I have this simple code :
var g= Task.Factory.StartNew<int> (() => 8)
.ContinueWith (ant =>{throw null;})
.ContinueWith (a =>{ Console.WriteLine("OK");},TaskContinuationOptions.NotOnFaulted);
try{
Console.WriteLine("1");
g.Wait();
Console.WriteLine("2");
}
catch (AggregateException ex)
{Console.WriteLine("catch"); }
The Output :
1
catch
System.AggregateException: A Task's exception(s) were not observed either by Waiting on the Task or accessing its Exception property. As a result, the unobserved exception was rethrown by the finalizer thread.
msdn :
TaskContinuationOptions.NotOnFaulted
Specifies that the continuation task should not be scheduled if its
antecedent threw an unhandled exception. This option is not valid for
multi-task continuations.
ok .
And it is ok - not showing this line cause the prev line DID throw exception.
Questions :
Do I get the AggregateException exception because I haven't inspected the Exception property ?
Must I always inspect if the antecedent throw an exception ( in each line ? ) ? ( I can't check each line ! it doesn't make any sense and very annoying)
Wasn't the try catch block should have swallow the exception ? ( I thought that all exceptions bubble up to the wait method....so ? )
Do I get the AggregateException exception because I haven't inspected
the Exception property ?
No, you get an exception, because task g cancels by TPL(because, as msdn stated, this task will not scheduled if antescendent task throws an exception).
We have 3 tasks here:
Original Task (that uses StartNew)
First Continuation Task (that throws an exception)
Second Continuation Task (that prints OK) (this is g task from your code).
The issue is that you ask TPL to start 3d task only if 2nd task will finished successfully. This means that if this condition will not met TPL will cancel your newly created task entirely.
You got unobserved task exception because you have temporary task (task 2 in my list) that you never observe. An because you never observe it faulted state it will throw in finalizer to tell you about it.
You can check this by printing task's status in catch block:
catch (AggregateException ex)
{
Console.WriteLine("catch");
// Will print: Status in catch: Canceled
Console.WriteLine("Status in catch: {0}", g.Status);
}
Must I always inspect if the antecedent throw an exception ( in each
line ? ) ? ( I can't check each line ! it doesn't make any sense and
very annoying)
Yes you should observe antecedent tasks exception to avoid this issue:
static class TaskEx
{
public static Task ObserverExceptions(this Task task)
{
task.ContinueWith(t => { var ignore = t.Exception; },
TaskContinuationOptions.OnlyOnFaulted);
return task;
}
}
And then use it as following:
var g= Task.Factory.StartNew<int> (() => 8)
.ContinueWith (ant =>{throw null;})
.ObserveExceptions()
.ContinueWith (a =>{ Console.WriteLine("OK");});
try{
Console.WriteLine("1");
g.Wait();
Console.WriteLine("2");
}
catch (AggregateException ex)
{Console.WriteLine("catch"); }
UPDATE: Added solution to last bullet
Wasn't the try catch block should have swallow the exception ? ( I
thought that all exceptions bubble up to the wait method....so ? )
We have set of extension method (called TransformWith) in our project that can solve this particular issue and gain following:
Exception would bubble up to the catch block and
We'll not crash application with TaskUnobservedException
Here the usage
var g = Task.Factory.StartNew(() => 8)
.ContinueWith(ant => { throw null; })
// Using our extension method instead of simple ContinueWith
.TransformWith(t => Console.WriteLine("OK"));
try
{
Console.WriteLine("1");
// Will fail with NullReferenceException (inside AggregateExcpetion)
g.Wait();
Console.WriteLine("2");
}
catch (AggregateException ex)
{
// ex.InnerException is a NullReferenceException
Console.WriteLine(ex.InnerException);
}
And here is a extension method:
static class TaskEx
{
public static Task TransformWith(this Task future, Action<Task> continuation)
{
var tcs = new TaskCompletionSource<object>();
future
.ContinueWith(t =>
{
if (t.IsCanceled)
{
tcs.SetCanceled();
}
else if (t.IsFaulted)
{
tcs.SetException(t.Exception.InnerExceptions);
}
else
{
try
{
continuation(future);
tcs.SetResult(null);
}
catch (Exception e)
{
tcs.SetException(e);
}
}
}, TaskContinuationOptions.ExecuteSynchronously);
return tcs.Task;
}
}
Do I get the AggregateException exception because I haven't inspected
the Exception property ?
Tasks always throw AggregateException : http://msdn.microsoft.com/en-us/library/system.threading.tasks.task.exception.aspx
You can get the original exception using :
var myTask = Task.Factory.StartNew(() => { throw new NotImplementedException(); });
var myException = myTask.Exception.Flatten().InnerException as NotImplementedException;
Must I always inspect if the antecedent throw an exception ( in each
line ? ) ? ( I can't check each line ! it doesn't make any sense and
very annoying)
Yes it is anoying, you should create two continuations for each task to check exceptions : one that checks if there has been an exception to handle it, and another one to continue the operation if there was no exception see TaskContinuationOptions.OnlyOnFaulted and TaskContinuationOptions.OnlyOnRanToCompletion.
