我有两个任务要执行,一个任务要获取数据,大约需要2秒钟才能完成,另一个任务要处理数据,大约需要3秒钟。我必须针对大约13.000个不同的元素运行这两个任务。如果我安全地运行此任务,则大约需要18个小时才能完成。
现在我正在考虑使用模式WaitAllOneByOne,但是在每个元素上调用Task.Run
会使我的数据库崩溃,因此我想到了实现SemaphoreSlim来控制每个任务上应该运行多少个线程。
到目前为止,我已经能够创建该模式的实现,但是我不知道它是否是线程安全的,并且我对如何对其进行测试一无所知。
代码如下:
class Program
{
static async Task Main(string[] args)
{
List<Guid> idsCombinacion = GetIdsCombinacion(10);
List<Task<string>> todo = new List<Task<string>>(idsCombinacion.Count);
List<Task> processing = new List<Task>(idsCombinacion.Count);
//this starts callings to GetData, up to 2 threads.
SemaphoreSlim semaphoreGetData = new SemaphoreSlim(2);
var dataTask = GetDataAsync(idsCombinacion, todo, semaphoreGetData);
SemaphoreSlim semaphoreProcessData = new SemaphoreSlim(3);
while (todo.Count > 0)
{
Task<string>[] tasks = todo.ToArray();
int idx = Task.WaitAny(tasks);
todo.RemoveAt(idx);
//once the tasks in the to-do list goes finishing, we start a new task to perform the Processing, up to 3 threads
semaphoreProcessData.Wait();
var result = tasks[idx].Result;
tasks[idx].Dispose();
var t = Task.Factory.StartNew(() =>
{
Task.Delay(3000).Wait();
Console.WriteLine($"todo: {todo.Count}; processing: {processing.Count}; guid: {result}");
semaphoreProcessData.Release();
}, TaskCreationOptions.LongRunning);
processing.Add(t);
}
var aux = processing.ToArray();
//we wait for all the jobs to complete
Task.WaitAll(aux);
await dataTask;
semaphoreGetData.Dispose();
semaphoreProcessData.Dispose();
}
private static async Task GetDataAsync(List<Guid> idsCombinacion, List<Task<string>> todo, SemaphoreSlim semaphoreGetData)
{
foreach (var idComb in idsCombinacion)
{
await semaphoreGetData.WaitAsync();
var t = Task<string>.Factory.StartNew(() =>
{
Task.Delay(2000).Wait();
semaphoreGetData.Release();
return "Guid: " + idComb;
}, TaskCreationOptions.LongRunning);
todo.Add(t);
}
}
private static List<Guid> GetIdsCombinacion(int howMany)
{
var idsCombinacion = new List<Guid>(howMany);
for (int i = 0; i < howMany; i++)
idsCombinacion.Add(Guid.NewGuid());
return idsCombinacion;
}
}
Joe Hummel博士在此pluralsight course中的模式WaitAllOneByOne。