Develop Reference

async call in async void EventHandler leads to a deadlock

Is there a way to call SendAsync in OnConnect without leading to a deadlock? I'm not using .Wait or .Result and it still leads to a deadlock.
Edit:
The actual problem is that SendAsync is being called twice (once at OnConnect and once at Main). If I put a await Task.Delay(10000) before the second call in Main, it actually works good. How can I fix it? If there is no task delay, it basically hangs on await tcs.Task.ConfigureAwait(false), because it's being called twice and async void OnConnect is kinda "fire and forget", meaning that it's not waiting for the first SendAsync to complete, before it goes for the second call.
// Program.cs
var client = new Client(key, secret);
await client.StartAsync().ConfigureAwait(false);
await Task.Delay(3000); // This line fixes it, but it's kinda fake fix
await client.SendAsync(request).ConfigureAwait(false);
await client.SendAsync(request2).ConfigureAwait(false);
Console.ReadLine();
// Client.cs
public class Client
{
private static long _nextId;
private readonly WebSocketClient _webSocket;
private readonly ConcurrentDictionary<long, TaskCompletionSource<string>> _outstandingRequests = new();
...
public event EventHandler<ConnectEventArgs>? Connected;
public event EventHandler<MessageReceivedEventArgs>? MessageReceived;
public ValueTask StartAsync()
{
_client.Connected += OnConnect;
_client.MessageReceived += OnMessageReceived;
return _webSocket.StartAsync(); // there is a long-running `Task.Run` inside it, which keeps the web socket connection and its pipelines open.
}
private async void OnConnect(object? sender, ConnectEventArgs e)
{
await AuthAsync(...); // the problematic line
}
private void OnMessageReceived(object? sender, MessageReceivedEventArgs e)
{
... deserialization stuff
if (_requests.TryRemove(response.Id, out var tcs))
{
tcs.TrySetResult(message);
}
}
public ValueTask<TResponse?> SendAsync<TResponse>(JsonRpcRequest request)
{
var tcs = new TaskCompletionSource<string>(TaskCreationOptions.RunContinuationsAsynchronously);
_requests.TryAdd(request.Id, tcs);
return SendRequestAndWaitForResponseAsync();
async ValueTask<TResponse?> SendRequestAndWaitForResponseAsync()
{
var message = JsonSerializer.Serialize(request);
await _client.SendAsync(message).ConfigureAwait(false);
var response = await tcs.Task.ConfigureAwait(false); // it hangs here (deadlock)
return JsonSerializer.Deserialize<TResponse>(response);
}
}
public ValueTask<JsonRpcResponse?> AuthAsync(JsonRpcRequest request)
{
return SendAsync<JsonRpcResponse>(request);
}
private static long NextId()
{
return Interlocked.Increment(ref _nextId);
}
}
public sealed class WebSocketClient
{
private readonly AsyncManualResetEvent _sendSemaphore = new(false); // Nito.AsyncEx
private readonly WebSocketPipe _webSocket; // SignalR Web Socket Pipe
...
public event EventHandler<ConnectEventArgs>? Connected;
public event EventHandler<DisconnectEventArgs>? Disconnected;
public event EventHandler<MessageReceivedEventArgs>? MessageReceived;
public ValueTask StartAsync()
{
_ = Task.Run(async () =>
{
try
{
await CreatePolicy()
.ExecuteAsync(async () =>
{
await _webSocket.StartAsync(new Uri(_url), CancellationToken.None).ConfigureAwait(false);
Connected?.Invoke(this, new ConnectEventArgs());
_sendSemaphore.Set();
await ReceiveLoopAsync().ConfigureAwait(false);
})
.ConfigureAwait(false);
}
catch (Exception ex)
{
// Failed after all retries
Disconnected?.Invoke(this, new DisconnectEventArgs(ex));
}
});
return ValueTask.CompletedTask;
}
public async ValueTask SendAsync(string message)
{
await _sendSemaphore.WaitAsync().ConfigureAwait(false);
var encoded = Encoding.UTF8.GetBytes(message);
await _webSocket.Transport!.Output
.WriteAsync(new ArraySegment<byte>(encoded, 0, encoded.Length), CancellationToken.None)
.ConfigureAwait(false);
}
private IAsyncPolicy CreatePolicy()
{
var retryPolicy = Policy
.Handle<WebSocketException>()
.WaitAndRetryForeverAsync(_ => ReconnectInterval,
(exception, retryCount, calculatedWaitDuration) =>
{
_sendSemaphore.Reset();
Reconnecting?.Invoke(this, new ReconnectingEventArgs(exception, retryCount, calculatedWaitDuration));
return Task.CompletedTask;
});
return retryPolicy;
}
private async Task ReceiveLoopAsync()
{
while (true)
{
var result = await _webSocket.Transport!.Input.ReadAsync(CancellationToken.None).ConfigureAwait(false);
var buffer = result.Buffer;
...
}
}
}

