How to Catch Exceptions for an array of Tasks - c#

I would like to understand how what I should be using inside of my "try" block when trying to await an Array of Tasks.
I want all tasks to be awaited, regardless if one of them threw an Exception, so that they can all complete.
Should I use:
var tasks = new Task<CasApiRouterModelExtendedInfo>[mbis.Length];
for (int i = 0; i < mbis.Length; i++)
{
tasks[i] = CAS.Service.GetAllRouterInterfacesAsync(mbis[i], false, 2);
}
try
{
Task.WaitAll(tasks);
}
catch (AggregateException ex)
{
Trace.TraceError("Some interface discoveries failed: ");
foreach (var innerEx in ex.InnerExceptions)
{
Trace.TraceError(innerEx.Message);
}
}
foreach (var task in tasks)
{
if (task.Status == TaskStatus.RanToCompletion && task.Result != null)
{
returnResults.Add(task.Result);
}
}
OR
var tasks = new Task<CasApiRouterModelExtendedInfo>[mbis.Length];
for (int i = 0; i < mbis.Length; i++)
{
tasks[i] = Task.Run(() => CAS.Service.GetAllRouterInterfacesAsync(mbis[i], true, 2));
}
try
{
for (int i = 0; i < tasks.Length; i++)
{
tasks[i].Wait();
}
}
catch (AggregateException ex)
{
Trace.TraceError("Some interface discoveries failed: ");
foreach (var innerEx in ex.InnerExceptions)
{
Trace.TraceError(innerEx.Message);
}
}
foreach (var task in tasks)
{
if (task.Status == TaskStatus.RanToCompletion && task.Result != null)
{
returnResults.Add(task.Result);
}
}
Also, does that "task.Status == TaskStatus.RanToCompletion" return true as long as the task didn't throw an Exception (is this a good way to do this check)?

Is there any reason you would prefer B? I can't think of one.
B is incorrect in case of error.
Also, it is bad style. Your intention is to wait for all tasks, so say that in the code. No need for manual loops obscuring what you want to accomplish.
Also, does that "task.Status == TaskStatus.RanToCompletion" return true as long as the task didn't throw an Exception (is this a good way to do this check)?
This checks for successful completion. Likely, this is what you want. Tasks can end up cancelled but likely you will consider that as an error case.
task.Result != null
Is that really what you want? task.Result is never set to null by the system. This can only happen if GetAllRouterInterfacesAsync makes the result null.

Related

Finish two tasks then printing something

I have three tasks, one is producer, then consumer and the last one is to print something after finishing the first two. However the code doesn't reach the last task, which means no printing.
while (true)
{
ThreadEvent.WaitOne(waitingTime, false);
lock (SyncVar)
{
collection = new BlockingCollection<string>(4);
Task producer = Task.Run(() =>
{
if (list.Count > 0)
Console.WriteLine("Block begin");
while (!collection.IsAddingCompleted)
{
var firstItem = list.FirstOrDefault();
collection.TryAdd(firstItem);
list.Remove(firstItem);
}
collection.CompleteAdding();
});
Task consumer = Task.Run(() => DoConsume());
Task endTask = consumer.ContinueWith(i => Console.WriteLine("Block end"));// not print this line, why?
Task.WaitAll(producer, consumer, endTask);
if (ThreadState != State.Running) break;
}
}
Please look at my code logic.
EDIT:
For `DoConsume', it is complicated.
public void DoConsume()
{
if (collection.Count > 0)
Console.WriteLine("There are {0} channels to be processed.", collection.Count);
var workItemBlock = new ActionBlock<string>(
workItem =>
{
bool result =ProcessEachChannel(workItem);
});
foreach (var workItem in collection.GetConsumingEnumerable())
{
workItemBlock.Post(workItem);
}
workItemBlock.Complete();
}
The problem is that your producer will never complete:
// This will run until after CompleteAdding is called
while (!collection.IsAddingCompleted)
{
var firstItem = list.FirstOrDefault();
collection.TryAdd(firstItem);
list.Remove(firstItem);
}
//... which doesn't happen until after the loop
collection.CompleteAdding();
It looks like you're just trying to add all of the items in your list, which should be as simple as:
Task producer = Task.Run(() =>
{
if (list.Count > 0)
Console.WriteLine("Block begin");
while(list.Any())
{
var firstItem = list.First();
collection.TryAdd(firstItem);
list.Remove(firstItem);
}
collection.CompleteAdding();
});
Or, a simpler method:
Task producer = Task.Run(() =>
{
if (list.Count > 0)
Console.WriteLine("Block begin");
foreach(var item in list)
{
collection.TryAdd(item);
}
list.Clear();
collection.CompleteAdding();
});
I used Reed Copsey's code but the error is still there. Just can't figure it out why.
I think that my code has the flaw at while (!collection.IsAddingCompleted).
Because the collection has the boundary of 4, suppose there are two item left in the collection. The condition collection.IsAddingCompleted is never met therefore the code could not jump out of the while loop.
I rewrote the code, it seems fine. The code is similar MSDN. I used Take to retrieve the element in the collection.
while (true)
{
ThreadEvent.WaitOne(waitingTime, false);
lock (SyncVar)
{
collection = new BlockingCollection<string>(4);
DoWork dc = new DoWork();
Task consumer = Task.Run(() =>
{
while (!collection.IsCompleted)
{
string data = "";
try
{
if (collection.Count > 0)
data = collection.Take();
}
catch (InvalidOperationException e)
{
Console.WriteLine(e.Message);
}
if (data != "")
{
bool result = dc.DoConsume(data);
}
}
});
Task producer = Task.Run(() =>
{
if (list.Count > 0)
Console.WriteLine("Block begin");
foreach (var item in list)
{
collection.Add(item);
}
list.Clear();
collection.CompleteAdding();
});
Task endTask = consumer.ContinueWith(i => Console.WriteLine("Block end"));
Task.WaitAll(producer, consumer, endTask);
if (ThreadState != State.Running) break;
}

