This is a two part question:
I am working on a big project where multiple plugins developed by different teams are loaded inside one common container shell. At times I can see that my UI updates are blocked as there are multiple parallel UI updates, i want to know if there is a way to find which component is blocking the ui thread
In .net how can i create a separate UI thread which requires dedicated UI intensive work?
Much Appreciate your help. Thanks.
Use the debugger. Debug + Break All when you notice it blocking. Then Debug + Windows + Threads and select the main thread. The call stack window shows you what it is doing.
A corner case is where these plugins are using a lot of calls to Control.Begin/Invoke or Dispatcher.Begin/Invoke. Your UI thread is not blocked in this case, it is just being overwhelmed by requests to dispatch the delegate targets. And doesn't get around to doing its normal duties anymore, like repainting the windows and responding mouse and keyboard events. There's little you can do about this beyond working with the plugin authors to get them to mend their ways.
You've already got an UI thread, the thread that created the first window. Creating additional threads that have their own windows is possible but causes unsolvable problems with window Z-order (a window will disappear underneath the window of another app) and generous helpings of window interop threading misery.
Visual Studio 2010's (in the higher SKUs) include features to check for this. If you run your program under the Concurrency Profiler, you can see exactly which threads are waiting on which locks when the deadlock occurs. In addition, it will highlight the deadlock (I believe in bright red) to make it easy to track down.
One approach you can take (though it may require a bit of redesign) is to disallow all plugin logic from running in the UI thread. All operations that require updates to the UI must be routed through well-defined service interfaces that can interpret, dispatch and perhaps even throttle the UI updates. This is only practical if your plugins are not deeply UI-centric and you have a service model that allows you to isolate the data being manipulated by the plugins from the visualization of that data. Without knowing more about your application, I can't give more concrete recommendations.
Here are two possible solutions to the problem that I came up with quickly. I am sure there are other equally valid solutions though.
Option 1: Instead of using the push model (via the ISynchronizeInvoke methods) switch to a pull (or poll) model in which the UI queries the plugin for updates. This has the following advantages.
It breaks the tight coupling between the UI and worker/plugin threads that Control.Invoke imposes.
It puts the responsibility of updating the UI thread on the UI thread where it should belong anyway.
The UI thread gets to dictate when and how often the update should take place.
There is no risk of the UI message pump being overrun as would be the case with the marshaling techniques initiated by the worker/plugin thread.
The worker/plugin thread does not have to wait for an acknowledgement that the update was performed before proceeding with its next steps (ie. you get more throughput on both the UI and worker/plugin threads).
Option 2: Have the plugin accept an ISynchronizeInvoke instance instead of an actual Form or Control. This special synchronizing object will be implemented using a dedicated thread and a queue that acts as buffer between the plugin and the UI. It will accept update messages via the normal Invoke or BeginInvoke methods, which means you can keep the plugin architecture and interfaces mostly intact, and then forward those messages on to the UI after some type of filtering, merging, and throttling operations have occurred. The number of update messages existing in the synchronizing object will ebb and flow as the UI and plugin threads work load changes. It could be smart enough to change its forwarding strategy as the rate of messages increase.
Related
I have an application in active use with a rarely occurring 'freeze' of the UI thread. This of course results in the application becoming unresponsive and requires the user to terminate it manually.
Due to how difficult this issue is to reproduce, I'd like to develop some tools to help gather data on when it's happening in the field and what's occurring at the time. What I'm thinking about is some sort of background watchdog task that would monitor a 'heartbeat' from a periodically scheduled UI thread task. If it goes quiet for too long, I can say that the UI is effectively locked up (or at least stalled for way longer than I'd ever want) and hence gather data.
So, in preparation for this, a few questions:
Does something like this already exist? This seems like a reasonably common problem, so if there are existing tools to help diagnose this, it might be worth using those rather than rolling my own solution.
I'm still debating what information I should try to gather when the freeze is detected. Is there some way for me to easily grab the stack trace of the UI thread so that it can be logged? Possibly grab stack traces from all active threads? Is there some way I can capture a complete debug dump?