As mentioned in the comments, those web socket wrappers are using System.IO.Pipelines, which is incorrect. System.IO.Pipelines is a stream of bytes, so it's appropriate for (non-web) sockets; a web socket is a stream of messages, so something like System.Threading.Channels would be more appropriate.
You could try something like this, which I just typed up and haven't even run:
public sealed class ChannelWebSocket : IDisposable
{
private readonly WebSocket _webSocket;
private readonly Channel<Message> _input;
private readonly Channel<Message> _output;
public ChannelWebSocket(WebSocket webSocket, Options options)
{
_webSocket = webSocket;
_input = Channel.CreateBounded(new BoundedChannelOptions(options.InputCapacity)
{
FullMode = options.InputFullMode,
}, options.InputMessageDropped);
_output = Channel.CreateBounded(new BoundedChannelOptions(options.OutputCapacity)
{
FullMode = options.OutputFullMode,
}, options.OutputMessageDropped);
}
public ChannelReader<Message> Input => _input.Reader;
public ChannelWriter<Message> Output => _output.Writer;
public void Dispose() => _webSocket.Dispose();
public async void Start()
{
var inputTask = InputLoopAsync(default);
var outputTask = OutputLoopAsync(default);
var completedTask = await Task.WhenAny(inputTask, outputTask);
if (completedTask.Exception != null)
{
try { await _webSocket.CloseAsync(WebSocketCloseStatus.InternalServerError, statusDescription: null, default); } catch { /* ignore */ }
try { _input.Writer.Complete(completedTask.Exception); } catch { /* ignore */ }
try { _output.Writer.Complete(completedTask.Exception); } catch { /* ignore */ }
}
}
public sealed class Message
{
public WebSocketMessageType MessageType { get; set; }
public OwnedMemorySequence<byte> Payload { get; set; } = null!;
}
private async Task InputLoopAsync(CancellationToken cancellationToken)
{
while (true)
{
var payload = new OwnedMemorySequence<byte>();
var buffer = MemoryPool<byte>.Shared.Rent();
ValueWebSocketReceiveResult result;
do
{
result = await _webSocket.ReceiveAsync(buffer.Memory, cancellationToken);
if (result.MessageType == WebSocketMessageType.Close)
{
_input.Writer.Complete();
return;
}
payload.Append(buffer.Slice(0, result.Count));
} while (!result.EndOfMessage);
await _input.Writer.WriteAsync(new Message
{
MessageType = result.MessageType,
Payload = payload,
}, cancellationToken);
}
}
private async Task OutputLoopAsync(CancellationToken cancellationToken)
{
await foreach (var message in _output.Reader.ReadAllAsync())
{
var sequence = message.Payload.ReadOnlySequence;
if (sequence.IsEmpty)
continue;
while (!sequence.IsSingleSegment)
{
await _webSocket.SendAsync(sequence.First, message.MessageType, endOfMessage: false, cancellationToken);
sequence = sequence.Slice(sequence.First.Length);
}
await _webSocket.SendAsync(sequence.First, message.MessageType, endOfMessage: true, cancellationToken);
message.Payload.Dispose();
}
await _webSocket.CloseAsync(WebSocketCloseStatus.NormalClosure, statusDescription: null, cancellationToken);
}
public sealed class Options
{
