I need to implement a library to request vk.com API. The problem is that API supports only 3 requests per second. I would like to have API asynchronous.
Important: API should support safe accessing from multiple threads.
My idea is implement some class called throttler which allow no more than 3 request/second and delay other request.
The interface is next:
public interface IThrottler : IDisposable
{
Task<TResult> Throttle<TResult>(Func<Task<TResult>> task);
}
The usage is like
var audio = await throttler.Throttle(() => api.MyAudio());
var messages = await throttler.Throttle(() => api.ReadMessages());
var audioLyrics = await throttler.Throttle(() => api.AudioLyrics(audioId));
/// Here should be delay because 3 requests executed
var photo = await throttler.Throttle(() => api.MyPhoto());
How to implement throttler?
Currently I implemented it as queue which is processed by background thread.
public Task<TResult> Throttle<TResult>(Func<Task<TResult>> task)
{
/// TaskRequest has method Run() to run task
/// TaskRequest uses TaskCompletionSource to provide new task
/// which is resolved when queue processed til this element.
var request = new TaskRequest<TResult>(task);
requestQueue.Enqueue(request);
return request.ResultTask;
}
This is shorten code of background thread loop which process the queue:
private void ProcessQueue(object state)
{
while (true)
{
IRequest request;
while (requestQueue.TryDequeue(out request))
{
/// Delay method calculates actual delay value and calls Thread.Sleep()
Delay();
request.Run();
}
}
}
Is it possible to implement this without background thread?
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);
}
Here is one solution that uses a Stopwatch:
public class Throttler : IThrottler
{
private readonly Stopwatch m_Stopwatch;
private int m_NumberOfRequestsInLastSecond;
private readonly int m_MaxNumberOfRequestsPerSecond;
public Throttler(int max_number_of_requests_per_second)
{
m_MaxNumberOfRequestsPerSecond = max_number_of_requests_per_second;
m_Stopwatch = Stopwatch.StartNew();
}
public async Task<TResult> Throttle<TResult>(Func<Task<TResult>> task)
{
var elapsed = m_Stopwatch.Elapsed;
if (elapsed > TimeSpan.FromSeconds(1))
{
m_NumberOfRequestsInLastSecond = 1;
m_Stopwatch.Restart();
return await task();
}
if (m_NumberOfRequestsInLastSecond >= m_MaxNumberOfRequestsPerSecond)
{
TimeSpan time_to_wait = TimeSpan.FromSeconds(1) - elapsed;
await Task.Delay(time_to_wait);
m_NumberOfRequestsInLastSecond = 1;
m_Stopwatch.Restart();
return await task();
}
m_NumberOfRequestsInLastSecond++;
return await task();
}
}
Here is how this code can be tested:
class Program
{
static void Main(string[] args)
{
DoIt();
Console.ReadLine();
}
static async Task DoIt()
{
Func<Task<int>> func = async () =>
{
await Task.Delay(100);
return 1;
};
Throttler throttler = new Throttler(3);
for (int i = 0; i < 10; i++)
{
var result = await throttler.Throttle(func);
Console.WriteLine(DateTime.Now);
}
}
}
You can use this as Generic
public TaskThrottle(int maxTasksToRunInParallel)
{
_semaphore = new SemaphoreSlim(maxTasksToRunInParallel);
}
public void TaskThrottler<T>(IEnumerable<Task<T>> tasks, int timeoutInMilliseconds, CancellationToken cancellationToken = default(CancellationToken)) where T : class
{
// Get Tasks as List
var taskList = tasks as IList<Task<T>> ?? tasks.ToList();
var postTasks = new List<Task<int>>();
// When the first task completed, it will flag
taskList.ForEach(x =>
{
postTasks.Add(x.ContinueWith(y => _semaphore.Release(), cancellationToken));
});
taskList.ForEach(x =>
{
// Wait for open slot
_semaphore.Wait(timeoutInMilliseconds, cancellationToken);
cancellationToken.ThrowIfCancellationRequested();
x.Start();
});
Task.WaitAll(taskList.ToArray(), cancellationToken);
}
Edit: this solution works but use it only if it is ok to process all request in serial (in one thread). Otherwise use solution accepted as answer.
Well, thanks to Best way in .NET to manage queue of tasks on a separate (single) thread
My question is almost duplicate except adding delay before execution, which is actually simple.
The main helper here is SemaphoreSlim class which allows to restrict degree of parallelism.
