Develop Reference

.NET Async in shutdown methods?

I have an application that connects to a REST API using async methods. I have this set up using async/await pretty much everywhere that connects to the API, however I have a question and some strange behavior that I don't completely understand. What I want to do is simply return a license in certain scenarios when the program shuts down. This is initiated by a window closing event; the event handler is as follows:
async void Window_Closing(object sender, System.ComponentModel.CancelEventArgs e)
{
...other synchronous code...
//Check for floating licensing
if (KMApplication.License != null && KMApplication.License.Scope != Enums.LicenseScope.Standalone)
{
for (int i = 0; i < 3; i++)
{
try
{
await KMApplication.License.ShutDown(KMApplication.Settings == null
? Enums.LicenseReturnModes.PromptOnShutdown
: KMApplication.Settings.LicenseReturnMode)
.ConfigureAwait(false);
break;
}
catch (Exception ex)
{
_logger.Warn("Exception in license release, attempt " + i, ex);
}
}
}
await KMApplication.ApiService.Disconnect().ConfigureAwait(false);
_logger.Info("Shutdown Complete");
Application.Current?.Shutdown();
}
When this runs I can step through in the debugger and it gets to the first license shutdown call which is the first async awaited call. Then when I press F10 to step to the next line of code it just shuts down and is gone. I verified that the license release that is supposed to be happening in that line is in face happening so it appears to run to completion of that line but then shuts down or crashes or something. I also looked at the logs and it never gets to the Shutdown Complete line and I don't believe it's getting to the ApiService.Disconnect either.
I also tried running this as a sync method using Task.Run(() => ...the method...).GetAwaiter().GetResult() but that just deadlocks on the first call.
How do I handle this and have it run the async release, wait for it to be done, then shut down?

The fundamental problem in what you're trying to do is that async/await assumes the main application thread continues running. This assumption directly conflicts with the shutdown action, whose job is to terminate all running tasks.
If you examine the documentation on Window_Closing, it states the following (and only the following):
Occurs directly after Close() is called, and can be handled to cancel window closure.
This is important. The only thing this is supposed to do is allow you to programmatically cancel the window closure, thus prompting some additional user action.
Your expectations are befuddled because of how async/await works. Async/await appears to run in a linear fashion; however, what actually happens is that control is passed back to the caller at the first await. The framework assumes at that point that you do not wish to cancel the form close, and the program is allowed to terminate, taking all other actions with it.
Fundamentally, all C-style programs have a main entry point, which runs a loop. It's been that way since the early days of C, and continues that way in WPF. However, in WPF, Microsoft got a bit clever, and decided to hide this from the programmer. There are a couple of options to deal with things that need to happen after main window closing:
Re-hijack the main loop from your program, and put the code there. The details on how to do this may be found here.
Set an explicit shutdown mode, and kick off the task to initiate that. Call Application.Shutdown() as the very last line of code you need to execute.

Here is an async version of the FormClosing event. It delays the closing of the form until the completion of the supplied Task. The user is prevented from closing the form before the completion of the task.
The OnFormClosingAsync event passes an enhanced version of the FormClosingEventArgs class to the handling code, with two additional properties: bool HideForm and int Timeout. These properties are read/write, much like the existing Cancel property. Setting HideForm to true has the effect of hiding the form while the async operation is in progress, to avoid frustrating the user. Setting Timeout to a value > 0 has the effect of abandoning the async operation after the specified duration in msec, and closing the form. Otherwise it is possible that the application could be left running indefinitely with a hidden UI, which could certainly be a problem. The Cancel property is still usable, and can be set to true by the handler of the event, to prevent the form from closing.
static class WindowsFormsAsyncExtensions
{
public static IDisposable OnFormClosingAsync(this Form form,
Func<object, FormClosingAsyncEventArgs, Task> handler)
{
Task compositeTask = null;
form.FormClosing += OnFormClosing; // Subscribe to the event
return new Disposer(() => form.FormClosing -= OnFormClosing);
async void OnFormClosing(object sender, FormClosingEventArgs e)
{
if (compositeTask != null)
{
// Prevent the form from closing before the task is completed
if (!compositeTask.IsCompleted) { e.Cancel = true; return; }
// In case of success allow the form to close
if (compositeTask.Status == TaskStatus.RanToCompletion) return;
// Otherwise retry calling the handler
}
e.Cancel = true; // Cancel the normal closing of the form
var asyncArgs = new FormClosingAsyncEventArgs(e.CloseReason);
var handlerTask = await Task.Factory.StartNew(
() => handler(sender, asyncArgs),
CancellationToken.None, TaskCreationOptions.DenyChildAttach,
TaskScheduler.Default); // Start in a thread-pool thread
var hideForm = asyncArgs.HideForm;
var timeout = asyncArgs.Timeout;
if (hideForm) form.Visible = false;
compositeTask = Task.WhenAny(handlerTask, Task.Delay(timeout)).Unwrap();
try
{
await compositeTask; // Await and then continue in the UI thread
}
catch (OperationCanceledException) // Treat this as Cancel = true
{
if (hideForm) form.Visible = true;
return;
}
catch // On error don't leave the form hidden
{
if (hideForm) form.Visible = true;
throw;
}
if (asyncArgs.Cancel) // The caller requested to cancel the form close
{
compositeTask = null; // Forget the completed task
if (hideForm) form.Visible = true;
return;
}
await Task.Yield(); // Ensure that form.Close will run asynchronously
form.Close(); // Finally close the form
}
}
private struct Disposer : IDisposable
{
private readonly Action _action;
public Disposer(Action disposeAction) => _action = disposeAction;
void IDisposable.Dispose() => _action?.Invoke();
}
}
public class FormClosingAsyncEventArgs : EventArgs
{
public CloseReason CloseReason { get; }
private volatile bool _cancel;
public bool Cancel { get => _cancel; set => _cancel = value; }
private volatile bool _hideForm;
public bool HideForm { get => _hideForm; set => _hideForm = value; }
private volatile int _timeout;
public int Timeout { get => _timeout; set => _timeout = value; }
public FormClosingAsyncEventArgs(CloseReason closeReason) : base()
{
this.CloseReason = closeReason;
this.Timeout = System.Threading.Timeout.Infinite;
}
}
Since OnFormClosingAsync is an extension method and not a real event, it can only have a single handler.
Usage example:
public Form1()
{
InitializeComponent();
this.OnFormClosingAsync(Window_FormClosingAsync);
}
async Task Window_FormClosingAsync(object sender, FormClosingAsyncEventArgs e)
{
e.HideForm = true; // Optional
e.Timeout = 5000; // Optional
await KMApplication.License.ShutDown();
//e.Cancel = true; // Optional
}
The Window_FormClosingAsync handler will run in a thread-pool thread, so it should not include any UI manipulation code.
Unsubscribing from the event is possible, by keeping a reference of the IDisposable return value, and disposing it.
Update: After reading this answer, I realized that it is possible to add a real event FormClosingAsync in the form, without creating a class that inherits from the form. This can be achieved by adding the event, and then running an initialization method that hooks the event to the native FormClosing event. Something like this:
public event Func<object, FormClosingAsyncEventArgs, Task> FormClosingAsync;
public Form1()
{
InitializeComponent();
this.InitFormClosingAsync(() => FormClosingAsync);
this.FormClosingAsync += Window_FormClosingAsync_A;
this.FormClosingAsync += Window_FormClosingAsync_B;
}
Inside the initializer, in the internal handler of the native FormClosing event, all the subscribers of the event can be retrieved
using the GetInvocationList method:
var eventDelegate = handlerGetter();
if (eventDelegate == null) return;
var invocationList = eventDelegate.GetInvocationList()
.Cast<Func<object, FormClosingAsyncEventArgs, Task>>().ToArray();
...and then invoked appropriately. All this adds complexity, while the usefulness of allowing multiple handlers is debated. So I would probably stick with the original single-handler design.
Update: It is still possible to have multiple handlers using the original method OnFormClosingAsync. It is quite easy actually. The Func<T>
class inherits from Delegate, so it has invocation list like a real event:
Func<object, FormClosingAsyncEventArgs, Task> aggregator = null;
aggregator += Window_FormClosingAsync_A;
aggregator += Window_FormClosingAsync_B;
this.OnFormClosingAsync(aggregator);
No modification in the OnFormClosingAsync method is required.