public int InputCapacity { get; set; } = 16;
public BoundedChannelFullMode InputFullMode { get; set; } = BoundedChannelFullMode.Wait;
public Action<Message>? InputMessageDropped { get; set; }
public int OutputCapacity { get; set; } = 16;
public BoundedChannelFullMode OutputFullMode { get; set; } = BoundedChannelFullMode.Wait;
public Action<Message>? OutputMessageDropped { get; set; }
}
}
It uses this type for building a memory sequence:
public sealed class MemorySequence<T>
{
private MemorySegment? _head;
private MemorySegment? _tail;
public MemorySequence<T> Append(ReadOnlyMemory<T> buffer)
{
if (_tail == null)
_head = _tail = new MemorySegment(buffer, runningIndex: 0);
else
_tail = _tail.Append(buffer);
return this;
}
public ReadOnlySequence<T> ReadOnlySequence => CreateReadOnlySequence(0, _tail?.Memory.Length ?? 0);
public ReadOnlySequence<T> CreateReadOnlySequence(int firstBufferStartIndex, int lastBufferEndIndex) =>
_tail == null ? new ReadOnlySequence<T>(Array.Empty<T>()) :
new ReadOnlySequence<T>(_head!, firstBufferStartIndex, _tail, lastBufferEndIndex);
private sealed class MemorySegment : ReadOnlySequenceSegment<T>
{
public MemorySegment(ReadOnlyMemory<T> memory, long runningIndex)
{
Memory = memory;
RunningIndex = runningIndex;
}
public MemorySegment Append(ReadOnlyMemory<T> nextMemory)
{
var next = new MemorySegment(nextMemory, RunningIndex + Memory.Length);
Next = next;
return next;
}
}
}
and this type for building an owned memory sequence:
public sealed class OwnedMemorySequence<T> : IDisposable
{
private readonly CollectionDisposable _disposable = new();
private readonly MemorySequence<T> _sequence = new();
public OwnedMemorySequence<T> Append(IMemoryOwner<T> memoryOwner)
{
_disposable.Add(memoryOwner);
_sequence.Append(memoryOwner.Memory);
return this;
}
public ReadOnlySequence<T> ReadOnlySequence => _sequence.ReadOnlySequence;
public ReadOnlySequence<T> CreateReadOnlySequence(int firstBufferStartIndex, int lastBufferEndIndex) =>
_sequence.CreateReadOnlySequence(firstBufferStartIndex, lastBufferEndIndex);
public void Dispose() => _disposable.Dispose();
}
which depends on owned memory span extension methods I stole from here:
public static class MemoryOwnerSliceExtensions
{
public static IMemoryOwner<T> Slice<T>(this IMemoryOwner<T> owner, int start, int length)
{
if (start == 0 && length == owner.Memory.Length)
return owner;
return new SliceOwner<T>(owner, start, length);
}
public static IMemoryOwner<T> Slice<T>(this IMemoryOwner<T> owner, int start)
{
if (start == 0)
return owner;
return new SliceOwner<T>(owner, start);
}
private sealed class SliceOwner<T> : IMemoryOwner<T>
{
private readonly IMemoryOwner<T> _owner;
public Memory<T> Memory { get; }
public SliceOwner(IMemoryOwner<T> owner, int start, int length)
{
_owner = owner;
Memory = _owner.Memory.Slice(start, length);
}
public SliceOwner(IMemoryOwner<T> owner, int start)
{
_owner = owner;
Memory = _owner.Memory[start..];
}
public void Dispose() => _owner.Dispose();
}
}
This code is all completely untested; use at your own risk.