So, first create a semaphore:
// Semaphore allows run 1 thread concurrently.
private readonly SemaphoreSlim semaphore = new SemaphoreSlim(1, 1);
And, final version of throttle looks like
public async Task<TResult> Throttle<TResult>(Func<Task<TResult>> task)
{
await semaphore.WaitAsync();
try
{
await delaySource.Delay();
return await task();
}
finally
{
semaphore.Release();
}
}
Delay source is also pretty simple:
private class TaskDelaySource
{
private readonly int maxTasks;
private readonly TimeSpan inInterval;
private readonly Queue<long> ticks = new Queue<long>();
public TaskDelaySource(int maxTasks, TimeSpan inInterval)
{
this.maxTasks = maxTasks;
this.inInterval = inInterval;
}
public async Task Delay()
{
// We will measure time of last maxTasks tasks.
while (ticks.Count > maxTasks)
ticks.Dequeue();
if (ticks.Any())
{
var now = DateTime.UtcNow.Ticks;
var lastTick = ticks.First();
// Calculate interval between last maxTasks task and current time
var intervalSinceLastTask = TimeSpan.FromTicks(now - lastTick);
if (intervalSinceLastTask < inInterval)
await Task.Delay((int)(inInterval - intervalSinceLastTask).TotalMilliseconds);
}
ticks.Enqueue(DateTime.UtcNow.Ticks);
}
}
Related
I have a need in my asp.net webapi (framework .Net 4.7.2) to call Redis (using StackExchange.Redis) in order to delete a key in a fire and forget way and I am making some stress test.
As I am comparing the various way to have the max speed :
I have already test executing the command with the FireAndForget flag,
I have also measured a simple command to Redis by await it.
And I am now searching a way to collect a list of commands received in a window of 15ms and execute them all in one go by pipeling them.
I have first try to use a Task.Run Action to call Redis but the problem that I am observing is that under stress, the memory of my webapi keep climbing.
The memory is full of System.Threading.IThreadPoolWorkItem[] objects with the folowing code :
[HttpPost]
[Route("api/values/testpostfireforget")]
public ApiResult<int> DeleteFromBasketId([FromBody] int basketId)
{
var response = new DeleteFromBasketResponse<int>();
var cpt = Interlocked.Increment(ref counter);
Task.Run(async () => {
await db.StringSetAsync($"BASKET_TO_DELETE_{cpt}",cpt.ToString())
.ConfigureAwait(false);
});
return response;
}
So I think that under stress my api keep enqueing background task in memory and execute them one after the other as fast as it can but less than the request coming in...
So I am searching for a way to have only one long lived background thread running with the asp.net webapi, that could capture the commands to send to Redis and execute them by pipeling them.
I was thinking in runnning a background task by implementing IHostedService interface, but it seems that in this case the background task would not share any state with my current http request. So implementing a IhostedService would be handy for a scheduled background task but not in my case, or I do not know how...
Based on StackExchange.Redis documentation you can use CommandFlags.FireAndForget flag:
[HttpPost]
[Route("api/values/testpostfireforget")]
public ApiResult<int> DeleteFromBasketId([FromBody] int basketId)
{
var response = new DeleteFromBasketResponse<int>();
var cpt = Interlocked.Increment(ref counter);
db.StringSet($"BASKET_TO_DELETE_{cpt}", cpt.ToString(), flags: CommandFlags.FireAndForget);
return response;
}
Edit 1: another solution based on comment
You can use pub/sub approach. Something like this should work:
public class MessageBatcher
{
private readonly IDatabase target;
private readonly BlockingCollection<Action<IDatabaseAsync>> tasks = new();
private Task worker;
public MessageBatcher(IDatabase target) => this.target = target;
public void AddMessage(Action<IDatabaseAsync> task) => tasks.Add(task);
public IDisposable Start(int batchSize)
{
var cancellationTokenSource = new CancellationTokenSource();
worker = Task.Factory.StartNew(state =>
{
var count = 0;
var tokenSource = (CancellationTokenSource) state;
var box = new StrongBox<IBatch>(target.CreateBatch());
tokenSource.Token.Register(b => ((StrongBox<IBatch>)b).Value.Execute(), box);
foreach (var task in tasks.GetConsumingEnumerable(tokenSource.Token))
{
var batch = box.Value;
task(batch);
if (++count == batchSize)
{
batch.Execute();
box.Value = target.CreateBatch();
count = 0;
}
}
}, cancellationTokenSource, cancellationTokenSource.Token, TaskCreationOptions.LongRunning, TaskScheduler.Current);
return new Disposer(worker, cancellationTokenSource);
}
private class Disposer : IDisposable
{
private readonly Task worker;
private readonly CancellationTokenSource tokenSource;
public Disposer(Task worker, CancellationTokenSource tokenSource) => (this.worker, this.tokenSource) = (worker, tokenSource);
public void Dispose()
{
tokenSource.Cancel();
worker.Wait();
tokenSource.Dispose();
}
}
}
Usage:
private readonly MessageBatcher batcher;
ctor(MessageBatcher batcher) // ensure that passed `handler` is singleton and already already started
{
this.batcher= batcher;
}
[HttpPost]
[Route("api/values/testpostfireforget")]
public ApiResult<int> DeleteFromBasketId([FromBody] int basketId)
{
var response = new DeleteFromBasketResponse<int>();
var cpt = Interlocked.Increment(ref counter);
batcher.AddMessage(db => db.StringSetAsync($"BASKET_TO_DELETE_{cpt}", cpt.ToString(), flags: CommandFlags.FireAndForget));
return response;
}
I have a class which creates a Task which runs during its whole lifetime, that is, until Dispose() is called on it:
In the constructor I call:
_worker = Task.Run(() => ProcessQueueAsync(_workerCancellation.Token), _workerCancellation.Token);
The way I currently do it (which I am also not sure is the right way) is cancelling the CancellationToken, and waiting on the task.
public void Dispose()
{
if (_isDisposed)
{
return;
}
_workerCancellation.Cancel();
_worker.GetAwaiter().GetResult();
_isDisposed = true;
}
When I do the same in the AsyncDispose method like so:
public async ValueTask DisposeAsync()
{
await _worker;
}
I get this warning
How do I correctly dispose of such a worker? Thanks!
