Implement C# Generic Timeout
The really tricky part here was killing the long running task through passing the executor thread from the Action back to a place where it could be aborted. I accomplished this with the use of a wrapped delegate that passes out the thread to kill into a local variable in the method that created the lambda.
I submit this example, for your enjoyment. The method you are really interested in is CallWithTimeout. This will cancel the long running thread by aborting it, and swallowing the ThreadAbortException:
Usage:
class Program
{
static void Main(string[] args)
{
//try the five second method with a 6 second timeout
CallWithTimeout(FiveSecondMethod, 6000);
//try the five second method with a 4 second timeout
//this will throw a timeout exception
CallWithTimeout(FiveSecondMethod, 4000);
}
static void FiveSecondMethod()
{
Thread.Sleep(5000);
}
The static method doing the work:
static void CallWithTimeout(Action action, int timeoutMilliseconds)
{
Thread threadToKill = null;
Action wrappedAction = () =>
{
threadToKill = Thread.CurrentThread;
try
{
action();
}
catch(ThreadAbortException ex){
Thread.ResetAbort();// cancel hard aborting, lets to finish it nicely.
}
};
IAsyncResult result = wrappedAction.BeginInvoke(null, null);
if (result.AsyncWaitHandle.WaitOne(timeoutMilliseconds))
{
wrappedAction.EndInvoke(result);
}
else
{
threadToKill.Abort();
throw new TimeoutException();
}
}
}
We are using code like this heavily in production:
var result = WaitFor<Result>.Run(1.Minutes(), () => service.GetSomeFragileResult());
Implementation is open-sourced, works efficiently even in parallel computing scenarios and is available as a part of Lokad Shared Libraries
/// <summary>
/// Helper class for invoking tasks with timeout. Overhead is 0,005 ms.
/// </summary>
/// <typeparam name="TResult">The type of the result.</typeparam>
[Immutable]
public sealed class WaitFor<TResult>
{
readonly TimeSpan _timeout;
/// <summary>
/// Initializes a new instance of the <see cref="WaitFor{T}"/> class,
/// using the specified timeout for all operations.
/// </summary>
/// <param name="timeout">The timeout.</param>
public WaitFor(TimeSpan timeout)
{
_timeout = timeout;
}
/// <summary>
/// Executes the spcified function within the current thread, aborting it
/// if it does not complete within the specified timeout interval.
/// </summary>
/// <param name="function">The function.</param>
/// <returns>result of the function</returns>
/// <remarks>
/// The performance trick is that we do not interrupt the current
/// running thread. Instead, we just create a watcher that will sleep
/// until the originating thread terminates or until the timeout is
/// elapsed.
/// </remarks>
/// <exception cref="ArgumentNullException">if function is null</exception>
/// <exception cref="TimeoutException">if the function does not finish in time </exception>
public TResult Run(Func<TResult> function)
{
if (function == null) throw new ArgumentNullException("function");
var sync = new object();
var isCompleted = false;
WaitCallback watcher = obj =>
{
var watchedThread = obj as Thread;
lock (sync)
{
if (!isCompleted)
{
Monitor.Wait(sync, _timeout);
}
}
// CAUTION: the call to Abort() can be blocking in rare situations
// http://msdn.microsoft.com/en-us/library/ty8d3wta.aspx
// Hence, it should not be called with the 'lock' as it could deadlock
// with the 'finally' block below.
if (!isCompleted)
{
watchedThread.Abort();
}
};
try
{
ThreadPool.QueueUserWorkItem(watcher, Thread.CurrentThread);
return function();
}
catch (ThreadAbortException)
{
// This is our own exception.
Thread.ResetAbort();
throw new TimeoutException(string.Format("The operation has timed out after {0}.", _timeout));
}
finally
{
lock (sync)
{
isCompleted = true;
Monitor.Pulse(sync);
}
}
}
/// <summary>
/// Executes the spcified function within the current thread, aborting it
/// if it does not complete within the specified timeout interval.
/// </summary>
/// <param name="timeout">The timeout.</param>
/// <param name="function">The function.</param>
/// <returns>result of the function</returns>
/// <remarks>
/// The performance trick is that we do not interrupt the current
/// running thread. Instead, we just create a watcher that will sleep
/// until the originating thread terminates or until the timeout is
/// elapsed.