detect when each task is complete

I want to update a progressbar as each task is completed below.
The method var continuation2 = Task.Factory.ContinueWhenAny(..... doesnt work.
What is the correct way to do this?
C# Code
private void radButtonInsertManyErrors_Click(object sender, EventArgs e)
{
try
{
radProgressBarStatus.Maximum = int.Parse(radTextBoxNumberofErrorsInsert.Text);
radProgressBarStatus.Value1 = 0;
Task<int>[] tasks = new Task<int>[int.Parse(radTextBoxNumberofErrorsInsert.Text)];
for (int i = 0; i < int.Parse(radTextBoxNumberofErrorsInsert.Text); i++)
{
int x = i;
tasks[i] = new Task<int>(() =>
{
//insert the error into table FA_Errors
Accessor.Insert_FAErrors(BLLErrorCodes.BLL_Error_Codes.Error_Log_Event_Login.ToString(),
(int)BLLErrorCodes.BLL_Error_Codes.Error_Log_Event_Login,
"Some Error", "",
MethodBase.GetCurrentMethod().DeclaringType.Namespace.ToString(),
MethodBase.GetCurrentMethod().Name.ToString(),
BLLErrorCategory.BLL_Error_Category.WEB_APP.ToString(),
"pc source", "damo",
sConn.ToString());
return 1;
});
}
var continuation = Task.Factory.ContinueWhenAll(
tasks,
(antecedents) =>
{
RadMessageBox.Show("Finished inserting errors ");
});
var continuation2 = Task.Factory.ContinueWhenAny(
tasks,
(antecedents) =>
{
radProgressBarStatus.Value1++;
});
for (int i = 0; i < int.Parse(radTextBoxNumberofErrorsInsert.Text); i++)
tasks[i].Start();
// Use next line if you want to block the main thread until all the tasks are complete
//continuation.Wait();
}
catch (Exception ex)
{
MessageBox.Show(ex.Message.ToString());
}
}
You can use this function:
public static void TaskProgress(IEnumerable<Task> tasks, Action<int> callback)
{
int count = 0;
foreach (var task in tasks)
task.ContinueWith(t => callback(Interlocked.Increment(ref count)));
}
It will call the callback each time a task completes with the number of currently completed tasks. Note that the callbacks are not synchronized, so it can be called while the previous callback is still running.
Set up a continuation with each of the tasks. Keep a (thread-safe) counter on how many completed and update the UI on completion of each task.
Actually, Task.WhenAll does keep such a counter under the hood. It is just not accessible.

Aggregate and forward exceptions from multiple threads

In my method I start multiple threads and then wait until they finish their work (something like fork-join pattern).
using (var countdownEvent = new CountdownEvent(runningThreadsCount))
{
for (int i = 0; i < threadsCount; i++)
{
var thread = new Thread(new ThreadStart(delegate
{
// Do something
countdownEvent.Signal();
}));
thread.Start();
}
countdownEvent.Wait();
}
Now I need to be able to catch exception in this threads (lets assume that // Do something may throw an exception), pass exception to the main thread, unblock it (since it is waiting on countdownEvent) and re-throw the exception.
What is the most elegant way to achieve that?
Solved my problem with Tasks API. Thanks flq for suggestion!
var cancellationTokenSource = new CancellationTokenSource();
var tasks = new Task[threadsCount]
for (int i = 0; i < threadsCount; i++)
{
tasks[i] = Task.Factory.StartNew(
delegate
{
// Do something
}, cancellationTokenSource.Token);
}
try
{
Task.WaitAll(tasks);
}
catch (AggregateException ae)
{
cancellationTokenSource.Cancel();
throw ae.InnerExceptions[0];
}