Ok here is what I ended up doing. Basically the window closing kicks off a task that will wait for the release to happen and then invoke the shutdown. This is what I was trying to do before but it didn't seem to work in async void method but it seems to be when done this way. Here is the new handler:
void Window_Closing(object sender, System.ComponentModel.CancelEventArgs e)
{
...other sync code...
Task.Run(async () =>
{
await InvokeKmShutdown();
(Dispatcher ?? Dispatcher.CurrentDispatcher).InvokeShutdown();
});
}
And the shutdown method looks like this:
async Task InvokeKmShutdown()
{
...other sync code...
await KMApplication.ApiService.Disconnect();
//Check for floating licensing
if (KMApplication.License != null && KMApplication.License.Scope != License.Core.Enums.LicenseScope.Standalone)
{
for (int i = 0; i < 3; i++)
{
try
{
await KMApplication.License.ShutDown(KMApplication.Settings == null
? Enums.LicenseReturnModes.PromptOnShutdown
: KMApplication.Settings.LicenseReturnMode);
break;
}
catch (Exception ex)
{
_logger.Warn("Exception in license release, attempt " + i, ex);
}
}
}
}
Hope it helps someone.
EDIT
Note that this is with an WPF app set to ShutdownMode="OnExplicitShutdown" in App.xaml so it won't shut down the actual app until I call the shutdown. If you are using WinForms or WPF is set to shut down on last window or main window close (main window close is the default I believe) you will end up with the race condition described in the comments below and may get the threads shut down before things run to completion.

Cross-thread operation not valid: Control 'textbox' accessed from a thread other than the thread it was created on