As requested, here is the full code of what I am trying to do:
public sealed class ActiveObjectWrapper<T, TS> : IAsyncDisposable
{
private bool _isDisposed = false;
private const int DefaultQueueCapacity = 1024;
private readonly Task _worker;
private readonly CancellationTokenSource _workerCancellation;
private readonly Channel<(T, TaskCompletionSource<TS>)> _taskQueue;
private readonly Func<T, TS> _onReceive;
public ActiveObjectWrapper(Func<T, TS> onReceive, int? queueCapacity = null)
{
_onReceive = onReceive;
_taskQueue = Channel.CreateBounded<(T, TaskCompletionSource<TS>)>(queueCapacity ?? DefaultQueueCapacity);
_workerCancellation = new CancellationTokenSource();
_worker = Task.Run(() => ProcessQueueAsync(_workerCancellation.Token), _workerCancellation.Token);
}
private async Task ProcessQueueAsync(CancellationToken cancellationToken)
{
await foreach (var (value, taskCompletionSource) in _taskQueue.Reader.ReadAllAsync(cancellationToken))
{
try
{
var result = _onReceive(value); // todo: do I need to propagate the cancellation token?
taskCompletionSource.SetResult(result);
}
catch (Exception exception)
{
taskCompletionSource.SetException(exception);
}
}
}
public async Task<TS> EnqueueAsync(T value)
{
// see: https://devblogs.microsoft.com/premier-developer/the-danger-of-taskcompletionsourcet-class/
var completionSource = new TaskCompletionSource<TS>(TaskCreationOptions.RunContinuationsAsynchronously);
await _taskQueue.Writer.WriteAsync((value, completionSource));
return await completionSource.Task;
}
public async ValueTask DisposeAsync()
{
if (_isDisposed)
{
return;
}
_taskQueue.Writer.Complete();
_workerCancellation.Cancel();
await _worker;
_isDisposed = true;
}
}
This is the pattern that I use when implementing both IDisposabe and IDisposeableAsync. It isn't strictly compliant with the .Net recommendations. I found that implementing DisposeAsyncCore() was unnecessary as my classes are sealed.
public void Dispose()
{
Dispose(disposing: true);
//GC.SuppressFinalize(this);
Worker.GetAwaiter().GetResult();
}
public void Dispose(bool disposing)
{
if (isDisposed)
{
return;
}
if (disposing)
{
lock (isDisposing)
{
if (isDisposed)
{
return;
}
Cts.Cancel();
Cts.Dispose();
isDisposed = true;
}
}
}
public async ValueTask DisposeAsync()
{
Dispose(disposing: true);
//GC.SuppressFinalize(this);
await Worker;
}
This looks like an attempt to build a TransformBlock<TIn,TOut> on top of Channels. Using a Channel properly wouldn't generate such a warning. There's no reason to use a single processing task, or store it in a field.
Using Blocks
First, the equivalent code using a TransformBlock<Tin,TOut> would be:
var block=new TransformBlock<TIn,TOut>(msgIn=>func(msgIn));
foreach(....)
{
block.Post(someMessage);
}
To stop it, block.Complete() would be enough. Any pending messages would still be processed. A TransformBlock is meant to forward its output to other blocks, eg an ActionBlock or BufferBlock.
var finalBlock=new ActionBlock<TOut>(msgOut=>Console.WriteLine(msgOut));
block.LinkTo(finalBlock,new DataflowLinkOptions{PropagateCompletion = true});
...
block.Complete();
await finalBlock.Completion;
Using Channels
Doing something similar with channels doesn't need explicit classes, or Enqueue/Dequeue methods. Channels are build with separate ChannelReader, ChannelWriter interfaces to make it easier to control ownership, concurrency and completion.