Better way to serialize with a timeout in c#.Net

My use case:
In a single threaded application, I need to serialize arbitrary classes for logging purposes.
The arbitrary classes are predominantly translated in an automated way from a massive VB6 application into .NET.
If serialized without a timeout, the serialization method will loop until it runs out of memory.
This is what I have currently:
internal class Serializer
{
private readonly log4net.ILog log = log4net.LogManager.GetLogger(System.Reflection.MethodBase.GetCurrentMethod().DeclaringType);
public volatile string result = null;
public volatile Func<string> toExecute = null;
public Thread thread;
public ManualResetEventSlim messageToSender = new ManualResetEventSlim(false);
public ManualResetEventSlim messageToReceiver = new ManualResetEventSlim(false);
public Serializer()
{
thread = new Thread(new ThreadStart(run));
thread.Start();
}
~Serializer()
{
try
{
if (messageToSender != null) messageToSender.Dispose();
}
catch { };
try
{
if (messageToReceiver != null) messageToReceiver.Dispose();
}
catch { };
}
public volatile bool ending = false;
public void run()
{
while (!ending)
{
try
{
if (toExecute != null)
{
result = toExecute();
}
messageToReceiver.Reset();
messageToSender.Set();
messageToReceiver.Wait();
}
catch (ThreadInterruptedException)
{
log.Warn("Serialization interrupted");
break;
}
catch (ThreadAbortException)
{
Thread.ResetAbort();
result = null;
}
catch (Exception ex)
{
log.Error("Error in Serialization", ex);
Console.WriteLine(ex);
break;
}
}
}
}
public class LocalStructuredLogging
{
private static volatile Serializer _serializer;
private static Serializer serializer
{
get
{
if (_serializer == null)
{
_serializer = new Serializer();
}
return _serializer;
}
}
public void LogStucturedEnd()
{
try
{
if (serializer != null)
{
serializer.ending = true;
serializer.thread.Interrupt();
}
}
catch { }
}
internal ConcurrentDictionary<long, bool> disallowedToSerialize = new ConcurrentDictionary<long, bool>();
public string TrySerialize<T>(T payload, [CallerLineNumber] int line = 0)
{
long hashEl = typeof(T).Name.GetHashCode() * line;
bool dummy;
unchecked
{
if (disallowedToSerialize.TryGetValue(hashEl, out dummy))
{
return "°,°";
}
}
serializer.toExecute = () =>
{
try
{
return Newtonsoft.Json.JsonConvert.SerializeObject(payload, new Newtonsoft.Json.JsonSerializerSettings() { ReferenceLoopHandling = Newtonsoft.Json.ReferenceLoopHandling.Ignore });
}
catch (Exception)
{
disallowedToSerialize.TryAdd(hashEl, false);
return "°°°";
}
};
try
{
serializer.messageToSender.Reset();
serializer.messageToReceiver.Set();
if (serializer.messageToSender.Wait(6000))
{
return Interlocked.Exchange(ref serializer.result, null);
}
serializer.toExecute = null;
serializer.thread.Abort();
serializer.messageToSender.Wait(2000);
disallowedToSerialize.TryAdd(hashEl, false);
return "°§°";
}
catch (Exception)
{
disallowedToSerialize.TryAdd(hashEl, false);
return "°-°";
}
}
}
The code is called as in the following (test is an arbitrary class instance):
var logger = new LocalStructuredLogging();
var rr5 = logger.TrySerialize(test);
Although it seems to do the job, there are some issues with it:
it has a dependency on Thread.Abort
it is time dependent, so it will thus produce varied results on a loaded system
every class instance is treated like every other class instance - no tweaking
...
So, are there any better solutions available ?