I need some help. I started c# and not very familiar with event handling and threading yet. As a beginner and as time and exposure progresses, I would like to learn more on these advanced topics and improved and hope all of you here can help me.
I ran onto this problem of "Cross-thread operation not valid: Control 'textbox control called stackStatus' accessed from a thread other than the thread it was created on". I have tried to troubleshoot this whole day but simply no avail. I am stuck. :-( The program hits an exception and cannot continue to execute smoothly.
I have read the following threads and tried a few things but I guess I am still missing something. Appreciate if someone can help me out here. Thanks.
Cross-thread operation not valid: Control accessed from a thread other than the thread it was created on
Cross-thread operation not valid: Control 'textBox1' accessed from a thread other than the thread it was created on
Here's are most of the portion of the code:
private void createCloud_Click(object sender, EventArgs e)
{
CreateCloud(); //start creation method
stackStatus.Text = "Creating stack..."; //updates the cloud status textbox
stackStatus.Refresh();
Cursor.Current = Cursors.WaitCursor; //change the cursor to wait state
Start_Describestack(); //call describe method to find out the status of cloud creation progress
Task.Delay(12000); // wait 12s in case not ready
Start_Describestack(); // call again describe method to find out the cloud creation progress status
Cursor.Current = Cursors.Default; //put cursor on wait
describeevents(); // call method to get all cloud creation event data and publish on the datagridview
}
private void Start_Describestack()
{
//method making use of timer to call
_timer = new System.Timers.Timer(15000);
_timer.Elapsed += new ElapsedEventHandler(describeStack);
_timer.Enabled = true;
}
delegate void describeStackCallBack(object sender, ElapsedEventArgs e);
private void describeStack(object sender, ElapsedEventArgs e)
{
//this method makes api calls through cloudclient to describe the stack
//this is where the "Cross-thread operation not valid: Control 'stackStatus' accessed from a thread other than the thread it was created on"
var client = new cloudclient();
var request2 = new StacksRequest();
request2.Cloudstackname = stackid;
try
{
var response = client.DescribeCloudStacks(request2);
foreach (var stack in response.Stacks)
{
//something is wrong here but I do not know how to fix it. Please help
if (this.stackStatus.InvokeRequired)
{
describeStackCallBack d = new describeStackCallBack(describeStack);
this.Invoke(d, new object[] { sender, e });
stackStatus.Refresh();
describevents();
}
else
{
stackStatus.Text = stack.StackStatus;
stackStatus.Refresh();
describeevents();
}
}
}
catch (Exception)
{
if (this.stackStatus.InvokeRequired)
{
describeStackCallBack d = new describeStackCallBack(describeStack);
this.Invoke(d, new object[] { sender, e });
stackStatus.Text = "Stack not found/Deleted";
}
else
{ stackStatus.Text = "Stack not found/Deleted"; }
}
describeevents();
}
private void describeevents()
{
var newclient = new cloudclient();
var request3 = new eventrequest();
request3.Cloudstackname = stackid;
try
{
var response = newclient.eventstack(request3);
dataGridView3.Rows.Clear();
foreach (var events in response.sevents)
{
dataGridView3.Rows.Add(events.Timestamp, events.ResourceStatus, events.ResourceType);
}
}
catch (Exception)
{
dataGridView3.Rows.Clear();
MessageBox.Show("Stack not ready!");
}
dataGridView3.Refresh();
}

Rather than doing :
stackStatus.Text = "some text";
Try :
stackStatus.Invoke((Action)delegate
{
stackStatus.Text = "some text";
});
Note that GUI element assignment outside the thread or they are declared is deprecated because the controls may no longer be available at any time.

There are two issues in your approach, which conspire to prevent your attempt to imitate the solution to the exception from working:
You have failed to note that the proposed solution calls itself, and in so doing, causes the foreach to be restarted for each time it's invoked from the worker thread.
You are following Microsoft canonical implementation of cross-thread-friendly Invoke()-based code, which IMHO is lame.
It is my opinion that there is no point in ever checking InvokeRequired. The standard pattern always involves situations where on the first entry, you know you will require Invoke(), and even if you didn't, there's no real harm in calling Invoke() when it's not necessary.
Instead, you should always keep separate the code that should run in the UI thread, and the code that does not. Then, in the code that does not, always use Invoke() to execute the code that does.
For example:
private void Start_Describestack()
{
//method making use of timer to call
_timer = new System.Timers.Timer(15000);
_timer.Elapsed += new ElapsedEventHandler(_timer_Elapsed);
_timer.Enabled = true;
}
private void _timer_Elapsed(object sender, ElapsedEventArgs e)
{
Invoke((MethodInvoker)describeStack);
}
private void describeStack()
{
var client = new cloudclient();
var request2 = new StacksRequest();
request2.Cloudstackname = stackid;
try
{
var response = client.DescribeCloudStacks(request2);
foreach (var stack in response.Stacks)
{
stackStatus.Text = stack.StackStatus;
stackStatus.Refresh();
describeevents();
}
}
catch (Exception)
{
stackStatus.Text = "Stack not found/Deleted";
}
describeevents();
}
That said, an improvement on the above would be to use System.Windows.Forms.Timer instead of System.Timers.Timer. The latter raises the Elapsed event on a worker thread, but the former raises its event on the UI thread, right where you want it. No Invoke() required at all.
You have at least one other problem with your code as well:
private void createCloud_Click(object sender, EventArgs e)
{
CreateCloud(); //start creation method
stackStatus.Text = "Creating stack..."; //updates the cloud status textbox
stackStatus.Refresh();
Cursor.Current = Cursors.WaitCursor; //change the cursor to wait state
Start_Describestack(); //call describe method to find out the status of cloud creation progress
Task.Delay(12000); // wait 12s in case not ready
Start_Describestack(); // call again describe method to find out the cloud creation progress status
Cursor.Current = Cursors.Default; //put cursor on wait
describeevents(); // call method to get all cloud creation event data and publish on the datagridview
}
In the above, the call to Task.Delay(12000); accomplishes nothing. The Task.Delay() method doesn't actually block the current thread. Instead, it returns an awaitable task object. The code in which it appears only is delayed if you wait on the returned object.
It's also questionable to call Start_Describestack() twice, because this method doesn't do anything except start the timer. Calling it twice means now you have two timers running.
Finally, you should also not have all those calls to Refresh() in your code. Correctly written Windows Forms code will not need anything like that. Updates to control properties will cause control invalidation automatically, and the control will update as needed at its next opportunity, which as long as the code is written correctly, will be soon enough for the user to not notice any significant delay.
Now, putting all of the above together, it seems to me that you should avoid using the timer altogether. There is still the potential problem that your call to DescribeCloudStacks() is a lengthy one, and could cause the UI to momentarily appear "stuck", which obviously isn't a desirable thing. In addition, the timer-based code, whether you require Invoke() or not, can be harder to understand, especially for someone new to asynchronous programming and threading.
Using the async/await feature, you can write the code in a conventional, procedural way, while still ensuring that the UI remains responsive, and that the UI-related code is always executed in the UI thread where it belongs. That might look something like this:
private async void createCloud_Click(object sender, EventArgs e)
{
CreateCloud(); //start creation method
stackStatus.Text = "Creating stack..."; //updates the cloud status textbox
Cursor.Current = Cursors.WaitCursor; //change the cursor to wait state
await describeStack(); //call describe method to find out the status of cloud creation progress
await Task.Delay(12000); // wait 12s in case not ready
await describeStack(); // call again describe method to find out the cloud creation progress status
Cursor.Current = Cursors.Default; //put cursor on wait
describeevents(); // call method to get all cloud creation event data and publish on the datagridview
}
private async Task describeStack()
{
var client = new cloudclient();
var request2 = new StacksRequest();
request2.Cloudstackname = stackid;
try
{
var response = await Task.Run(() => client.DescribeCloudStacks(request2));
foreach (var stack in response.Stacks)
{
stackStatus.Text = stack.StackStatus;
describeevents();
}
}
catch (Exception)
{
stackStatus.Text = "Stack not found/Deleted";
}
describeevents();
}
The above executes most of the describeStacks() method in the UI thread. The exception would be the DescribeCloudStacks() method call, which is run as a worker task. While it's running, the UI thread is free to operate normally. Execution of the describeStacks() method is temporarily put "on hold" (without blocking the UI thread) while the worker task runs, and then is resumed when it completes.
It's not clear from your original example whether you really wanted a repeating timer or not. The above doesn't use any loops; it calls the describeStack() method only twice, with a 12-second delay in between. But if you want a loop, you can do that as well. Just use the await Task.Delay() for the delay and await describeStack() for the operation, and put that in a loop as you like.