A similar pipeline using channels would require only some methods:
ChannelReader<string> FolderToChannel(string path,CancellationToken token=default)
{
Channel<int> channel=Channel.CreateUnbounded();
var writer=channel.Writer;
_ = Task.Run(async ()=>{
foreach(var path in Directory.EnumerateFiles(path))
{
await _writer.SendAsync(path);
if (token.CancellationRequested)
{
return;
}
}
},token).ContinueWith(t=>_writer.TryComplete(t.Exception));
return channel;
}
This produces a reader that can be passed to a processing method. One that could generate another reader with the results:
static ChannelReader<MyClass> ParseFile(this ChannelReader<string> reader,CancellationToken token)
{
Channel<int> channel=Channel.CreateUnbounded();
var writer=channel.Writer;
_ = Task.Run(async ()=>{
await foreach(var path in reader.ReadAllAsync(token))
{
var json= await File.ReadAllTextAsync(path);
var dto= JsonSerializer.DeserializeObject<MyClass>(json);
await _writer.SendAsync(dto);
}
},token).ContineWith(t=>writer.TryComplete(t.Exception);
return channel;
}
And a final step that only consumes a channel:
static async Task LogIt(this ChannelReader<MyClass> reader,CancellationToken token)
{
await Task.Run(async ()=>{
await foreach(var dto in reader.ReadAllAsync(token))
{
Console.WriteLine(dto);
}
},token);
}
The three steps can be combined very easily:
var cts=new CancellationTokenSource();
var complete=FolderToChannel(somePath,cts.Token)
.ParseFile(cts.Token)
.LogIt(cts.Token);
await complete;
By encapsulating the channel itself and the processing in a method there's no ambiguity about who owns the channel, who is responsible for completion or cancellation
I am trying to create a helper class for sending some information periodically to backend server.
Attaching the code below.
public class HeartBeatService
{
private CancellationToken _cancellationToken;
private CancellationTokenSource _cancellationTokenSource;
public void StartHeartBeatService(TimeSpan timeSpan)
{
_cancellationTokenSource = new CancellationTokenSource();
_cancellationToken = _cancellationTokenSource.Token;
Task.Run(async () =>
{
while (!_cancellationToken.IsCancellationRequested)
{
SendHeartBeatToAzure();
try
{
await Task.Delay(timeSpan, _cancellationToken);
}
catch
{
break;
}
}
});
}
public void SuspendHeartBeatService()
{
_cancellationTokenSource?.Cancel();
}
private async void SendHeartBeatToAzure()
{
var platformService = ServiceLocator.Get<IPlatformService>();
var location = await platformService?.GetPositionAsync();
if (!double.IsNaN(location.Item1) && !double.IsNaN(location.Item2))
{
Debug.WriteLine($"Triggering Heartbeat with location{location.Item1},{location.Item2}");
//TODO Invoke heartbeat api call.
}
}
}
The code for sending the information to server is working fine.
But there is some issue with CancellationToken which is not working/it is not cancelling.
not sure what's wrong with the implementation.
Change the signature of the SendHeartBeatToAzure to return a Task, so that it can be awaited:
private async Task SendHeartBeatToAzure()
Then await the task returned by the method inside the loop. To achieve a stable and consisted heartbeat, it is a good idea to create the Task.Delay task before calling the method:
Task.Run(async () =>
{
while (true)
{
var delayTask = Task.Delay(timeSpan, _cancellationToken);
await SendHeartBeatToAzure();
await delayTask;
}
});
As a side note, you should probably store the task returned by Task.Run as a readonly property of the HeartBeatService class, so that the status of the task can be monitored.
I'm implementing a worker engine with an upper limit to concurrency. I'm using a semaphore to wait until concurrency drops below the maximum, then use Task.Factory.StartNew to wrap the async handler in a try/catch, with a finally which releases the semaphore.
I realise this creates threads on the thread pool - but my question is, when one of those task-running threads actually awaits (on a real IO call or wait handle), is the thread returned to the pool, as I'd hope it would be?
If there's a better way to implement a task scheduler with limited concurrency where the work handler is an async method (returns Task), I'd love to hear it too. Or, let's say ideally, if there's a way to queue up an async method (again, it's a Task-returning async method) that feels less dodgy than wrapping it in a synchronous delegate and passing it to Task.Factory.StartNew, that would seem perfect..?
(This also makes me think that there are two kinds of parallelism here: how many tasks are being processed overall, but also how many continuations are running on different threads concurrently. Might be cool to have configurable options for both, though not a fixed requirement..)
Edit: snippet:
concurrencySemaphore.Wait(cancelToken);
deferRelease = false;
try
{
var result = GetWorkItem();
if (result == null)
{ // no work, wait for new work or exit signal
signal = WaitHandle.WaitAny(signals);
continue;
}
deferRelease = true;
tasks.Add(Task.Factory.StartNew(() =>
{
try
{
DoWorkHereAsync(result); // guess I'd think to .GetAwaiter().GetResult() here.. not run this yet
}
finally
{
concurrencySemaphore.Release();
}
}, cancelToken));
}
finally
{
if (!deferRelease)
{
concurrencySemaphore.Release();
}
}
Here an example of a TaskWorker, that will not produce countless worker threads.