Does something like this already exist?
Yes. It is imperative to do this outside of the process since a deadlocked one makes it very likely that the diagnostic code is going to deadlock as well. The DebugDiag utility has explicit support for hung apps.
I'm still debating what information I should try to gather
The minidump you get out of DebugDiag should be enough to give you a shot a diagnosing the cause. A sample debug session that demonstrates tackling a deadlock is shown here.
And of course a good hard look at your code never hurts. UI thread deadlocks are frequently caused by:
Displaying a window on a worker thread. The SystemEvents class is a very significant troublemaker when you do this. It needs to fire its events on the UI thread but that requires it to guess which specific thread in your program is actually the UI thread. Once it guesses wrong, you are set for a spontaneous deadlock any time later. Do note that this doesn't require you using the SystemEvents class in your own code, many controls subscribe the ThemeChanged event to repaint themselves. Fwiw, that debug session I linked to demonstrates such a deadlock. Beware of home-spun splash screens, "progress" windows that are created on the worker thread instead of the UI thread and of course any worker thread that displays UI.
Data bound controls whose underlying binding source is updated on a worker thread. Very common of course since dbase queries tend to be slow. Such controls have to be unbound explicitly first.
My application is heavily depended on threads to do the complex processing of very large data. The UI needs to be updated as the processing is being done. I know and tried to used BackgroundWorker's OnProgressChanged and RunWorkerCompleted methods to update the UI. Also using Invoke method of the UI thread to update. Everything seems to work fine on Win XP 32 bit and 64 bit OS. On Win Vista and Win 7 (32 and 64 bit), the application randomly hangs while updating the UI by using Invoke method.
Does the behavior of Invoke changes on different Win OS?
What are the other ways of updating UI from thread apart from Invoke?
Thanks
Does the behavior of Invoke changes on different Win OS?
It should not, no. But, threading problems can materialize in very unpredictable ways. It is possible that you have an unidentified problem.
What are the other ways of updating UI from thread apart from Invoke?
Using Invoke or BeginInvoke are way overused especially when trying to report simple progress information to the UI thread. If you search for some of my answers related to the subject you will see that I rip on this approach all of the time. And for good reason since their are many disadvantages to using this technique. It is unfortunate that BackgroundWorker uses this mechanism exclusively for updating the UI through its ProgressChanged event.
One alternate method is to have your worker thread publish progress information into a shared variable and have the UI thread poll for it periodically via a timer. Here are some of my usual talking points in justifying this approach over the marshaling techniques.
Invoke and BeginInvoke are expensive operations.
The UI thread gets to dictate when and how often to update the form and its controls.
It eliminates the tight coupling between UI and worker threads that ISynchronizeInvoke imposes.
There is no risk of overrunning or saturating the UI message queue with a bunch of marshaling operations.
You get more throughput on the worker thread since it does not have to wait for response as would be the case with Invoke.
Not sure what's going wrong, but you could always just have a System.Windows.Forms.Timer running that updates the GUI periodically; use some member variables to pass raw data between threads, inside locks if necessary. Not the most elegant solution, but it might give you a different perspective into what's hanging since the threads are more independent this way, rather than relying on a background thread to Invoke your main thread.
You could try using one of the Invoke() or BeginInvoke() overloads that takes a Dispatcher.Priority enum as a parameter. If you select a parameter such as 'Background' you should see your application is still responsive. The only issue then becomes ensuring you are servicing your incoming data at an adequate rate without an ever increasing queue.
Another option would be to forego the multithreading entirely. If your long-running operation can be broken up into chunks, do it in the GUI thread and use Application.DoEvents() when you want to update the GUI.
I used to dislike use of that method since it could not only update the GUI but also start responding to user input, kicking off timers, etc., but ultimately it's no less safe than using a background thread, which allows the GUI to start doing anything at any time. So probably after each call of Application.DoEvents() you'd need to check _canceled or whatever. (I eventually decided I dislike the existence of this method, since it eliminates the guarantees of linear execution order, than the use of it).