I don't see where the stackStatus object is created so I'm just guessing that you are creating it through a contructor for the class containing describeStack() and then you are registering an event handler for the click. I think what is happening is the event handler is being run on a different thread from the one in which the instance was created so you might have to change how you create the stackStatus object. That error is likely happening because whatever type the stackStatus was created from is known to not be reentrant so when the runtime detects access between threads it raises an exception so you are aware and can either prevent or recover from race-conditions or deadlocks.

Deadlock when updating a UI control from a worker thread

To simplify the explanation of the strange behavior I am experiencing, I have this simple class named Log which fires 1 log events every 1000msec.
public static class Log
{
public delegate void LogDel(string msg);
public static event LogDel logEvent;
public static void StartMessageGeneration ()
{
for (int i = 0; i < 1000; i++)
{
logEvent.Invoke(i.ToString());
Task.Delay(1000);
}
}
}
I have the Form class below which is subscribed to the log events of the Log class so it can handle them and display in a simple text box.
Once a log message arrives, it is added to a list. Every 500msec, a timer object access that list so its content can be displayed in a text box.
public partial class Form1 : Form
{
private SynchronizationContext context;
private System.Threading.Timer guiTimer = null;
private readonly object syncLock = new object();
private List<string> listOfMessages = new List<string>();
public Form1()
{
InitializeComponent();
context = SynchronizationContext.Current;
guiTimer = new System.Threading.Timer(TimerProcessor, this, 0, 500);
Log.logEvent += Log_logEvent;
}
private void Log_logEvent(string msg)
{
lock (syncLock)
listOfMessages.Add(msg);
}
private void TimerProcessor(object obj)
{
Form1 myForm = obj as Form1;
lock (myForm.syncLock)
{
if (myForm.listOfMessages.Count == 0)
return;
myForm.context.Send(new SendOrPostCallback(delegate
{
foreach (string item in myForm.listOfMessages)
myForm.textBox1.AppendText(item + "\n");
}), null);
listOfMessages.Clear();
}
}
private void button1_Click(object sender, EventArgs e)
{
Log.StartMessageGeneration();
}
}
The problem I see is that sometimes, there is a dead lock (application stuck). Seems that the 2 locks (1st one for adding to the list and the 2nd one for "retrieving" from the list) are somehow blocking each others.
Hints:
1) reducing the rate of sending the messages from 1 sec to 200msec seems to help (not sure why)
2) Somehow something happens when returning to the GUI thread (using the synchronization context) and accessing the GUI control. If I don't return to the GUI thread, the 2 locks are working fine together...
Thanks everyone!