The magic is done by awaiting SemaphoreSlim.WaitAsync() which is an IO task (and there is no thread).
class TaskWorker
{
private readonly SemaphoreSlim _semaphore;
public TaskWorker(int maxDegreeOfParallelism)
{
if (maxDegreeOfParallelism <= 0)
{
throw new ArgumentOutOfRangeException(nameof(maxDegreeOfParallelism));
}
_semaphore = new SemaphoreSlim(maxDegreeOfParallelism, maxDegreeOfParallelism);
}
public async Task RunAsync(Func<Task> taskFactory, CancellationToken cancellationToken = default(CancellationToken))
{
// No ConfigureAwait(false) here to keep the SyncContext if any
// for the real task
await _semaphore.WaitAsync(cancellationToken);
try
{
await taskFactory().ConfigureAwait(false);
}
finally
{
_semaphore.Release(1);
}
}
public async Task<T> RunAsync<T>(Func<Task<T>> taskFactory, CancellationToken cancellationToken = default(CancellationToken))
{
await _semaphore.WaitAsync(cancellationToken);
try
{
return await taskFactory().ConfigureAwait(false);
}
finally
{
_semaphore.Release(1);
}
}
}
and a simple console app to test
class Program
{
static void Main(string[] args)
{
var worker = new TaskWorker(1);
var cts = new CancellationTokenSource();
var token = cts.Token;
var tasks = Enumerable.Range(1, 10)
.Select(e => worker.RunAsync(() => SomeWorkAsync(e, token), token))
.ToArray();
Task.WhenAll(tasks).GetAwaiter().GetResult();
}
static async Task SomeWorkAsync(int id, CancellationToken cancellationToken)
{
Console.WriteLine($"Some Started {id}");
await Task.Delay(2000, cancellationToken).ConfigureAwait(false);
Console.WriteLine($"Some Finished {id}");
}
}
Update
TaskWorker implementing IDisposable
class TaskWorker : IDisposable
{
private readonly CancellationTokenSource _cts = new CancellationTokenSource();
private readonly SemaphoreSlim _semaphore;
private readonly int _maxDegreeOfParallelism;
public TaskWorker(int maxDegreeOfParallelism)
{
if (maxDegreeOfParallelism <= 0)
{
throw new ArgumentOutOfRangeException(nameof(maxDegreeOfParallelism));
}
_maxDegreeOfParallelism = maxDegreeOfParallelism;
_semaphore = new SemaphoreSlim(maxDegreeOfParallelism, maxDegreeOfParallelism);
}
public async Task RunAsync(Func<Task> taskFactory, CancellationToken cancellationToken = default(CancellationToken))
{
ThrowIfDisposed();
using (var cts = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken, _cts.Token))
{
// No ConfigureAwait(false) here to keep the SyncContext if any
// for the real task
await _semaphore.WaitAsync(cts.Token);
try
{
await taskFactory().ConfigureAwait(false);
}
finally
{
_semaphore.Release(1);
}
}
}
public async Task<T> RunAsync<T>(Func<Task<T>> taskFactory, CancellationToken cancellationToken = default(CancellationToken))
{
ThrowIfDisposed();
using (var cts = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken, _cts.Token))
{
await _semaphore.WaitAsync(cts.Token);
try
{
return await taskFactory().ConfigureAwait(false);
}
finally
{
_semaphore.Release(1);
}
}
}
private void ThrowIfDisposed()
{
if (disposedValue)
{
throw new ObjectDisposedException(this.GetType().FullName);
}
}
#region IDisposable Support
private bool disposedValue = false;
protected virtual void Dispose(bool disposing)
{
if (!disposedValue)
{
if (disposing)
{
_cts.Cancel();
// consume all semaphore slots
for (int i = 0; i < _maxDegreeOfParallelism; i++)
{
_semaphore.WaitAsync().GetAwaiter().GetResult();
}
_semaphore.Dispose();
_cts.Dispose();
}
disposedValue = true;
}
}
public void Dispose()
{
Dispose(true);
}
#endregion
}
You can think that thread is returned to a ThreadPool even thought it is not actauly a return. The thread simply picks next queued item when async operation starts.
I would suggest you to look at Task.Run instead of Task.Factory.StartNew Task.Run vs Task.Factory.StartNew.
And also have a look at TPL DataFlow. I think it will match your requirements.
I have some methods returning Task<T> on which I can await at will. I'd like to have those Tasks executed on a custom TaskScheduler instead of the default one.
var task = GetTaskAsync ();
await task;
I know I can create a new TaskFactory (new CustomScheduler ()) and do a StartNew () from it, but StartNew () takes an action and create the Task, and I already have the Task (returned behind the scenes by a TaskCompletionSource)
How can I specify my own TaskScheduler for await ?
I think what you really want is to do a Task.Run, but with a custom scheduler. StartNew doesn't work intuitively with asynchronous methods; Stephen Toub has a great blog post about the differences between Task.Run and TaskFactory.StartNew.
So, to create your own custom Run, you can do something like this:
private static readonly TaskFactory myTaskFactory = new TaskFactory(
CancellationToken.None, TaskCreationOptions.DenyChildAttach,
TaskContinuationOptions.None, new MyTaskScheduler());
private static Task RunOnMyScheduler(Func<Task> func)
{
return myTaskFactory.StartNew(func).Unwrap();
}
private static Task<T> RunOnMyScheduler<T>(Func<Task<T>> func)
{
return myTaskFactory.StartNew(func).Unwrap();
}
private static Task RunOnMyScheduler(Action func)
{
return myTaskFactory.StartNew(func);
}
private static Task<T> RunOnMyScheduler<T>(Func<T> func)
{
return myTaskFactory.StartNew(func);
}
Then you can execute synchronous or asynchronous methods on your custom scheduler.
