Queue of async tasks with throttling which supports muti-threading
So we'll start out with a solution to a simpler problem, that of creating a queue that process up to N tasks concurrently, rather than throttling to N tasks started per second, and build on that:
public class TaskQueue
{
private SemaphoreSlim semaphore;
public TaskQueue()
{
semaphore = new SemaphoreSlim(1);
}
public TaskQueue(int concurrentRequests)
{
semaphore = new SemaphoreSlim(concurrentRequests);
}
public async Task<T> Enqueue<T>(Func<Task<T>> taskGenerator)
{
await semaphore.WaitAsync();
try
{
return await taskGenerator();
}
finally
{
semaphore.Release();
}
}
public async Task Enqueue(Func<Task> taskGenerator)
{
await semaphore.WaitAsync();
try
{
await taskGenerator();
}
finally
{
semaphore.Release();
}
}
}
We'll also use the following helper methods to match the result of a TaskCompletionSource
to a `Task:
public static void Match<T>(this TaskCompletionSource<T> tcs, Task<T> task)
{
task.ContinueWith(t =>
{
switch (t.Status)
{
case TaskStatus.Canceled:
tcs.SetCanceled();
break;
case TaskStatus.Faulted:
tcs.SetException(t.Exception.InnerExceptions);
break;
case TaskStatus.RanToCompletion:
tcs.SetResult(t.Result);
break;
}
});
}
public static void Match<T>(this TaskCompletionSource<T> tcs, Task task)
{
Match(tcs, task.ContinueWith(t => default(T)));
}
Now for our actual solution what we can do is each time we need to perform a throttled operation we create a TaskCompletionSource
, and then go into our TaskQueue
and add an item that starts the task, matches the TCS to its result, doesn't await it, and then delays the task queue for 1 second. The task queue will then not allow a task to start until there are no longer N tasks started in the past second, while the result of the operation itself is the same as the create Task
:
public class Throttler
{
private TaskQueue queue;
public Throttler(int requestsPerSecond)
{
queue = new TaskQueue(requestsPerSecond);
}
public Task<T> Enqueue<T>(Func<Task<T>> taskGenerator)
{
TaskCompletionSource<T> tcs = new TaskCompletionSource<T>();
var unused = queue.Enqueue(() =>
{
tcs.Match(taskGenerator());
return Task.Delay(TimeSpan.FromSeconds(1));
});
return tcs.Task;
}
public Task Enqueue<T>(Func<Task> taskGenerator)
{
TaskCompletionSource<bool> tcs = new TaskCompletionSource<bool>();
var unused = queue.Enqueue(() =>
{
tcs.Match(taskGenerator());
return Task.Delay(TimeSpan.FromSeconds(1));
});
return tcs.Task;
}
}
I solved a similar problem using a wrapper around SemaphoreSlim. In my scenario, I had some other throttling mechanisms as well, and I needed to make sure that requests didn't hit the external API too often even if request number 1 took longer to reach the API than request number 3. My solution was to use a wrapper around SemaphoreSlim that had to be released by the caller, but the actual SemaphoreSlim would not be released until a set time had passed.
public class TimeGatedSemaphore
{
private readonly SemaphoreSlim semaphore;
public TimeGatedSemaphore(int maxRequest, TimeSpan minimumHoldTime)
{
semaphore = new SemaphoreSlim(maxRequest);
MinimumHoldTime = minimumHoldTime;
}
public TimeSpan MinimumHoldTime { get; }
public async Task<IDisposable> WaitAsync()
{
await semaphore.WaitAsync();
return new InternalReleaser(semaphore, Task.Delay(MinimumHoldTime));
}
private class InternalReleaser : IDisposable
{
private readonly SemaphoreSlim semaphoreToRelease;
private readonly Task notBeforeTask;
public InternalReleaser(SemaphoreSlim semaphoreSlim, Task dependantTask)
{
semaphoreToRelease = semaphoreSlim;
notBeforeTask = dependantTask;
}
public void Dispose()
{
notBeforeTask.ContinueWith(_ => semaphoreToRelease.Release());
}
}
}
Example usage:
private TimeGatedSemaphore requestThrottler = new TimeGatedSemaphore(3, TimeSpan.FromSeconds(1));
public async Task<T> MyRequestSenderHelper(string endpoint)
{
using (await requestThrottler.WaitAsync())
return await SendRequestToAPI(endpoint);
}