Is it safe to signal and immediately close a ManualResetEvent?

Question

I feel like I should know the answer to this, but I\'m going to ask anyway just in case I\'m making a potentially catastrophic mistake.

The following code executes a

Answer 1

When you signal with a ManualResetEvent.Set you are guaranteed that all the threads that are waiting for that event (i.e. in a blocking state on flag.WaitOne) will be signaled before returning control to the caller.

Of course there are circumstances when you might set the flag and your thread doesn't see it because it's doing some work before it checks the flag (or as nobugs suggested if you're creating multiple threads):

ThreadPool.QueueUserWorkItem(s =>
{
    QueryDB();
    flag.WaitOne();
    Console.WriteLine("Work Item 1 Executed");
});

There is contention on the flag and now you can create undefined behavior when you close it. Your flag is a shared resource between your threads, you should create a countdown latch on which every thread signals upon completion. This will eliminate contention on your flag.

public class CountdownLatch
{
    private int m_remain;
    private EventWaitHandle m_event;

    public CountdownLatch(int count)
    {
        Reset(count);
    }

    public void Reset(int count)
    {
        if (count < 0)
            throw new ArgumentOutOfRangeException();
        m_remain = count;
        m_event = new ManualResetEvent(false);
        if (m_remain == 0)
        {
            m_event.Set();
        }
    }

    public void Signal()
    {
        // The last thread to signal also sets the event.
        if (Interlocked.Decrement(ref m_remain) == 0)
            m_event.Set();
    }

    public void Wait()
    {
        m_event.WaitOne();
    }
}

1. Each thread signals on the countdown latch.
2. Your main thread waits on the countdown latch.
3. The main thread cleans up after the countdown latch signals.

Ultimately the time you sleep at the end is NOT a safe way to take care of your problems, instead you should design your program so it is 100% safe in a multithreaded environment.

UPDATE: Single Producer/Multiple Consumers
The presumption here is that your producer knows how many consumers will be created, After you create all your consumers, you reset the CountdownLatch with the given number of consumers:

// In the Producer
ManualResetEvent flag = new ManualResetEvent(false);
CountdownLatch countdown = new CountdownLatch(0);
int numConsumers = 0;
while(hasMoreWork)
{
    Consumer consumer = new Consumer(coutndown, flag);
    // Create a new thread with each consumer
    numConsumers++;
}
countdown.Reset(numConsumers);
flag.Set();
countdown.Wait();// your producer waits for all of the consumers to finish
flag.Close();// cleanup

Answer 2

It is not okay. You are getting lucky here because you only started two threads. They'll immediately start running when you call Set on a dual core machine. Try this instead and watch it bomb:

    static void Main(string[] args) {
        ManualResetEvent flag = new ManualResetEvent(false);
        for (int ix = 0; ix < 10; ++ix) {
            ThreadPool.QueueUserWorkItem(s => {
                flag.WaitOne();
                Console.WriteLine("Work Item Executed");
            });
        }
        Thread.Sleep(1000);
        flag.Set();
        flag.Close();
        Console.WriteLine("Finished");
        Console.ReadLine();
    }

Your original code will fail similarly on an old machine or your current machine when it is very busy with other tasks.

Answer 3

Looks like a risky pattern to me, even if due to the [current] implementation, it is OK. you are trying to dispose a resource which may still be in use.

It is like newing and constructing an object and deleting it blindly even before consumers of that object is done.

Even otherwise there is a problem here. Program may exit, even before other threads ever got a chance to run. Thread pool threads are background threads.

Given that you have to wait for other threads anyway, you might as well clean up afterwards.

Answer 4

My take is that there is a race condition. Having written event objects based on condition variables, you get code like this:

mutex.lock();
while (!signalled)
    condition_variable.wait(mutex);
mutex.unlock();

So while the event may be signalled, the code waiting on the event may still need access to parts of the event.

According to the documentation on Close, this only releases unmanaged resources. So if the event only uses managed resources, you may get lucky. But that could change in the future so I would err on the side of precaution and not close the event until you know it is no longer being used.