/// </remarks>
/// <exception cref="ArgumentNullException">if function is null</exception>
/// <exception cref="TimeoutException">if the function does not finish in time </exception>
public static TResult Run(TimeSpan timeout, Func<TResult> function)
{
return new WaitFor<TResult>(timeout).Run(function);
}
}
This code is still buggy, you can try with this small test program:
static void Main(string[] args) {
// Use a sb instead of Console.WriteLine() that is modifying how synchronous object are working
var sb = new StringBuilder();
for (var j = 1; j < 10; j++) // do the experiment 10 times to have chances to see the ThreadAbortException
for (var ii = 8; ii < 15; ii++) {
int i = ii;
try {
Debug.WriteLine(i);
try {
WaitFor<int>.Run(TimeSpan.FromMilliseconds(10), () => {
Thread.Sleep(i);
sb.Append("Processed " + i + "\r\n");
return i;
});
}
catch (TimeoutException) {
sb.Append("Time out for " + i + "\r\n");
}
Thread.Sleep(10); // Here to wait until we get the abort procedure
}
catch (ThreadAbortException) {
Thread.ResetAbort();
sb.Append(" *** ThreadAbortException on " + i + " *** \r\n");
}
}
Console.WriteLine(sb.ToString());
}
}
There is a race condition. It is clearly possible that a ThreadAbortException gets raised after the method WaitFor<int>.Run() is being called. I didn't find a reliable way to fix this, however with the same test I cannot repro any problem with the TheSoftwareJedi accepted answer.
Well, you could do things with delegates (BeginInvoke, with a callback setting a flag - and the original code waiting for that flag or timeout) - but the problem is that it is very hard to shut down the running code. For example, killing (or pausing) a thread is dangerous... so I don't think there is an easy way to do this robustly.
I'll post this, but note it is not ideal - it doesn't stop the long-running task, and it doesn't clean up properly on failure.
static void Main()
{
DoWork(OK, 5000);
DoWork(Nasty, 5000);
}
static void OK()
{
Thread.Sleep(1000);
}
static void Nasty()
{
Thread.Sleep(10000);
}
static void DoWork(Action action, int timeout)
{
ManualResetEvent evt = new ManualResetEvent(false);
AsyncCallback cb = delegate {evt.Set();};
IAsyncResult result = action.BeginInvoke(cb, null);
if (evt.WaitOne(timeout))
{
action.EndInvoke(result);
}
else
{
throw new TimeoutException();
}
}
static T DoWork<T>(Func<T> func, int timeout)
{
ManualResetEvent evt = new ManualResetEvent(false);
AsyncCallback cb = delegate { evt.Set(); };
IAsyncResult result = func.BeginInvoke(cb, null);
if (evt.WaitOne(timeout))
{
return func.EndInvoke(result);
}
else
{
throw new TimeoutException();
}
}
Some minor changes to Pop Catalin's great answer:
- Func instead of Action
- Throw exception on bad timeout value
- Calling EndInvoke in case of timeout
Overloads have been added to support signaling worker to cancel execution:
public static T Invoke<T> (Func<CancelEventArgs, T> function, TimeSpan timeout) {
if (timeout.TotalMilliseconds <= 0)
throw new ArgumentOutOfRangeException ("timeout");
CancelEventArgs args = new CancelEventArgs (false);
IAsyncResult functionResult = function.BeginInvoke (args, null, null);
WaitHandle waitHandle = functionResult.AsyncWaitHandle;
if (!waitHandle.WaitOne (timeout)) {
args.Cancel = true; // flag to worker that it should cancel!
/* •————————————————————————————————————————————————————————————————————————•
| IMPORTANT: Always call EndInvoke to complete your asynchronous call. |
| http://msdn.microsoft.com/en-us/library/2e08f6yc(VS.80).aspx |
| (even though we arn't interested in the result) |
•————————————————————————————————————————————————————————————————————————• */
ThreadPool.UnsafeRegisterWaitForSingleObject (waitHandle,
(state, timedOut) => function.EndInvoke (functionResult),
null, -1, true);
throw new TimeoutException ();
}
else
return function.EndInvoke (functionResult);
}
public static T Invoke<T> (Func<T> function, TimeSpan timeout) {
return Invoke (args => function (), timeout); // ignore CancelEventArgs
}
public static void Invoke (Action<CancelEventArgs> action, TimeSpan timeout) {
Invoke<int> (args => { // pass a function that returns 0 & ignore result
action (args);
return 0;
}, timeout);
}
public static void TryInvoke (Action action, TimeSpan timeout) {
Invoke (args => action (), timeout); // ignore CancelEventArgs
}
This is how I'd do it:
public static class Runner
{
public static void Run(Action action, TimeSpan timeout)
{
IAsyncResult ar = action.BeginInvoke(null, null);
if (ar.AsyncWaitHandle.WaitOne(timeout))
action.EndInvoke(ar); // This is necesary so that any exceptions thrown by action delegate is rethrown on completion
else
throw new TimeoutException("Action failed to complete using the given timeout!");
}
}