You should even create a third continuation to deal with cancellation if needed.
Wasn't the try catch block should have swallow the exception ? ( I
thought that all exceptions bubble up to the wait method....so ? )
No it won't, exceptions are not thrown at higher level, you should use TaskContinuationOptions.OnlyOnFaulted on the task continuation to check if there was an exception. You can get tasks exceptions at caller's level only with the async keyword not available in .net 4
Handle AggregateExceptions like this:
catch(AggregateException aex)
{
aex.Handle(ex =>
{
// Do some handling and logging
return true;
}
}
Related
Problem
Several tasks are run in parallel, and all, none, or any of them might throw exceptions. When all the tasks have finalized, all the exceptions that might have happened must be reported (via log, email, console output.... whatever).
Expected behavior
I can build all the tasks via linq with async lambdas, and then await for them running in parallel with Task.WhenAll(tasks). Then I can catch an AggregateException and report each of the individual inner exceptions.
Actual behavior
An AggregateException is thrown, but it contains just one inner exception, whatever number of individual exceptions have been thrown.
Minimal complete verifiable example
static void Main(string[] args)
{
try
{
ThrowSeveralExceptionsAsync(5).Wait();
}
catch (AggregateException ex)
{
ex.Handle(innerEx =>
{
Console.WriteLine($"\"{innerEx.Message}\" was thrown");
return true;
});
}
Console.ReadLine();
}
private static async Task ThrowSeveralExceptionsAsync(int nExceptions)
{
var tasks = Enumerable.Range(0, nExceptions)
.Select(async n =>
{
await ThrowAsync(new Exception($"Exception #{n}"));
});
await Task.WhenAll(tasks);
}
private static async Task ThrowAsync(Exception ex)
{
await Task.Run(() => {
Console.WriteLine($"I am going to throw \"{ex.Message}\"");
throw ex;
});
}
Output
Note that the output order of the "I am going to throw" messages might change, due to race conditions.
I am going to throw "Exception #0"
I am going to throw "Exception #1"
I am going to throw "Exception #2"
I am going to throw "Exception #3"
I am going to throw "Exception #4"
"Exception #0" was thrown
That's because await "unwraps" aggregate exceptions and always throws just first exception (as described in documentation of await), even when you await Task.WhenAll which obviously can result in multiple errors. You can access aggregate exception for example like this:
var whenAll = Task.WhenAll(tasks);
try {
await whenAll;
}
catch {
// this is `AggregateException`
throw whenAll.Exception;
}
Or you can just loop over tasks and check status and exception of each.
Note that after that fix you need to do one more thing:
try {
ThrowSeveralExceptionsAsync(5).Wait();
}
catch (AggregateException ex) {
// flatten, unwrapping all inner aggregate exceptions
ex.Flatten().Handle(innerEx => {
Console.WriteLine($"\"{innerEx.Message}\" was thrown");
return true;
});
}
Because task returned by ThrowSeveralExceptionsAsync contains AggregateException we thrown, wrapped inside another AggregateException.
In the method below, when an exception is thrown in the TRY block, it is being swallowed. How can I make it throw the exception so that it gets written to log in the catch block? The log writer works fine. Thanks!
public static bool MonitorQueueEmptyTask(string queueName, CancellationTokenSource tokenSource)
{
try
{
Task<bool> task = Task.Factory.StartNew<bool>(() =>
{
while (!QueueManager.IsQueueEmpty(queueName))
{
if (tokenSource.IsCancellationRequested)
{
break;
}
Thread.Sleep(5000);
throw new Exception("Throwing an error!"); //THIS THROW IS SWALLOWED -- NO LOG WRITTEN ON CATCH
};
return true;
}, tokenSource.Token);
}
catch (Exception ex)
{
WriteExceptionToLog(ex.Stack); //it's not that this method doesn't work. it works fine.
return false;
}
return true;
}
If you want to fire and forget, you can attach a continuation using ContinueWith. The current try-catch will not help you at all, as the exception is encapsulated inside the Task. If this is "fire and forget", than you can log the exception:
public static Task MonitorQueueEmptyTask(
string queueName, CancellationTokenSource tokenSource)
{
return Task.Factory.StartNew<bool>(() =>
{
while (!QueueManager.IsQueueEmpty(queueName))
{
if (tokenSource.IsCancellationRequested)
{
break;
}
Thread.Sleep(5000);
throw new Exception("Throwing an error!");
};
}, tokenSource.Token, TaskCreationOptions.LongRunning).ContinueWith(faultedTask =>
{
WriteExceptionToLog(faultedTask.Exception);
}, TaskContinuationOptions.OnlyOnFaulted);
}
This, in turn, will not propagate the exception after it's thrown, but will provide a mechanism to log the error. If you want the exception to be properly handled, you can register to TaskScheduler.UnobservedTaskException. Additionally, you can set ThrowUnobservedTaskExceptions enabled="true" in your configuration if you want unhandled exceptions to terminate your application. ContinueWith will consider the exception "handled" once you look at the task.Exception property.