Based upon dbc's excellent answer, I managed to create a better timed serializer.
It resolves all 3 issues mentioned above:
public class TimedJsonTextWriter : JsonTextWriter
{
public int? MaxDepth { get; set; }
public TimeSpan? MaxTimeUsed { get; set; }
public int MaxObservedDepth { get; private set; }
private DateTime start = DateTime.Now;
public TimedJsonTextWriter(TextWriter writer, JsonSerializerSettings settings, TimeSpan? maxTimeUsed)
: base(writer)
{
this.MaxDepth = (settings == null ? null : settings.MaxDepth);
this.MaxObservedDepth = 0;
this.MaxTimeUsed = maxTimeUsed;
}
public TimedJsonTextWriter(TextWriter writer, TimeSpan? maxTimeUsed, int? maxDepth = null)
: base(writer)
{
this.MaxDepth = maxDepth;
this.MaxTimeUsed = maxTimeUsed;
}
public override void WriteStartArray()
{
base.WriteStartArray();
CheckDepth();
}
public override void WriteStartConstructor(string name)
{
base.WriteStartConstructor(name);
CheckDepth();
}
public override void WriteStartObject()
{
base.WriteStartObject();
CheckDepth();
}
uint checkDepthCounter = 0;
private void CheckDepth()
{
MaxObservedDepth = Math.Max(MaxObservedDepth, Top);
if (Top > MaxDepth)
throw new JsonSerializationException($"Depth {Top} Exceeds MaxDepth {MaxDepth} at path \"{Path}\"");
unchecked
{
if ((++checkDepthCounter & 0x3ff) == 0 && DateTime.Now - start > MaxTimeUsed)
throw new JsonSerializationException($"Time Usage Exceeded at path \"{Path}\"");
}
}
}
public class LocalStructuredLogging
{
public void LogStucturedEnd()
{
}
internal HashSet<long> disallowedToSerialize = new HashSet<long>();
public string TrySerialize<T>(T payload, int maxDepth = 100, int secondsToTimeout = 2, [CallerLineNumber] int line = 0)
{
long hashEl = typeof(T).Name.GetHashCode() * line;
if (disallowedToSerialize.Contains(hashEl))
{
return "°,°";
}
try
{
var settings = new JsonSerializerSettings { MaxDepth = maxDepth, ReferenceLoopHandling = Newtonsoft.Json.ReferenceLoopHandling.Ignore };
using (var writer = new StringWriter())
{
using (var jsonWriter = new TimedJsonTextWriter(writer, settings, new TimeSpan(0, 0, secondsToTimeout)))
{
JsonSerializer.Create(settings).Serialize(jsonWriter, payload);
// Log the MaxObservedDepth here, if you want to.
}
return writer.ToString();
}
}
catch (Exception)
{
disallowedToSerialize.Add(hashEl);
return "°-°";
}
}
}
The only issue remaining are the Hash collisions, which are easy to solve (e.g. by using the source file name as well or use another type of Collection).

The correct way to run an action timed would be to do something like the following. I would recommend taking a second look at how serialization should work as well :).
/// <summary>
/// Run an action timed.
/// </summary>
/// <param name="action">Action to execute timed.</param>
/// <param name="secondsTimout">Seconds before Task should cancel.</param>
/// <returns></returns>
public static async Task RunTimeout(Action action, int secondsTimout) {
var tokenSource = new CancellationTokenSource();
tokenSource.CancelAfter(TimeSpan.FromSeconds(secondsTimout));
await Task.Run(action, tokenSource.Token);
}
You may also want to return a variable upon the completion of your timed task. That can be done like so...
public static async Task<T> RunTimeout<T>(Func<T> action, int secondsTimout) {
var tokenSource = new CancellationTokenSource();
tokenSource.CancelAfter(TimeSpan.FromSeconds(secondsTimout));
var result = await Task.Run(action, tokenSource.Token);
return result;
}

How to make different calls to an async method waiting for the result of first call?

