceled, 789 WaitingForActivation
Tasks: 145 RanToCompletion, 81 Canceled, 774 WaitingForActivation
Tasks: 145 RanToCompletion, 101 Canceled, 754 WaitingForActivation

如您所见，每秒取消的任务少于处理的任务。以这种速度，您将等待一整分钟，以取消1000个待处理的任务。

使用新的Task（）作为替代方案，并调用Task.Start（）将导致取消取消机制，而不会引发每个正在运行的任务的未处理异常。

tasks.Where(t => t.Status < TaskStatus.Running)
     .Take(5 - tasks.Count(t => t.Status == TaskStatus.Running))
     .ToList()
     .ForEach(t => t.Start());

private static Task DoableWork(CancellationToken token)
        {
            return new Task(async () => {
                try {
                    await Task.Delay(200, token); // Do stuff
                } catch (OperationCanceledException) {
                    throw;
                } catch (Exception) {
                    // Logging
                    throw;
                }
            }, token);
        }

Answer 1

发现了问题。 VS 2019（或其设置之一）在调试模式下大大减慢了异常处理的速度。当我直接运行.exe时，取消是即时的。感谢@TheodorZoulias显示无法重现该问题。

用信号量取消异步任务花费的时间太长

1 个答案: