WhenAll on the large number of Task

Question

我需要您的帮助来找到最佳解决方案。这是我的原始代码：

public async Task Test()
{
    var tasks = new List<Task>();
    string line;
    using (var streamReader = File.OpenText(InputPath))
    {
        while ((line = streamReader.ReadLine()) != null)
        {
            tasks.Add(Process(line));
        }
    }
    await Task.WhenAll(tasks.ToArray());
}
private Task Process(string line)
{
    return Task.Run(() =>
    {
        Console.WriteLine(line);
    });
}

它将读取一个有行的文件，并通过一个任务处理每一行。然而，如果文件有100多万行，任务数组就更大，那么这个代码还不错吗？或者我应该找到另一个解决方案。请帮帮我。谢谢。

Answer 1

这是个坏主意。这可能会启动太多线程。

一个更好的方法是简单地使用Parallel.ForEach()，如下所示：

using System;
using System.IO;
using System.Threading.Tasks;
namespace Demo
{
    static class Program
    {
        static void Main()
        {
            string filename = @"Your test filename goes here";
            Parallel.ForEach(File.ReadLines(filename), process);
        }
        private static void process(string line)
        {
            Console.WriteLine(line);
        }
    }
}

然而，这并不使用async/await。但是，如果您愿意，您可以将对Parallel.ForEach()的整个调用封装在一个任务中。

或者，如果你想使用任务并行库（一个Microsoft NuGet包），你可以这样做：

using System;
using System.IO;
using System.Threading;
using System.Threading.Tasks;
using System.Threading.Tasks.Dataflow;
namespace Demo
{
    static class Program
    {
        static void Main()
        {
            Task.Run(test).Wait();
        }
        static async Task test()
        {
            string filename = @"Your filename goes here";
            await processFile(filename);
        }
        static async Task processFile(string filename)
        {
            var options = new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 8, BoundedCapacity = 100 };
            var action = new ActionBlock<string>(s => process(s), options);
            foreach (var line in File.ReadLines(filename))
                await action.SendAsync(line);
            action.Complete();
            await action.Completion;
        }
        static void process(string line)
        {
            Thread.Sleep(100);  // Simulate work.
            Console.WriteLine(Thread.CurrentThread.ManagedThreadId + " " + line);
        }
    }
}

这将为您提供async支持。

---

附录：线程池节流的演示。

（这是对shay__评论的回应。）

如果启动了许多长时间运行的任务，而该任务的运行时间超过一秒钟左右，则可能会看到线程池限制。

如果当前进程的线程池线程数等于或超过调用ThreadPool.GetMinThreads(out workers, out ports);返回的worker计数，就会发生这种情况。

如果发生这种情况，在创建新的线程池线程之前，新线程池线程的启动将延迟一小段时间（在我的系统上是一秒钟）。通常情况下，这将允许另一个线程池线程变得可用，而这将被使用（这当然是节流）。

以下代码演示了该问题：

int workers, ports;
ThreadPool.GetMinThreads(out workers, out ports);
Console.WriteLine("Min workers = " + workers); // Prints 8 on my system.
var sw = Stopwatch.StartNew();
for (int i = 0; i < 100; ++i)
{
    Task.Run(() =>
    {
        Console.WriteLine($"Thread {Thread.CurrentThread.ManagedThreadId} started at time {sw.Elapsed}");
        Thread.Sleep(10000);
    });
}
Console.ReadLine();

在我的系统上，这会打印以下内容：

Min workers = 8
Thread 3 started at time 00:00:00.0098651
Thread 6 started at time 00:00:00.0098651
Thread 8 started at time 00:00:00.0099841
Thread 5 started at time 00:00:00.0099680
Thread 7 started at time 00:00:00.0099918
Thread 4 started at time 00:00:00.0098739
Thread 10 started at time 00:00:00.0100828
Thread 9 started at time 00:00:00.0101833
Thread 11 started at time 00:00:01.0096247
Thread 12 started at time 00:00:02.0098105
Thread 13 started at time 00:00:03.0099824
Thread 14 started at time 00:00:04.0100671
Thread 15 started at time 00:00:05.0098035
Thread 16 started at time 00:00:06.0099449
Thread 17 started at time 00:00:07.0096293
Thread 18 started at time 00:00:08.0106774
Thread 19 started at time 00:00:09.0098193
Thread 20 started at time 00:00:10.0104156
Thread 3 started at time 00:00:10.0109315
Thread 8 started at time 00:00:10.0112171
Thread 7 started at time 00:00:10.0112531
Thread 9 started at time 00:00:10.0117256
Thread 4 started at time 00:00:10.0117920
Thread 10 started at time 00:00:10.0117298
Thread 6 started at time 00:00:10.0109381
Thread 5 started at time 00:00:10.0112276
Thread 21 started at time 00:00:11.0095859
Thread 11 started at time 00:00:11.0101189
Thread 22 started at time 00:00:12.0095421
Thread 12 started at time 00:00:12.0111173
Thread 23 started at time 00:00:13.0095932    ...

请注意，前8个线程的启动速度非常快，但随后新线程被限制在每秒一个左右，直到第一批线程终止，然后可以重用。

还要注意，只有当线程需要相对较长的时间才能终止时，才会出现这种效果。