Threadpool/WaitHandle resource leak/crash

I think I may need to re-think my design. I'm having a hard time narrowing down a bug that is causing my computer to completely hang, sometimes throwing an HRESULT 0x8007000E from VS 2010.

I have a console application (that I will later convert to a service) that handles transferring files based on a database queue.

I am throttling the threads allowed to transfer. This is because some systems we are connecting to can only contain a certain number of connections from certain accounts.

For example, System A can only accept 3 simultaneous connections (which means 3 separate threads). Each one of these threads has their own unique connection object, so we shouldn't run in to any synchronization problems since they aren't sharing a connection.

We want to process the files from those systems in cycles. So, for example, we will allow 3 connections that can transfer up to 100 files per connection. This means, to move 1000 files from System A, we can only process 300 files per cycle, since 3 threads are allowed with 100 files each. Therefore, over the lifetime of this transfer, we will have 10 threads. We can only run 3 at a time. So, there will be 3 cycles, and the last cycle will only use 1 thread to transfer the last 100 files. (3 threads x 100 files = 300 files per cycle)

The current architecture by example is:

1. A System.Threading.Timer checks the queue every 5 seconds for something to do by calling GetScheduledTask()
2. If there's nothing to, GetScheduledTask() simply does nothing
3. If there is work, create a ThreadPool thread to process the work [Work Thread A]
4. Work Thread A sees that there are 1000 files to transfer
5. Work Thread A sees that it can only have 3 threads running to the system it is getting files from
6. Work Thread A starts three new work threads [B,C,D] and transfers
7. Work Thread A waits for B,C,D [WaitHandle.WaitAll(transfersArray)]
8. Work Thread A sees that there are still more files in the queue (should be 700 now)
9. Work Thread A creates a new array to wait on [transfersArray = new TransferArray[3] which is the max for System A, but could vary on system
10. Work Thread A starts three new work threads [B,C,D] and waits for them [WaitHandle.WaitAll(transfersArray)]
11. The process repeats until there are no more files to move.
12. Work Thread A signals that it is done

I am using ManualResetEvent to handle the signaling.

My questions are:

1. Is there any glaring circumstance which would cause a resource leak or problem that I am experiencing?
2. Should I loop thru the array after every WaitHandle.WaitAll(array) and call array[index].Dispose()?
3. The Handle count under the Task Manager for this process slowly creeps up
4. I am calling the initial creation of Worker Thread A from a System.Threading.Timer. Is there going to be any problems with this? The code for that timer is:

(Some class code for scheduling)

private ManualResetEvent _ResetEvent;

private void Start()
{
    _IsAlive = true;
    ManualResetEvent transferResetEvent = new ManualResetEvent(false);
    //Set the scheduler timer to 5 second intervals
    _ScheduledTasks = new Timer(new TimerCallback(ScheduledTasks_Tick), transferResetEvent, 200, 5000);
}

private void ScheduledTasks_Tick(object state)
{
    ManualResetEvent resetEvent = null;
    try
    {
        resetEvent = (ManualResetEvent)state;
        //Block timer until GetScheduledTasks() finishes
        _ScheduledTasks.Change(Timeout.Infinite, Timeout.Infinite);
        GetScheduledTasks();
    }
    finally
    {
        _ScheduledTasks.Change(5000, 5000);
        Console.WriteLine("{0} [Main] GetScheduledTasks() finished", DateTime.Now.ToString("MMddyy HH:mm:ss:fff"));
        resetEvent.Set();
    }
}


private void GetScheduledTask()
{
    try 
    { 
        //Check to see if the database connection is still up
        if (!_IsAlive)
        {
            //Handle
            _ConnectionLostNotification = true;
            return;
        }

        //Get scheduled records from the database
        ISchedulerTask task = null;

        using (DataTable dt = FastSql.ExecuteDataTable(
                _ConnectionString, "hidden for security", System.Data.CommandType.StoredProcedure,
                new List<FastSqlParam>() { new FastSqlParam(ParameterDirection.Input, SqlDbType.VarChar, "@ProcessMachineName", Environment.MachineName) })) //call to static class
        {
            if (dt != null)
            {
                if (dt.Rows.Count == 1)
                {  //Only 1 row is allowed
                    DataRow dr = dt.Rows[0];