The TaskCompletionSource<T>.Task is constructed without any action and the scheduler
is assigned on the first call to ContinueWith(...) (from Asynchronous Programming with the Reactive Framework and the Task Parallel Library — Part 3).
Thankfully you can customize the await behavior slightly by implementing your own class deriving from INotifyCompletion and then using it in a pattern similar to await SomeTask.ConfigureAwait(false) to configure the scheduler that the task should start using in the OnCompleted(Action continuation) method (from await anything;).
Here is the usage:
TaskCompletionSource<object> source = new TaskCompletionSource<object>();
public async Task Foo() {
// Force await to schedule the task on the supplied scheduler
await SomeAsyncTask().ConfigureScheduler(scheduler);
}
public Task SomeAsyncTask() { return source.Task; }
Here is a simple implementation of ConfigureScheduler using a Task extension method with the important part in OnCompleted:
public static class TaskExtension {
public static CustomTaskAwaitable ConfigureScheduler(this Task task, TaskScheduler scheduler) {
return new CustomTaskAwaitable(task, scheduler);
}
}
public struct CustomTaskAwaitable {
CustomTaskAwaiter awaitable;
public CustomTaskAwaitable(Task task, TaskScheduler scheduler) {
awaitable = new CustomTaskAwaiter(task, scheduler);
}
public CustomTaskAwaiter GetAwaiter() { return awaitable; }
public struct CustomTaskAwaiter : INotifyCompletion {
Task task;
TaskScheduler scheduler;
public CustomTaskAwaiter(Task task, TaskScheduler scheduler) {
this.task = task;
this.scheduler = scheduler;
}
public void OnCompleted(Action continuation) {
// ContinueWith sets the scheduler to use for the continuation action
task.ContinueWith(x => continuation(), scheduler);
}
public bool IsCompleted { get { return task.IsCompleted; } }
public void GetResult() { }
}
}
Here's a working sample that will compile as a console application:
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
using System.Threading.Tasks;
namespace Example {
class Program {
static TaskCompletionSource<object> source = new TaskCompletionSource<object>();
static TaskScheduler scheduler = new CustomTaskScheduler();
static void Main(string[] args) {
Console.WriteLine("Main Started");
var task = Foo();
Console.WriteLine("Main Continue ");
// Continue Foo() using CustomTaskScheduler
source.SetResult(null);
Console.WriteLine("Main Finished");
}
public static async Task Foo() {
Console.WriteLine("Foo Started");
// Force await to schedule the task on the supplied scheduler
await SomeAsyncTask().ConfigureScheduler(scheduler);
Console.WriteLine("Foo Finished");
}
public static Task SomeAsyncTask() { return source.Task; }
}
public struct CustomTaskAwaitable {
CustomTaskAwaiter awaitable;
public CustomTaskAwaitable(Task task, TaskScheduler scheduler) {
awaitable = new CustomTaskAwaiter(task, scheduler);
}
public CustomTaskAwaiter GetAwaiter() { return awaitable; }
public struct CustomTaskAwaiter : INotifyCompletion {
Task task;
TaskScheduler scheduler;
public CustomTaskAwaiter(Task task, TaskScheduler scheduler) {
this.task = task;
this.scheduler = scheduler;
}
public void OnCompleted(Action continuation) {
// ContinueWith sets the scheduler to use for the continuation action
task.ContinueWith(x => continuation(), scheduler);
}
public bool IsCompleted { get { return task.IsCompleted; } }
public void GetResult() { }
}
}
public static class TaskExtension {
public static CustomTaskAwaitable ConfigureScheduler(this Task task, TaskScheduler scheduler) {
return new CustomTaskAwaitable(task, scheduler);
}
}
public class CustomTaskScheduler : TaskScheduler {
protected override IEnumerable<Task> GetScheduledTasks() { yield break; }
protected override bool TryExecuteTaskInline(Task task, bool taskWasPreviouslyQueued) { return false; }
protected override void QueueTask(Task task) {
TryExecuteTask(task);
}
}
}
There is no way to embed rich async features into a custom TaskScheduler. This class was not designed with async/await in mind. The standard way to use a custom TaskScheduler is as an argument to the Task.Factory.StartNew method. This method does not understand async delegates. It is possible to provide an async delegate, but it is treated as any other delegate that returns some result. To get the actual awaited result of the async delegate one must call Unwrap() to the task returned.