Let's say I have a method it gets data from server
Task<Result> GetDataFromServerAsync(...)
If there is an ongoing call in progress, I don't want to start a new request to server but wait for the original to finish.
Let's say I have
var result = await objet.GetDataFromServerAsync(...);
and in a different place, called almost at the same time I have a second call
var result2 = await objet.GetDataFromServerAsync(...);
I don't want the second to start a new request to server if the first didn't finish. I want both calls to get the same result as soon as first call finish. This is a proof of concept, I have options but I wanted to see how easy it's to do this.

Here is a quick example using Lazy<Task<T>>:
var lazyGetDataFromServer = new Lazy<Task<Result>>
(() => objet.GetDataFromServerAsync(...));
var result = await lazyGetDataFromServer.Value;
var result2 = await lazyGetDataFromServer.Value;
It doesn't matter if these 2 awaits are done from separate threads as Lazy<T> is thread-safe, so result2 if ran second will still wait and use the same output from result.
Using the code from here you can wrap this up in a class called AsyncLazy<T>, and add a custom GetAwaiter so that you can just await it without the need to do .Value, very tidy =)

You can use anything for syncrhonization in your method.
For example, I used SemaphoreSlim:
public class PartyMaker
{
private bool _isProcessing;
private readonly SemaphoreSlim _slowStuffSemaphore = new SemaphoreSlim(1, 1);
private DateTime _something;
public async Task<DateTime> ShakeItAsync()
{
try
{
var needNewRequest = !_isProcessing;
await _slowStuffSemaphore.WaitAsync().ConfigureAwait(false);
if (!needNewRequest) return _something;
_isProcessing = true;
_something = await ShakeItSlowlyAsync().ConfigureAwait(false);
return _something;
}
finally
{
_isProcessing = false;
_slowStuffSemaphore.Release();
}
}
private async Task<DateTime> ShakeItSlowlyAsync()
{
await Task.Delay(TimeSpan.FromSeconds(1)).ConfigureAwait(false);
return DateTime.UtcNow;
}
}
Usage:
var maker = new PartyMaker();
var tasks = new[] {maker.ShakeItAsync(), maker.ShakeItAsync()};
Task.WaitAll(tasks);
foreach (var task in tasks)
{
Console.WriteLine(task.Result);
}
Console.WriteLine(maker.ShakeItAsync().Result);
Here is result:
17.01.2017 22:28:39
17.01.2017 22:28:39
17.01.2017 22:28:41
UPD
With this modification you can call async methods with args:
public class PartyMaker
{
private readonly SemaphoreSlim _slowStuffSemaphore = new SemaphoreSlim(1, 1);
private readonly ConcurrentDictionary<int, int> _requestCounts = new ConcurrentDictionary<int, int>();
private readonly ConcurrentDictionary<int, DateTime> _cache = new ConcurrentDictionary<int, DateTime>();
public async Task<DateTime> ShakeItAsync(Argument argument)
{
var key = argument.GetHashCode();
DateTime result;
try
{
if (!_requestCounts.ContainsKey(key))
{
_requestCounts[key] = 1;
}
else
{
++_requestCounts[key];
}
var needNewRequest = _requestCounts[key] == 1;
await _slowStuffSemaphore.WaitAsync().ConfigureAwait(false);
if (!needNewRequest)
{
_cache.TryGetValue(key, out result);
return result;
}
_cache.TryAdd(key, default(DateTime));
result = await ShakeItSlowlyAsync().ConfigureAwait(false);
_cache[key] = result;
return result;
}
finally
{
_requestCounts[key]--;
if (_requestCounts[key] == 0)
{
int temp;
_requestCounts.TryRemove(key, out temp);
_cache.TryRemove(key, out result);
}
_slowStuffSemaphore.Release();
}
}
private async Task<DateTime> ShakeItSlowlyAsync()
{
await Task.Delay(TimeSpan.FromSeconds(1)).ConfigureAwait(false);
return DateTime.UtcNow;
}
}
public class Argument
{
public Argument(int value)
{
Value = value;
}
public int Value { get; }
public override int GetHashCode()
{
return Value.GetHashCode();
}
}