There's a few problems with your code, and a few... silly things.
First, your Log.StartMessageGeneration doesn't actually produce a log message every second, because you're not awaiting the task returned by Task.Delay - you're basically just creating a thousand timers very quickly (and pointlessly). The log generation is limited only by the Invoke. Using Thread.Sleep is a blocking alternative to Task.Delay if you don't want to use Tasks, await etc. Of course, therein lies your biggest problem - StartMessageGeneration is not asynchronous with respect to the UI thread!
Second, there's little point in using System.Threading.Timer on your form. Instead, just use the windows forms timer - it's entirely on the UI thread so there's no need for marshalling your code back to the UI thread. Since your TimerProcessor doesn't do any CPU work and it only blocks for a very short time, it's the more straight-forward solution.
If you decide to keep using System.Threading.Timer anyway, there's no point in manually dealing with synchronization contexts - just use BeginInvoke on the form; the same way, there's no point in passing the form as an argument to the method, since the method isn't static. this is your form. You can actually see this is the case since you omitted myForm in listOfMessages.Clear() - the two instances are the same, myForm is superfluous.
A simple pause in the debugger will easily tell you where the program is hung - learn to use the debugger well, and it will save you a lot of time. But let's just look at this logically. StartMessageGeneration runs on the UI thread, while System.Threading.Timer uses a thread-pool thread. When the timer locks syncLock, StartMessageGeneration can't enter the same lock, of course - that's fine. But then you Send to the UI thread, and... the UI thread can't do anything, since it's blocked by StartMessageGeneration, which never gives the UI an opportunity to do anything. And StartMessageGeneration can't proceed, because it's waiting on the lock. The only case where this "works" is when StartMessageGeneration runs fast enough to complete before your timer fires (thus freeing the UI thread to do its work) - which is very much possible due to your incorrect use of Task.Delay.
Now let's look on your "hints" with all we know. 1) is simply your bias in measurements. Since you never wait on the Task.Delay in any way, changing the interval does absolutely nothing (with a tiny change in case the delay is zero). 2) of course - that's where your deadlock is. Two pieces of code that depend on a shared resource, while they both require to take posession of another resource. It's a very typical case of a deadlock. Thread 1 is waiting for A to release B, and thread 2 is waiting for B to release A (in this case, A being syncLock and B being the UI thread). When you remove the Send (or replace it with Post), thread 1 no longer has to wait on B, and the deadlock disappears.
There's other things that make writing code like this simpler. There's little point in declaring your own delegate when you can just use Action<string>, for example; using await helps quite a bit when dealing with mixed UI/non-UI code, as well as managing any kind of asynchronous code. You don't need to use event where a simple function will suffice - you can just pass that delegate to a function that needs it if that makes sense, and it may make perfect sense not to allow multiple event handlers to be called. If you decide to keep with the event, at least make sure it conforms to the EventHandler delegate.
To show how your code can be rewritten to be a bit more up-to-date and actually work:
void Main()
{
Application.Run(new LogForm());
}
public static class Log
{
public static async Task GenerateMessagesAsync(Action<string> logEvent,
CancellationToken cancel)
{
for (int i = 0; i < 1000; i++)
{
cancel.ThrowIfCancellationRequested();
logEvent(i.ToString());
await Task.Delay(1000, cancel);
}
}
}
public partial class LogForm : Form
{
private readonly List<string> messages;
private readonly Button btnStart;
private readonly Button btnStop;
private readonly TextBox tbxLog;
private readonly System.Windows.Forms.Timer timer;
public LogForm()
{
messages = new List<string>();
btnStart = new Button { Text = "Start" };
btnStart.Click += btnStart_Click;
Controls.Add(btnStart);
btnStop =
new Button { Text = "Stop", Location = new Point(80, 0), Enabled = false };
Controls.Add(btnStop);
tbxLog = new TextBox { Height = 200, Multiline = true, Dock = DockStyle.Bottom };
Controls.Add(tbxLog);
timer = new System.Windows.Forms.Timer { Interval = 500 };
timer.Tick += TimerProcessor;
timer.Start();
}
private void TimerProcessor(object sender, EventArgs e)
{
foreach (var message in messages)
{
tbxLog.AppendText(message + Environment.NewLine);
}
messages.Clear();
}
private async void btnStart_Click(object sender, EventArgs e)
{
btnStart.Enabled = false;
var cts = new CancellationTokenSource();
EventHandler stopAction = (_, __) => cts.Cancel();
btnStop.Click += stopAction;
btnStop.Enabled = true;
try
{
await Log.GenerateMessagesAsync(message => messages.Add(message), cts.Token);
}
catch (TaskCanceledException)
{
messages.Add("Cancelled.");
}
finally
{
btnStart.Enabled = true;
btnStop.Click -= stopAction;
btnStop.Enabled = false;
}
}
protected override void Dispose(bool disposing)
{
if (disposing)
{
timer.Dispose();
btnStart.Dispose();
btnStop.Dispose();
tbxLog.Dispose();
}
base.Dispose(disposing);
}
}

SynchronizationContext.Send is run synchronously. When you call it, you actually block the UI thread until the operation is complete. But if UI thread is already in lock state, then it just make sense that you are in deadlock.
You can use SynchronizationContext.Post to avoid this.
I just answer on your question, but the truth is that your code need a "little" refactoring..

UI Thread .Invoke() causing handle leak?

In what circumstances would updating a UI control from a non-UI thread could cause the processes' handles to continually increase, when using a delegate and .InvokeRequired?
For example:
public delegate void DelegateUIUpdate();
private void UIUpdate()
{
if (someControl.InvokeRequired)
{
someControl.Invoke(new DelegateUIUpdate(UIUpdate));
return;
}
// do something with someControl
}
When this is called in a loop or on timer intervals, the handles for the program consistently increase.
EDIT:
If the above is commented out and amended as such:
public delegate void DelegateUIUpdate();
private void UIUpdate()
{
//if (someControl.InvokeRequired)
//{
// someControl.Invoke(new DelegateUIUpdate(UIUpdate));
// return;
//}
CheckForIllegalCrossThreadCalls = false;
// do something with someControl
}
...then the handles stop incrementing, however I don't want to allow cross thread calls, of course.
EDIT 2:
Here is a sample that shows the handles increase:
Thread thread;
private delegate void UpdateGUI();
bool UpdateTheGui = false;
public Form1()
{
InitializeComponent();
thread = new Thread(new ThreadStart(MyThreadLoop));
thread.Start();
}
private void MyThreadLoop()
{
while (true)
{
Thread.Sleep(500);
if (UpdateTheGui)
{
UpdateTheGui = false;
UpdateTheGuiNow();
}
}
}
private void UpdateTheGuiNow()
{
if (label1.InvokeRequired)
{
label1.Invoke(new UpdateGUI(UpdateTheGuiNow));
return;
}
label1.Text = DateTime.Now.ToString("MM-dd-yyyy HH:mm:ss");
label2.Text = DateTime.Now.ToString("MM-dd-yyyy HH:mm:ss");
label3.Text = DateTime.Now.ToString("MM-dd-yyyy HH:mm:ss");
}
private void btnInvoke_Click(object sender, EventArgs e)
{
UpdateTheGui = true;
}