                    //Get task information
                    TransferParam.TaskType taskType = (TransferParam.TaskType)Enum.Parse(typeof(TransferParam.TaskType), dr["TaskTypeId"].ToString());
                    task = ScheduledTaskFactory.CreateScheduledTask(taskType);

                    task.Description = dr["Description"].ToString();
                    task.IsEnabled = (bool)dr["IsEnabled"];
                    task.IsProcessing = (bool)dr["IsProcessing"];
                    task.IsManualLaunch = (bool)dr["IsManualLaunch"];
                    task.ProcessMachineName = dr["ProcessMachineName"].ToString();
                    task.NextRun = (DateTime)dr["NextRun"];
                    task.PostProcessNotification = (bool)dr["NotifyPostProcess"];
                    task.PreProcessNotification = (bool)dr["NotifyPreProcess"];
                    task.Priority = (TransferParam.Priority)Enum.Parse(typeof(TransferParam.SystemType), dr["PriorityId"].ToString());
                    task.SleepMinutes = (int)dr["SleepMinutes"];
                    task.ScheduleId = (int)dr["ScheduleId"];
                    task.CurrentRuns = (int)dr["CurrentRuns"];
                    task.TotalRuns = (int)dr["TotalRuns"];

                    SchedulerTask scheduledTask = new SchedulerTask(new ManualResetEvent(false), task);
                    //Queue up task to worker thread and start
                    ThreadPool.QueueUserWorkItem(ne开发者_开发技巧w WaitCallback(this.ThreadProc), scheduledTask);     
                }
            }
        }

    }
    catch (Exception ex)
    {
        //Handle
    }
}

private void ThreadProc(object taskObject)
{
    SchedulerTask task = (SchedulerTask)taskObject;
    ScheduledTaskEngine engine = null;
    try
    {
        engine = SchedulerTaskEngineFactory.CreateTaskEngine(task.Task, _ConnectionString);
        engine.StartTask(task.Task);    
    }
    catch (Exception ex)
    {
        //Handle
    }
    finally
    {
        task.TaskResetEvent.Set();
        task.TaskResetEvent.Dispose();
    }
}

0x8007000E is an out-of-memory error. That and the handle count seem to point to a resource leak. Ensure you're disposing of every object that implements IDisposable. This includes the arrays of ManualResetEvents you're using.

If you have time, you may also want to convert to using the .NET 4.0 Task class; it was designed to handle complex scenarios like this much more cleanly. By defining child Task objects, you can reduce your overall thread count (threads are quite expensive not only because of scheduling but also because of their stack space).

I'm looking for answers to a similar problem (Handles Count increasing over time).

I took a look at your application architecture and like to suggest you something that could help you out:

Have you heard about IOCP (Input Output Completion Ports).

I'm not sure of the dificulty to implement this using C# but in C/C++ it is a piece of cake. By using this you create a unique thread pool (The number of threads in that pool is in general defined as 2 x the number of processors or processors cores in the PC or server) You associate this pool to a IOCP Handle and the pool does the work. See the help for these functions: CreateIoCompletionPort(); PostQueuedCompletionStatus(); GetQueuedCompletionStatus();

In General creating and exiting threads on the fly could be time consuming and leads to performance penalties and memory fragmentation. There are thousands of literature about IOCP in MSDN and in google.

I think you should reconsider your architecture altogether. The fact that you can only have 3 simultaneously connections is almost begging you to use 1 thread to generate the list of files and 3 threads to process them. Your producer thread would insert all files into a queue and the 3 consumer threads will dequeue and continue processing as items arrive in the queue. A blocking queue can significantly simplify the code. If you are using .NET 4.0 then you can take advantage of the BlockingCollection class.

public class Example
{
    private BlockingCollection<string> m_Queue = new BlockingCollection<string>();

    public void Start()
    {
        var threads = new Thread[] 
            { 
                new Thread(Producer), 
                new Thread(Consumer), 
                new Thread(Consumer), 
                new Thread(Consumer) 
            };
        foreach (Thread thread in threads)
        {
            thread.Start();
        }
    }

    private void Producer()
    {
        while (true)
        {
            Thread.Sleep(TimeSpan.FromSeconds(5));
            ScheduledTask task = GetScheduledTask();
            if (task != null)
            {
                foreach (string file in task.Files)
                {
                    m_Queue.Add(task);
                }
            }
        }
    }

    private void Consumer()
    {
        // Make a connection to the resource that is assigned to this thread only.
        while (true)
        {
            string file = m_Queue.Take();
            // Process the file.
        }
    }
}

I have definitely oversimplified things in the example above, but I hope you get the general idea. Notice how this is much simpler as there is not much in the way of thread synchronization (most will be embedded in the blocking queue) and of course there is no use of WaitHandle objects. Obviously you would have to add in the correct mechanisms to shut down the threads gracefully, but that should be fairly easy.

It turns out the source of this strange problem was not related to architecture but rather because of converting the solution from 3.5 to 4.0. I re-created the solution, performing no code changes, and the problem never occurred again.