Cannot return IObservable<T> from a method marked async

This contrived example is roughly how my code is structured:
public abstract class SuperHeroBase
{
protected SuperHeroBase() { }
public async Task<CrimeFightingResult> FightCrimeAsync()
{
var result = new CrimeFightingResult();
result.State = CrimeFightingStates.Fighting;
try
{
await FightCrimeOverride(results);
}
catch
{
SetError(results);
}
if (result.State == CrimeFightingStates.Fighting)
result.State = CrimeFightingStates.GoodGuyWon;
return result;
}
protected SetError(CrimeFightingResult results)
{
result.State = CrimeFightingStates.BadGuyWon;
}
protected abstract Task FightCrimeOverride(CrimeFightingResult results);
}
public enum CrimeFightingStates
{
NotStarted,
Fighting,
GoodGuyWon, // success state
BadGuyWon // error state
}
public class CrimeFightingResult
{
internal class CrimeFightingResult() { }
public CrimeFightingStates State { get; internal set; }
}
Now I'm trying to build a collection that would hold multiple SuperHero objects and offer a AllHerosFightCrime method. The hero's should not all fight at once (the next one starts when the first is finished).
public class SuperHeroCollection : ObservableCollection<SuperHeroBase>
{
public SuperHeroCollection() { }
// I mark the method async...
public async IObservable<CrimeFightingResult> AllHerosFightCrime()
{
var heros = new List<SuperHeroBase>(this);
var results = new ReplaySubject<CrimeFightingResult>();
foreach (var hero in heros)
{
// ... so I can await on FightCrimeAsync and push
// the result to the subject when done
var result = await hero.FightCrimeAsync();
results.OnNext(result);
}
results.OnCompleted();
// I can't return the IObservable here because the method is marked Async.
// It expects a return type of CrimeFightingResult
return results;
}
}
How can I return the IObservable<CrimeFightingResults> and still have the call to FightCrimeAsync awaited?

You could turn your task into an observable and combine them using Merge:
public IObservable<CrimeFightingResult> AllHerosFightCrime()
{
var heros = new List<SuperHeroBase>(this);
return heros.Select(h => h.FightCrimeAsync().ToObservable())
.Merge();
}
If you want to maintain the order your events are received you can use Concat instead of Merge.

Parallel.ForEach stalled when integrated with BlockingCollection

I adopted my implementation of parallel/consumer based on the code in this question
class ParallelConsumer<T> : IDisposable
{
private readonly int _maxParallel;
private readonly Action<T> _action;
private readonly TaskFactory _factory = new TaskFactory();
private CancellationTokenSource _tokenSource;
private readonly BlockingCollection<T> _entries = new BlockingCollection<T>();
private Task _task;
public ParallelConsumer(int maxParallel, Action<T> action)
{
_maxParallel = maxParallel;
_action = action;
}
public void Start()
{
try
{
_tokenSource = new CancellationTokenSource();
_task = _factory.StartNew(
() =>
{
Parallel.ForEach(
_entries.GetConsumingEnumerable(),
new ParallelOptions { MaxDegreeOfParallelism = _maxParallel, CancellationToken = _tokenSource.Token },
(item, loopState) =>
{
Log("Taking" + item);
if (!_tokenSource.IsCancellationRequested)
{
_action(item);
Log("Finished" + item);
}
else
{
Log("Not Taking" + item);
_entries.CompleteAdding();
loopState.Stop();
}
});
},
_tokenSource.Token);
}
catch (OperationCanceledException oce)
{
System.Diagnostics.Debug.WriteLine(oce);
}
}
private void Log(string message)
{
Console.WriteLine(message);
}
public void Stop()
{
Dispose();
}
public void Enqueue(T entry)
{
Log("Enqueuing" + entry);
_entries.Add(entry);
}
public void Dispose()
{
if (_task == null)
{
return;
}
_tokenSource.Cancel();
while (!_task.IsCanceled)
{
}
_task.Dispose();
_tokenSource.Dispose();
_task = null;
}
}
And here is a test code
class Program
{
static void Main(string[] args)
{
TestRepeatedEnqueue(100, 1);
}
private static void TestRepeatedEnqueue(int itemCount, int parallelCount)
{
bool[] flags = new bool[itemCount];
var consumer = new ParallelConsumer<int>(parallelCount,
(i) =>
{
flags[i] = true;
}
);
consumer.Start();
for (int i = 0; i < itemCount; i++)
{
consumer.Enqueue(i);
}
Thread.Sleep(1000);
Debug.Assert(flags.All(b => b == true));
}
}
The test always fails - it always stuck at around 93th-item from the 100 tested. Any idea which part of my code caused this issue, and how to fix it?