The exception is not swallowed; it's just that it doesn't occur on the thread that executes the try/catch block, but on the separate Task thread.
If you don't observe the task's result or exception, when the task is eventually garbage collected, it will throw an exception saying that the task was not observed. Unless you catch that by handling the TaskScheduler.UnobservedTaskException, it will crash the process.
I also had a problem with this, and i really dislike the whole idea of App.config, so can provide another solution to prevent the exceptions disappearing :)
Save the exception then throw it after the Task.Run has completed, e.g.
private async void Function() {
Exception save_exception = null;
await Task.Run(() => {
try {
// Do Stuff
} catch (Exception ex) {
save_exception = ex;
}
}).ContinueWith(new Action<Task>(task => {
if (save_exception != null)
throw save_exception;
// Do Stuff
}));
}
I am in a strange situation, I am using task and continue with upon faulted i am calling one method to process upon faulted or success but the function does not get fired.
Below is my code but upon expcetion it does not executes the UdpateResult method, what I am missing here.
var task = new Task.Factory.StartNew(SomeMethod());
task.ContinueWith(
t1 => Handler.UpdateResult(t1.Result, t1.Exception),
TaskContinuationOptions.ExecuteSynchronously);
try
{
task.Wait();
}
catch (AggregateException exception)
{
foreach (var innerException in exception.Flatten().InnerExceptions)
{
if (innerException is InvalidOperationException)
{
throw innerException;
}
throw new InvalidOperationException(string.Empty, innerException);
}
}
You're trying to use t1.Result within your continuation. If t1 has faulted, then accessing t1.Result will itself throw an exception.
You either need a single continuation which can handle any kind of end result, or you should attach different continuations for success and failure cases.
I have read a lot on how to handle exceptions in TPL but don't really understand.
Lets take this example code:
var task1 = new Task(() => { throw new Exception("Throw 1"); });
var task2 = task1.ContinueWith(t => Console.WriteLine("Catch 1:{0}", t.Exception.Message),
TaskContinuationOptions.OnlyOnFaulted);
var task3 = task2.ContinueWith(t => Console.WriteLine("Continuation"));
task1.Start();
try {
task1.Wait();
}
catch (Exception ex) {
Console.WriteLine("Wait Exception: {0}", ex.Message);
}
I expected this to print
Catch 1
Continuation
But I get
Catch 1
Continuation
Wait Exception
This means that the exception is still considered unhandled when the task completes and the task finalizer will eventually tear down the application.
How do I handle the exception inside the continuation so the finalizer will not throw? At the same time I want the task to remain in the faulted state so wrapping the task in try/catch will not work.
The background is that I want to implement the async event pattern as specified here but with error handling. My complete code looks like this
public IAsyncResult Begin(AsyncCallback callback, object state, Action action) {
var task1 = new Task(action);
var task2 = task1.ContinueWith(t => HandleException(t.Exception),
TaskContinuationOptions.OnlyOnFaulted);
if (callback != null) {
var task3 = task2.ContinueWith(t => callback(t),
TaskScheduler.FromCurrentSynchronizationContext());
var task4 = task3.ContinueWith(t => HandleException(t.Exception),
TaskContinuationOptions.OnlyOnFaulted);
}
task1.Start();
return task;
}
You do your wait on the task that fails, and if you read the documentation on Task.Wait carefully you will see that wait will rethrow the exception in this case.
But if you wait on your task3 everything should work as expected.
Of course you should keep this in mind:
When you use the OnlyOnFaulted option, it is guaranteed that the
Exception property in the antecedent is not null. You can use that
property to catch the exception and see which exception caused the
task to fault. If you do not access the Exception property, the
exception will go unhandled. Also, if you attempt to access the Result
property of a task that has been canceled or has faulted, a new
exception will be raised.
(Reference here)
And finally yet another good source on How to handle exceptions thrown by tasks
I hope this helps.
I'm looking at the TPL exception handling example from MSDN #
http://msdn.microsoft.com/en-us/library/dd537614(v=VS.100).aspx
The basic form of the code is:
Task task1 = Task.Factory.StartNew(() => { throw new IndexOutOfRangeException(); });
try
{
task1.Wait();
}
catch (AggregateException ae)
{
throw ae.Flatten();
}
My question is: Is this a race condition? What happens if task1 throws before the try has executed? Am I missing something that stops this being a race?
Shouldn't it be written like this instead:
try
{
Task task1 = Task.Factory.StartNew(() => { throw new IndexOutOfRangeException(); });
task1.Wait();
}
catch (AggregateException ae)
{
throw ae.Flatten();
}
No, the first example is perfectly valid.
When the exception is raised in the Task it is wrapped in an AggregateException. Only when another thread joins the task, in this example by calling task1.Wait() is the exception propogated to the joining thread. Essentially the exception is 'stored' until it can be propogated back to a thread that is waiting for the feedback.