This is not the problem though. The problem is that the TaskScheduler infrastructure does not treat the async delegate as a single unit of work. Each task is split into multiple mini-tasks (using every await as a separator), and each mini-task is processed individually. This severely restricts the asynchronous functionality that can be implemented on top of this class. As an example here is a custom TaskScheduler that is intended to queue the supplied tasks one at a time (to limit the concurrency in other words):
public class MyTaskScheduler : TaskScheduler
{
private readonly SemaphoreSlim _semaphore = new SemaphoreSlim(1);
protected async override void QueueTask(Task task)
{
await _semaphore.WaitAsync();
try
{
await Task.Run(() => base.TryExecuteTask(task));
await task;
}
finally
{
_semaphore.Release();
}
}
protected override bool TryExecuteTaskInline(Task task,
bool taskWasPreviouslyQueued) => false;
protected override IEnumerable<Task> GetScheduledTasks() { yield break; }
}
The SemaphoreSlim should ensure that only one Task would run at a time. Unfortunately it doesn't work. The semaphore is released prematurely, because the Task passed in the call QueueTask(task) is not the task that represents the whole work of the async delegate, but only the part until the first await. The other parts are passed to the TryExecuteTaskInline method. There is no way to correlate these task-parts, because no identifier or other mechanism is provided. Here is what happens in practice:
var taskScheduler = new MyTaskScheduler();
var tasks = Enumerable.Range(1, 5).Select(n => Task.Factory.StartNew(async () =>
{
Console.WriteLine($"{DateTime.Now:HH:mm:ss.fff} Item {n} Started");
await Task.Delay(1000);
Console.WriteLine($"{DateTime.Now:HH:mm:ss.fff} Item {n} Finished");
}, default, TaskCreationOptions.None, taskScheduler))
.Select(t => t.Unwrap())
.ToArray();
Task.WaitAll(tasks);
Output:
05:29:58.346 Item 1 Started
05:29:58.358 Item 2 Started
05:29:58.358 Item 3 Started
05:29:58.358 Item 4 Started
05:29:58.358 Item 5 Started
05:29:59.358 Item 1 Finished
05:29:59.374 Item 5 Finished
05:29:59.374 Item 4 Finished
05:29:59.374 Item 2 Finished
05:29:59.374 Item 3 Finished
Disaster, all tasks are queued at once.
Conclusion: Customizing the TaskScheduler class is not the way to go when advanced async features are required.
Update: Here is another observation, regarding custom TaskSchedulers in the presence of an ambient SynchronizationContext. The await mechanism by default captures the current SynchronizationContext, or the current TaskScheduler, and invokes the continuation on either the captured context
or the scheduler. If both are present, the current SynchronizationContext is preferred, and the current TaskScheduler is ignored. Below is a demonstration of this behavior, in a WinForms application¹:
private async void Button1_Click(object sender, EventArgs e)
{
await Task.Factory.StartNew(async () =>
{
MessageBox.Show($"{Thread.CurrentThread.ManagedThreadId}, {TaskScheduler.Current}");
await Task.Delay(1000);
MessageBox.Show($"{Thread.CurrentThread.ManagedThreadId}, {TaskScheduler.Current}");
}, default, TaskCreationOptions.None,
TaskScheduler.FromCurrentSynchronizationContext()).Unwrap();
}
Clicking the button causes two messages to popup sequentially, with this information:
1, System.Threading.Tasks.SynchronizationContextTaskScheduler
1, System.Threading.Tasks.ThreadPoolTaskScheduler
This experiment shows that only the first part of the asynchronous delegate, the part before the first await, was scheduled on the non-default scheduler.
This behavior limits even further the practical usefulness of custom TaskSchedulers in an async/await-enabled environment.
¹ Windows Forms applications have a WindowsFormsSynchronizationContext installed automatically, when the Application.Run method is called.
Can you fit for this method call:
await Task.Factory.StartNew(
() => { /* to do what you need */ },
CancellationToken.None, /* you can change as you need */
TaskCreationOptions.None, /* you can change as you need */
customScheduler);
After the comments it looks like you want to control the scheduler on which the code after the await is run.
The compile creates a continuation from the await that runs on the current SynchronizationContext by default. So your best shot is to set up the SynchronizationContext before calling await.
There are some ways to await a specific context. See Configure Await from Jon Skeet, especially the part about SwitchTo, for more information on how to implement something like this.
EDIT:
The SwitchTo method from TaskEx has been removed, as it was too easy to misuse. See the MSDN Forum for reasons.
Faced with same issue, tried to use LimitedConcurrencyLevelTaskScheduler, but it does not support async tasks. So...
Just wrote my own small simple Scheduler, that allow to run async Tasks based on global ThreadPool (and Task.Run method) with ability to limit current max degree of parallelism. It is enough for my exact purposes, maybe will also help you, guys.