You cannot use Parallel.Foreach() with BlockingCollection.GetConsumingEnumerable(), as you have discovered.
For an explanation, see this blog post:
https://devblogs.microsoft.com/pfxteam/parallelextensionsextras-tour-4-blockingcollectionextensions/
Excerpt from the blog:
BlockingCollection’s GetConsumingEnumerable implementation is using BlockingCollection’s internal synchronization which already supports multiple consumers concurrently, but ForEach doesn’t know that, and its enumerable-partitioning logic also needs to take a lock while accessing the enumerable.
As such, there’s more synchronization here than is actually necessary, resulting in a potentially non-negligable performance hit.
[Also] the partitioning algorithm employed by default by both Parallel.ForEach and PLINQ use chunking in order to minimize synchronization costs: rather than taking the lock once per element, it'll take the lock, grab a group of elements (a chunk), and then release the lock.
While this design can help with overall throughput, for scenarios that are focused more on low latency, that chunking can be prohibitive.
That blog also provides the source code for a method called GetConsumingPartitioner() which you can use to solve the problem.
public static class BlockingCollectionExtensions
{
public static Partitioner<T> GetConsumingPartitioner<T>(this BlockingCollection<T> collection)
{
return new BlockingCollectionPartitioner<T>(collection);
}
public class BlockingCollectionPartitioner<T> : Partitioner<T>
{
private BlockingCollection<T> _collection;
internal BlockingCollectionPartitioner(BlockingCollection<T> collection)
{
if (collection == null)
throw new ArgumentNullException("collection");
_collection = collection;
}
public override bool SupportsDynamicPartitions
{
get { return true; }
}
public override IList<IEnumerator<T>> GetPartitions(int partitionCount)
{
if (partitionCount < 1)
throw new ArgumentOutOfRangeException("partitionCount");
var dynamicPartitioner = GetDynamicPartitions();
return Enumerable.Range(0, partitionCount).Select(_ => dynamicPartitioner.GetEnumerator()).ToArray();
}
public override IEnumerable<T> GetDynamicPartitions()
{
return _collection.GetConsumingEnumerable();
}
}
}

The reason for failure is because of the following reason as explained here
The partitioning algorithm employed by default by both
Parallel.ForEach and PLINQ use chunking in order to minimize
synchronization costs: rather than taking the lock once per element,
it'll take the lock, grab a group of elements (a chunk), and then
release the lock.
To get it to work, you can add a method on your ParallelConsumer<T> class to indicate that the adding is completed, as below
public void StopAdding()
{
_entries.CompleteAdding();
}
And now call this method after your for loop , as below
consumer.Start();
for (int i = 0; i < itemCount; i++)
{
consumer.Enqueue(i);
}
consumer.StopAdding();
Otherwise, Parallel.ForEach() would wait for the threshold to be reached so as to grab the chunk and start processing.