very unique exception for BlockingCollection on .net 4.0

i'm using the BlockingCollection for a Producer Consumer pattern and i got an excecption i think to write a patent on it- only two results in google!
the expection is "CompleteAdding may not be used concurrently with additions to the collection" and it happens when i TryAdd on th BlockingCollection as Follows:
public void EnqueueTask(T item)
{
if (!_cancellationTokenSource.IsCancellationRequested)
{
_workerQueue.Add(item);
}
}
the CompleteAdding is called on the dispose of the Consumer-Producer wrapper class:
public void Dispose()
{
if (!_IsActive)
return;
_IsActive = false;
_cancellationTokenSource.Cancel();
_workerQueue.CompleteAdding();
// Wait for the consumer's thread to finish.
for (int i = 0; i < _workers.Length; ++i)
{
Task t1 = Task.Factory.StartNew(() =>
{
try
{
if (!_workers[i].Join(4000))
LogWriter.Trace("Failed to join thread", "ThreadFailureOnDispose");
}
catch (Exception ex)
{
OnLogged(ex.Message + ex.StackTrace);
}
});
}
// Release any OS resources.
}
Anyone from microsoft got an idea? should i sleep after the cancelation and before calling the CompleteAdding?
Look at this piece of the code:
for (int i = 0; i < _workers.Length; ++i)
{
Task t1 = Task.Factory.StartNew(() =>
{
try
{
if (!_workers[i].Join(4000)) << == Here
LogWriter.Trace("Failed to join thread", "ThreadFailureOnDispose");
}
In _workers[i].Join(4000), the value of i is not what you think it is. Try again with:
for (int i = 0; i < _workers.Length; ++i)
{
int j = i; // copy
Task t1 = Task.Factory.StartNew(() =>
{
try
{
if (!_workers[j].Join(4000)) // j
LogWriter.Trace("Failed to join thread", "ThreadFailureOnDispose");
}
In your version, the variable 'i' is captured and all Tasks use the same var. All but the first few will see i == _workers.Length because they are executed after the for-loop is completed.
It is a classic lambda + captured var problem.

Is this goto expressive?