I had the same problem with
this.Invoke(new DelegateClockUpdate(ChangeClock), sender, e);
creating one handle each call.
The handle increments because Invoke is Synchronous and effectively the handle has been left hanging.
Either a Wait Handle should be used to process the result or the Asynchronous BeginInvoke method as shown below.
this.BeginInvoke(new DelegateClockUpdate(ChangeClock), sender, e);

The Control.Invoke() method doesn't consume any handles. However, this code is clearly called from a thread. A Thread does consume handles, 5 of them.
The Thread class doesn't have a Dispose() method, although it ought to have one. That was probably by design, it would be very difficult to call reliably, impossibly so for threadpool threads. The 5 handles that a thread requires are released by the finalizer. Your program will require ever increasing amounts of handles if the finalizer never runs.
Not getting the finalizer to run is quite unusual. You would have to have a program that starts a lot of threads but doesn't allocate a lot of memory. This tends to only happen in static tests. You can diagnose this condition with Perfmon.exe, use the .NET memory performance counters and check if gen #0 collections are being done.
If this happens in a production program then you'll have to call GC.Collect() yourself to avoid a runaway handle leak.

I've seen the same thing in my code. I fixed it by replacing Invoke with BeginInvoke. The handle leak went away.
Doron.

I actually see the same problem occuring as JYelton. I have the same call from within a thread to update the UI.
As soon as the line someControl.Invoke(new DelegateUIUpdate(UIUpdate)); is called, the handle increases by one. There is certainly a leak of some kind on the invoke, but I have no idea what is causing it. This has been verified on several systems.

Aync call with explicit handle finalize. Exapmle:
public static class ActionExtensions
{
private static readonly ILog log = LogManager.GetLogger(typeof(ActionExtensions));
/// <summary>
/// Async exec action.
/// </summary>
/// <param name="action">Action.</param>
public static void AsyncInvokeHandlers(
this Action action)
{
if (action == null)
{
return;
}
foreach (Action handler in action.GetInvocationList())
{
// Initiate the asychronous call. Include an AsyncCallback
// delegate representing the callback method, and the data
// needed to call EndInvoke.
handler.BeginInvoke(
ar =>
{
try
{
// Retrieve the delegate.
var handlerToFinalize = (Action)ar.AsyncState;
// Call EndInvoke to free resources.
handlerToFinalize.EndInvoke(ar);
var handle = ar.AsyncWaitHandle;
if (handle.SafeWaitHandle != null && !handle.SafeWaitHandle.IsInvalid && !handle.SafeWaitHandle.IsClosed)
{
((IDisposable)handle).Dispose();
}
}
catch (Exception exception)
{
log.Error("Async Action exec error.", exception);
}
},
handler);
}
}
}
See http://msdn.microsoft.com/en-us/library/system.iasyncresult.asyncwaithandle.aspx note:
When you use the BeginInvoke method of a delegate to call a method asynchronously and obtain a wait handle from the resulting IAsyncResult, we recommend that you close the wait handle as soon as you are finished using it, by calling the WaitHandle.Close method. If you simply release all references to the wait handle, system resources are freed when garbage collection reclaims the wait handle, but garbage collection works more efficiently when disposable objects are explicitly closed or disposed. For more information, see the AsyncResult.AsyncWaitHandle property.

Here's an extension method which functions similarly to the normal Invoke call, but will clean up the handle after:
namespace ExtensionMethods
{
public static class ExtensionMethods
{
public static void InvokeAndClose(this Control self, MethodInvoker func)
{
IAsyncResult result = self.BeginInvoke(func);
self.EndInvoke(result);
result.AsyncWaitHandle.Close();
}
}
}
You can then call it very similarly to a normal invoke:
myForm.InvokeAndClose((MethodInvoker)delegate
{
someControl.Text = "New Value";
});
It will block and wait for the delegate to execute, then close the handle before returning.

This is the standard pattern for using Invoke to marshall updates to the UI thread.
Are you sure your problem is not being caused by other code in your application that is not included in your question?

I don't think it is related. Perhaps just waiting for the garbage collector to dispose the newly allocated object(s) inside Invoke().