Main demo code (console app, dotnet core 3.1) :
static async Task Main(string[] args)
{
//5 tasks to run per time
int concurrentLimit = 5;
var scheduler = new ThreadPoolConcurrentScheduler(concurrentLimit);
//catch all errors in separate event handler
scheduler.OnError += Scheduler_OnError;
// just monitor "live" state and output to console
RunTaskStateMonitor(scheduler);
// simulate adding new tasks "on the fly"
SimulateAddingTasksInParallel(scheduler);
Console.WriteLine("start adding 50 tasks");
//add 50 tasks
for (var i = 1; i <= 50; i++)
{
scheduler.StartNew(myAsyncTask);
}
Console.WriteLine("50 tasks added to scheduler");
Thread.Sleep(1000000);
}
Supporting code (place it in the same place) :
private static void Scheduler_OnError(Exception ex)
{
Console.WriteLine(ex.ToString());
}
private static int currentTaskFinished = 0;
//your sample of async task
static async Task myAsyncTask()
{
Console.WriteLine("task started ");
using (HttpClient httpClient = new HttpClient())
{
//just make http request to ... wikipedia!
//sorry, Jimmy Wales! assume,guys, you will not DDOS wiki :)
var uri = new Uri("https://wikipedia.org/");
var response = await httpClient.GetAsync(uri);
string result = await response.Content.ReadAsStringAsync();
if (string.IsNullOrEmpty(result))
Console.WriteLine("error, await is not working");
else
Console.WriteLine($"task result : site length is {result.Length}");
}
//or simulate it using by sync sleep
//Thread.Sleep(1000);
//and for tesing exception :
//throw new Exception("my custom error");
Console.WriteLine("task finished ");
//just incrementing total ran tasks to output in console
Interlocked.Increment(ref currentTaskFinished);
}
static void SimulateAddingTasksInParallel(ThreadPoolConcurrentScheduler taskScheduler)
{
int runCount = 0;
Task.Factory.StartNew(() =>
{
while (true)
{
runCount++;
if (runCount > 5)
break;
//every 10 sec 5 times
Thread.Sleep(10000);
//adding new 5 tasks from outer task
Console.WriteLine("start adding new 5 tasks!");
for (var i = 1; i <= 5; i++)
{
taskScheduler.StartNew(myAsyncTask);
}
Console.WriteLine("new 5 tasks added!");
}
}, TaskCreationOptions.LongRunning);
}
static void RunTaskStateMonitor(ThreadPoolConcurrentScheduler taskScheduler)
{
int prev = -1;
int prevQueueSize = -1;
int prevFinished = -1;
Task.Factory.StartNew(() =>
{
while (true)
{
// getting current thread count in working state
var currCount = taskScheduler.GetCurrentWorkingThreadCount();
// getting inner queue state
var queueSize = taskScheduler.GetQueueTaskCount();
//just output overall state if something changed
if (prev != currCount || queueSize != prevQueueSize || prevFinished != currentTaskFinished)
{
Console.WriteLine($"Monitor : running tasks:{currCount}, queueLength:{queueSize}. total Finished tasks : " + currentTaskFinished);
prev = currCount;
prevQueueSize = queueSize;
prevFinished = currentTaskFinished;
}
// check it every 10 ms
Thread.Sleep(10);
}
}
, TaskCreationOptions.LongRunning);
}
Scheduler :
public class ThreadPoolConcurrentScheduler
{
private readonly int _limitParallelThreadsCount;
private int _threadInProgressCount = 0;
public delegate void onErrorDelegate(Exception ex);
public event onErrorDelegate OnError;
private ConcurrentQueue<Func<Task>> _taskQueue;
private readonly object _queueLocker = new object();
public ThreadPoolConcurrentScheduler(int limitParallelThreadsCount)
{
//set maximum parallel tasks to run
_limitParallelThreadsCount = limitParallelThreadsCount;
// thread-safe queue to store tasks
_taskQueue = new ConcurrentQueue<Func<Task>>();
}
//main method to start async task
public void StartNew(Func<Task> task)
{
lock (_queueLocker)
{
// checking limit
if (_threadInProgressCount >= _limitParallelThreadsCount)
{
//waiting new "free" threads in queue
_scheduleTask(task);
}
else
{
_startNewTask(task);
}
}
}
private void _startNewTask(Func<Task> task)
{
Interlocked.Increment(ref _threadInProgressCount);
Task.Run(async () =>
{
try
{
await task();
}
catch (Exception e)
{
//Console.WriteLine(e);
OnError?.Invoke(e);
}
}).ContinueWith(_onTaskEnded);
}
//will be called on task end
private void _onTaskEnded(Task task)
{
lock (_queueLocker)
{
Interlocked.Decrement(ref _threadInProgressCount);
//queue has more priority, so if thread is free - let's check queue first
if (!_taskQueue.IsEmpty)
{
if (_taskQueue.TryDequeue(out var result))
{
_startNewTask(result);
}
}
}
}
private void _scheduleTask(Func<Task> task)
{
_taskQueue.Enqueue(task);
}
//returning in progress task count
public int GetCurrentWorkingThreadCount()
{
return _threadInProgressCount;
}
//return number of tasks waiting to run
public int GetQueueTaskCount()
{
lock (_queueLocker) return _taskQueue.Count;
}
}
Few notes :
First - check comments to it, maybe it is the worst code ever!
Did not test in prod
Did not implement cancellation tokens and any other functionality, that should be there, but i'm too lazy. Sorry