The following code was a proof of concept for a message batching routine. Do I avoid goto like the plague and rewrite this code? Or do you think the goto is an expressive way to get this done?
If you'd rewrite please post some code...
var queue = new Queue<TraceItem>(this.batch);
while (this.connected)
{
byte[] buffer = null;
try
{
socket.Recv(out buffer);
}
catch
{
// ignore the exception we get when the socket is shut down from another thread
// the connected flag will be set to false and we'll break the loop
}
HaveAnotherMessage:
if (buffer != null)
{
try
{
var item = TraceItemSerializer.FromBytes(buffer);
if (item != null)
{
queue.Enqueue(item);
buffer = null;
if (queue.Count < this.batch && socket.Recv(out buffer, ZMQ.NOBLOCK))
{
goto HaveAnotherMessage;
}
}
}
catch (Exception ex)
{
this.ReceiverPerformanceCounter.IncrementDiagnosticExceptions();
this.tracer.TraceException(TraceEventType.Error, 0, ex);
}
}
// queue processing code
}
Pretty much sums up my thoughts on "goto."
Goto is bad programming practice for many reasons. Chief among them is that there is almost never a reason for it. Someone posted a do..while loop, use that. Use a boolean to check if you should continue. Use a while loop. Goto's are for interpreted languages and a call back to assembler days (JMP anyone?). You're using a high level language for a reason. So that you and everyone else doesn't look at your code and get lost.
To keep this answer somewhat current I'd like to point out that a combination of goto and bracing errors caused a major SSL bug in iOS and OS X.
Replace the goto with a do-while, or simply a while loop if you don't want the "always run once" functionality you have right now.
var queue = new Queue<TraceItem>(this.batch);
while (this.connected)
{
byte[] buffer = null;
try
{
socket.Recv(out buffer);
}
catch
{
// ignore the exception we get when the socket is shut down from another thread
// the connected flag will be set to false and we'll break the loop
}
do {
if (buffer != null)
{
try
{
var item = TraceItemSerializer.FromBytes(buffer);
if (item != null)
{
queue.Enqueue(item);
buffer = null;
}
}
catch (Exception ex)
{
this.ReceiverPerformanceCounter.IncrementDiagnosticExceptions();
this.tracer.TraceException(TraceEventType.Error, 0, ex);
}
}
} while(queue.Count < this.batch && socket.Recv(out buffer, ZMQ.NOBLOCK))
// queue processing code
}
It's so amazingly easy to rid yourself of GOTO in this situation it makes me cry:
var queue = new Queue<TraceItem>(this.batch);
while (this.connected)
{
byte[] buffer = null;
try
{
socket.Recv(out buffer);
}
catch
{
// ignore the exception we get when the socket is shut down from another thread
// the connected flag will be set to false and we'll break the loop
}
bool hasAnotherMessage = true
while(hasAnotherMessage)
{
hasAnotherMessage = false;
if (buffer != null)
{
try
{
var item = TraceItemSerializer.FromBytes(buffer);
if (item != null)
{
queue.Enqueue(item);
buffer = null;
if (queue.Count < this.batch && socket.Recv(out buffer, ZMQ.NOBLOCK))
{
hasAnotherMessage = true;
}
}
}
catch (Exception ex)
{
this.ReceiverPerformanceCounter.IncrementDiagnosticExceptions();
this.tracer.TraceException(TraceEventType.Error, 0, ex);
}
}
}
// queue processing code
}
I guess the goto is SLIGHTLY more readable intuitively... But if you WANTED to avoid it I think all you'd have to do is throw the code in a while(true) loop, and then have a break statement at the end of the loop for a normal iteration. And the goto could be replaced with a continue statement.
Eventually you just learn to read and write loops and other control flow structures instead of using goto statements, at least in my experience.
Kind of related to Josh K post but I'm writing it here since comments doesn't allow code.
I can think of a good reason: While traversing some n-dimensional construct to find something. Example for n=3 //...
for (int i = 0; i < X; i++)
for (int j = 0; j < Y; j++)
for (int k = 0; k < Z; k++)
if ( array[i][j][k] == someValue )
{
//DO STUFF
goto ENDFOR; //Already found my value, let's get out
}
ENDFOR: ;
//MORE CODE HERE...
I know you can use "n" whiles and booleans to see if you should continue.. or you can create a function that maps that n-dimensional array to just one dimension and just use one while but i believe that the nested for its far more readable.
By the way I'm not saying we should all use gotos but in this specific situation i would do it the way i just mentioned.
You could refactor is to something like this.
while (queue.Count < this.batch && buffer != null)
{
try
{
var item = TraceItemSerializer.FromBytes(buffer);
buffer = null;
if (item != null)
{
queue.Enqueue(item);
socket.Recv(out buffer, ZMQ.NOBLOCK)
}
}
catch (Exception ex)
{
this.ReceiverPerformanceCounter.IncrementDiagnosticExceptions();
this.tracer.TraceException(TraceEventType.Error, 0, ex);
}
}
Umm, I'm not really sure you want to goto out of a try block. I'm pretty sure that is not a safe thing to do, though I'm not 100% sure on that. That just doesn't look very safe...
Wrap the "HaveAnotherMessage" into a method that takes in the buffer and may call itself recursively. That would seem to be the easiest way to fix this.
I would avoid goto in this case, and refactor it. The method reads too long in my opinion.
I think your method is too big. It mixes different levels of abstraction, like error processing, message retrieval and message processing.
If you refactor it in different methods, the goto naturally goes away (note: I assume your main method is called Process):
...
private byte[] buffer;
private Queue<TraceItem> queue;
public void Process() {
queue = new Queue<TraceItem>(batch);
while (connected) {
ReceiveMessage();
TryProcessMessage();
}
}
private void ReceiveMessage() {
try {
socket.Recv(out buffer);
}
catch {
// ignore the exception we get when the socket is shut down from another thread
// the connected flag will be set to false and we'll break the processing
}
}
private void TryProcessMessage() {
try {
ProcessMessage();
}
catch (Exception ex) {
ProcessError(ex);
}
}
private void ProcessMessage() {
if (buffer == null) return;
var item = TraceItemSerializer.FromBytes(buffer);
if (item == null) return;
queue.Enqueue(item);
if (HasMoreData()) {
TryProcessMessage();
}
}
private bool HasMoreData() {
return queue.Count < batch && socket.Recv(out buffer, ZMQ.NOBLOCK);
}
private void ProcessError(Exception ex) {
ReceiverPerformanceCounter.IncrementDiagnosticExceptions();
tracer.TraceException(TraceEventType.Error, 0, ex);
}
...

Categories

Resources