UI update with multiple concurrent operations

I am developing an application in C# using National Instruments Daqmx for performing measurements on certain hardware.
My setup consists of several detectors from which I have to get data during a set period of time, all the while updating my UI with this data.
public class APD : IDevice
{
// Some members and properties go here, removed for clarity.
public event EventHandler ErrorOccurred;
public event EventHandler NewCountsAvailable;
// Constructor
public APD(
string __sBoardID,
string __sPulseGenCtr,
string __sPulseGenTimeBase,
string __sPulseGenTrigger,
string __sAPDTTLCounter,
string __sAPDInputLine)
{
// Removed for clarity.
}
private void APDReadCallback(IAsyncResult __iaresResult)
{
try
{
if (this.m_daqtskRunningTask == __iaresResult.AsyncState)
{
// Get back the values read.
UInt32[] _ui32Values = this.m_rdrCountReader.EndReadMultiSampleUInt32(__iaresResult);
// Do some processing here!
if (NewCountsAvailable != null)
{
NewCountsAvailable(this, new EventArgs());
}
// Read again only if we did not yet read all pixels.
if (this.m_dTotalCountsRead != this.m_iPixelsToRead)
{
this.m_rdrCountReader.BeginReadMultiSampleUInt32(-1, this.m_acllbckCallback, this.m_daqtskAPDCount);
}
else
{
// Removed for clarity.
}
}
}
catch (DaqException exception)
{
// Removed for clarity.
}
}
private void SetupAPDCountAndTiming(double __dBinTimeMilisec, int __iSteps)
{
// Do some things to prepare hardware.
}
public void StartAPDAcquisition(double __dBinTimeMilisec, int __iSteps)
{
this.m_bIsDone = false;
// Prepare all necessary tasks.
this.SetupAPDCountAndTiming(__dBinTimeMilisec, __iSteps);
// Removed for clarity.
// Begin reading asynchronously on the task. We always read all available counts.
this.m_rdrCountReader.BeginReadMultiSampleUInt32(-1, this.m_acllbckCallback, this.m_daqtskAPDCount);
}
public void Stop()
{
// Removed for clarity.
}
}
The object representing the detector basically calls a BeginXXX operation with a callback that holds the EndXXX en also fires an event indicating data available.
I have up to 4 of these detector objects as members of my UI form. I call the Start() method on all of them in sequence to start my measurement. This works and the NewCountsAvailable event fires for all four of them.
Due to the nature of my implementation, the BeginXXX method is called on the UI thread and the Callback and the Event are also on this UI thread. Therefore I cannot use some kind of while loop inside my UI thread to constantly update my UI with the new data because the events constantly fire (I tried this). I also do not want to use some kind of UpdateUI() method in each of the four NewCountsAvailable eventhandlers since this will load my system too much.
Since I am new to threaded programming in C# I am now stuck;
1) What is the "proper" way to handle a situation like this?
2) Is my implementation of the detector object sound? Should I call the Start() methods on these four detector objects from yet another thread?
3) Could I use a timer to update my UI every few hundred miliseconds, irrespective of what the 4 detector objects are doing?
I really have no clue!

I'd use a simple deferred update system.
1) Worker threads signal "data ready" by raising an event
2) UI thread listens for the event. When it is received, it just sets a "data needs updating" flag and returns, so minimal processing occurs on the event itself.
3) UI thread uses a timer (or sits on Application.Idle events) to check the "data needs updating" flag and, if necessary, update the UI. In many cases, UI only needs to be updated once or twice a second, so this need not burn a lot of CPU time.
This allows the UI to continue running as normal all the time (remaining interactive for the user), but within a short period of some data being ready, it is displayed in the UI.
Additionally, and most importantly for good UI, this approach can be used to allow multiple "data ready" events to fire and be rolled into a single UI update. This means that if 10 pieces of data are completed in close succession, the UI updates once rather than your window flickering for several seconds as the UI redraws (unnecessarily) 10 times.

I would try moving the IDevice monitoring logic to seperate threads for each device. The UI can then poll for values via a timer event, button click or some other UI related event. That way your UI will remain responsive and your threads are doing all the heavy lifting. Here's a basic example of this using a continuous loop. Obviously, this is a brutally simple example.
public partial class Form1 : Form
{
int count;
Thread t = null;
public Form1()
{
InitializeComponent();
}
private void ProcessLogic()
{
//CPU intensive loop, if this were in the main thread
//UI hangs...
while (true)
{
count++;
}
}
private void Form1_Load(object sender, EventArgs e)
{
//Cannot directly call ProcessLogic, hangs UI thread.
//ProcessLogic();
//instead, run it in another thread and poll needed values
//see button1_Click
t = new Thread(ProcessLogic);
t.Start();
}
private void Form1_FormClosing(object sender, FormClosingEventArgs e)
{
t.Abort();
}
private void button1_Click(object sender, EventArgs e)
{
button1.Text = count.ToString();
}
}

Some updates to reflect the new data you've provided:
Although I have my doubts that your EndXXX methods are happening on the UI thread, I still think you should spawn off the work to a background thread and then update the UI either as events are fired or as needed.
Because you've added a tight while loop in your UI, you need to call Application.DoEvents to allow your other events to be called.
Here's an updated sample that shows results in the UI as they occur:
public class NewCountArgs : EventArgs
{
public NewCountArgs(int count)
{
Count = count;
}
public int Count
{
get; protected set;
}
}
public class ADP
{
public event EventHandler<NewCountArgs> NewCountsAvailable;
private double _interval;
private double _steps;
private Thread _backgroundThread;
public void StartAcquisition(double interval, double steps)
{
_interval = interval;
_steps = steps;
// other setup work
_backgroundThread = new Thread(new ThreadStart(StartBackgroundWork));
_backgroundThread.Start();
}
private void StartBackgroundWork()
{
// setup async calls on this thread
m_rdrCountReader.BeginReadMultiSampleUInt32(-1, Callback, _steps);
}
private void Callback(IAsyncResult result)
{
int counts = 0;
// read counts from result....
// raise event for caller
if (NewCountsAvailable != null)
{
NewCountsAvailable(this, new NewCountArgs(counts));
}
}
}
public class Form1 : Form
{
private ADP _adp1;
private TextBox txtOutput; // shows updates as they occur
delegate void SetCountDelegate(int count);
public Form1()
{
InitializeComponent(); // assume txtOutput initialized here
}
public void btnStart_Click(object sender, EventArgs e)
{
_adp1 = new ADP( .... );
_adp1.NewCountsAvailable += NewCountsAvailable;
_adp1.StartAcquisition(....);
while(!_adp1.IsDone)
{
Thread.Sleep(100);
// your NewCountsAvailable callbacks will queue up
// and will need to be processed
Application.DoEvents();
}
// final work here
}
// this event handler will be called from a background thread
private void NewCountsAvailable(object sender, NewCountArgs newCounts)
{
// don't update the UI here, let a thread-aware method do it
SetNewCounts(newCounts.Count);
}
private void SetNewCounts(int counts)
{
// if the current thread isn't the UI thread
if (txtOutput.IsInvokeRequired)
{
// create a delegate for this method and push it to the UI thread
SetCountDelegate d = new SetCountDelegate(SetNewCounts);
this.Invoke(d, new object[] { counts });
}
else
{
// update the UI
txtOutput.Text += String.Format("{0} - Count Value: {1}", DateTime.Now, counts);
}
}
}

I don't know if I fully understand. What if you update you an object that contains the current data. So the callback don't directly interact with the UI. Then you could update the UI at a fixed rate, e.g. n times per second from another thread. See this post on updating UI from a background thread. I am assuming that you are using Windows Forms and not WPF.

