On the subject of waiting until tasks are complete and thread synchronisation.
I currently have an iteration i have enclosed within a Parallel.ForEach. In the Example below I have posed some questions in the comments about how best to handle the graceful termination of the loop (.NET 4.0);
private void myFunction()
{
IList<string> iListOfItems = new List<string>();
// populate iListOfItems
CancellationTokenSource cts = new CancellationTokenSource();
ParallelOptions po = new ParallelOptions();
po.MaxDegreeOfParallelism = 20; // max threads
po.CancellationToken = cts.Token;
try
{
var myWcfProxy = new myWcfClientSoapClient();
if (Parallel.ForEach(iListOfItems, po, (item, loopsate) =>
{
try
{
if (_requestedToStop)
loopsate.Stop();
// long running blocking WS call, check before and after
var response = myWcfProxy.ProcessIntervalConfiguration(item);
if (_requestedToStop)
loopsate.Stop();
// perform some local processing of the response object
}
catch (Exception ex)
{
// cannot continue game over.
if (myWcfProxy.State == CommunicationState.Faulted)
{
loopsate.Stop();
throw;
}
}
// else carry on..
// raise some events and other actions that could all risk an unhanded error.
}
).IsCompleted)
{
RaiseAllItemsCompleteEvent();
}
}
catch (Exception ex)
{
// if an unhandled error is raised within one of the Parallel.ForEach threads, do all threads in the
// ForEach abort? or run to completion? Is loopsate.Stop (or equivalent) called as soon as the framework raises an Exception?
// Do I need to call cts.Cancel here?
// I want to wait for all the threads to terminate before I continue at this point. Howe do we achieve that?
// do i need to call cts.Dispose() ?
MessageBox.Show(Logging.FormatException(ex));
}
finally
{
if (myWcfProxy != null)
{
// possible race condition with the for-each threads here unless we wait for them to terminate.
if (myWcfProxy.State == System.ServiceModel.CommunicationState.Faulted)
myWcfProxy.Abort();
myWcfProxy.Close();
}
// possible race condition with the for-each threads here unless we wait for them to terminate.
_requestedToStop = false;
}
}
Any help would be most appreciated. The MSDN documentation talks of ManualResetEventSlim's and cancellationToken.WaitHandle's. but not sure how to wire them into this, seem to be struggling understanding the MSDN examples as most dont apply.
I have mocked up some code below that may answer your question. The basic point is that you get fork/join parallelism with Parallel.ForEach, so you don't need to worry about race conditions outside of the parallel task (the calling thread blocks until the tasks have completed, successfully or otherwise). You just want to make sure to use the LoopState variable (the second argument to the lambda) to control your loop state.
If any iteration of the loop threw an unhandled exception, the overall loop will raise the AggregateException caught at the end.
Other links that mention this topic:
Parallel.ForEach throws exception when processing extremely large sets of data
http://msdn.microsoft.com/en-us/library/dd460720.aspx
Does Parallel.ForEach limits the number of active threads?