The B* * *dy captcha system decided it was a good idea to lose my answer I spent half an hour typing without so much as a warning or a chance to correct... so here we go again:
public class APD : IDevice
{
// Some members and properties go here, removed for clarity.
public event EventHandler ErrorOccurred;
public event EventHandler NewCountsAvailable;
public UInt32[] BufferedCounts
{
// Get for the _ui32Values returned by the EndReadMultiSampleUInt32()
// after they were appended to a list. BufferdCounts therefore supplies
// all values read during the experiment.
}
public bool IsDone
{
// This gets set when a preset number of counts is read by the hardware or when
// Stop() is called.
}
// Constructor
public APD( some parameters )
{
// Removed for clarity.
}
private void APDReadCallback(IAsyncResult __iaresResult)
{
try
{
if (this.m_daqtskRunningTask == __iaresResult.AsyncState)
{
// Get back the values read.
UInt32[] _ui32Values = this.m_rdrCountReader.EndReadMultiSampleUInt32(__iaresResult);
// Do some processing here!
if (NewCountsAvailable != null)
{
NewCountsAvailable(this, new EventArgs());
}
// Read again only if we did not yet read all pixels.
if (this.m_dTotalCountsRead != this.m_iPixelsToRead)
{
this.m_rdrCountReader.BeginReadMultiSampleUInt32(-1, this.m_acllbckCallback, this.m_daqtskAPDCount);
}
else
{
// Removed for clarity.
}
}
}
catch (DaqException exception)
{
// Removed for clarity.
}
}
private void SetupAPDCountAndTiming(double __dBinTimeMilisec, int __iSteps)
{
// Do some things to prepare hardware.
}
public void StartAPDAcquisition(double __dBinTimeMilisec, int __iSteps)
{
this.m_bIsDone = false;
// Prepare all necessary tasks.
this.SetupAPDCountAndTiming(__dBinTimeMilisec, __iSteps);
// Removed for clarity.
// Begin reading asynchronously on the task. We always read all available counts.
this.m_rdrCountReader.BeginReadMultiSampleUInt32(-1, this.m_acllbckCallback, this.m_daqtskAPDCount);
}
public void Stop()
{
// Removed for clarity.
}
}
Note I added some things I mistakenly left out in the original post.
Now on my form I have code like this;
public partial class Form1 : Form
{
private APD m_APD1;
private APD m_APD2;
private APD m_APD3;
private APD m_APD4;
private DataDocument m_Document;
public Form1()
{
InitializeComponent();
}
private void Button1_Click()
{
this.m_APD1 = new APD( ... ); // times four for all APD's
this.m_APD1.NewCountsAvailable += new EventHandler(m_APD1_NewCountsAvailable); // times 4 again...
this.m_APD1.StartAPDAcquisition( ... );
this.m_APD2.StartAPDAcquisition( ... );
this.m_APD3.StartAPDAcquisition( ... );
this.m_APD4.StartAPDAcquisition( ... );
while (!this.m_APD1.IsDone) // Actually I have to check all 4
{
Thread.Sleep(200);
UpdateUI();
}
// Some more code after the measurement is done.
}
private void m_APD1_NewCountsAvailable(object sender, EventArgs e)
{
this.m_document.Append(this.m_APD1.BufferedCounts);
}
private void UpdateUI()
{
// use the data contained in this.m_Document to fill the UI.
}
}
phew, I hope I dod not forget anything yping this a second time (that'll teach me not copying it before hitting Post).
What I see running this code is that;
1) The APD object works as advertised, it measures.
2) The NewCountsAvailable events fire and their handlers get executed
3) APD.StartAPDAcquisition() is called on the UI thread. Thus also BeginXXX is called on this thread. Therefore, by design, the callback is also on this thread and obviously also the NewCountsAvailable eventhandlers run on the UI thread. The only thing that is not on the UI thread is waiting for the hardware to return values to the BeginXXX EndXXX pair of calls.
4) Because the NewCountsAvailable events fire quite a lot, the while loop I intended to use for updating the UI does not run. Typically it runs once in the beginning and then somehow gets interupted by the eventhandlers that need to process. I do not fully understand this though, but it does not work...
I was thinking to solve this by getting rid of the while loop and putting a Forms.Timer on the form where UpdateUI() would be called from the Tick eventhandler. However, I do not know if this would be deemed "best practice". I also do not know if all these eventhandlers will eventually bring the UI thread to a crawl, I might need to add a few more of these APD objects in the future. Also UpdateUI() might contain some heavier code for calculating an image based on the values in m_Document. So the tick eventhandler might also be a resource drain in the timer approach. In case I use this solution I would also need to have a "Done" event in my APD class to notify when each APD finishes.
Should I perhaps not be working with events for notifying that new counts are available but instead work with some kind of "on demand" reading of APD.BufferedCounts and put the whole thing in yet another thread? I really haven't a clue...
I basically need a clean, lightweight solution that scales well should I add yet more APD's :)