Of course you lose multicore support this way, so it'd affect performance if you're trying to run lots of background ops at the same time.
I'm currently developing a project with XNA that is pulling information (ID, name, file location, etc) about each of my objects (each object will be displayed on screen) from a local SQL database.
I'd like to run my database queries on a separate thread so the rendered screen doesn't freeze if the database hangs or some other unforeseen event occurs. I'm using XNA 4.0 and the application will only be running on windows. Is this possible, and if so, how?
There are a number of options available. Generally speaking you need the query to run in a separate thread. You can use
Thread pool
QueueUserWorkItem
Tasks
Background worker
Async calls to the database
Parallel invoke
Manually created threads here and here
I would start with thread pooling and see how that works, dedicated manual threads are not that robust in terms of memory management and reuse.
Not to do it at all. Seriously. There are good reasons for using threads, but your reasons are bogus:
the rendered screen doesn't freeze if the database hangs or some other unforeseen event occur
Databases dont hang and unforseen events are unforseen events. How you can cope with the database not answering for 3 minutes, for example? Show a screen with objects that are unknown?
How do you mean "best"? There are a lot of ways to use threads and they all have strengths and weaknesses.
Declaring a new thread explicitly and starting it gives you the most direct control over the execution state of that thread:
var myDbThread = new Thread(()=>myDbRepo.GetRecordById<MyEntity>(idString));
myDbThread.Start();
Now, as long as you have a reference to myDbThread, you can abort it, pause it, join on it, etc. BUT, with control comes responsibility; you have to manage the threads you create yourself.
For most parallel tasks, using the ThreadPool is recommended. However, you lose some of the control:
Action myDbLambda = () => myEntityProperty = myDbRepo.GetRecordById<MyEntity>(idString);
var asyncResult = myDbLambda.BeginInvoke();
Once asyncResult.IsComplete returns true, myEntityProperty has the value. You can also architect it as a Func, and use a callback to set the value (this is recommended). The Asynchronous Model is built in to the BeginInvoke()/EndInvoke() method pair, and many exceptions like timeouts are expected by the ThreadPool, which will simply restart the timed-out thread. However, you can't "give up" and terminate a ThreadPool thread, "joining" on a ThreadPool thread is a little trickier, and if you're launching a lot of threads, the ThreadPool will start them in 250ms intervals which may not be the best use of processor.
There are many ways to use the ThreadPool; before delegates became even more important to .NET programming in v3.5, ThreadPool.QueueUserWorkItem was the main method. Now, as I said, delegates have BeginInvoke and EndInvoke methods allowing you to kick off background processes with the asynchronous model built in behind the scenes. In WinForms/WPF, you can also create BackgroundWorker components which are event-driven, allowing you to monitor progress and completion in a GUI element.
One thing to be aware of; it is virtually never a good idea to use background threads in ASP.NET. Unless you really know what you're doing, best-case you won't get the results of the behavior you sent to the worker thread, and worst-case you can crash your site trying.
This is possibly related to ProgressBar updates in blocked UI thread but is a little different.
While troubleshooting a crashing WinForms control (a DevExpress treelist) a member of our team encountered an unusual situation, and I'm wondering if anyone can help us understand what's going on. English is not his first language, so I'm posting on his behalf.
Please see this screenshot from Visual Studio 2005.
Note the following points:
The main UI thread is stopped and is currently in a DevExpress control draw method.
The code shown on screen is from a point earlier in the same call-stack. This code is in the data layer and was called in response to the control's request for an image to display for the tree node. (perhaps also originating from a Paint handler)
The displayed code is from earlier in the callstack, on the main UI thread, and is currently waiting on a lock! Since remote systems can send events which are processed on background threads in the data model (i.e., data models are sync'd between client and servers), we lock to keep the data collections thread safe.
As the callstack shows, we continued to process paint messages on the UI thread, while we would expect the thread to be blocked.
This is very difficult to replicate, and I have not been able to do so using a simpler test project on my own box. When this situation arises, however, the result is that the DevExpress control's internal state can be messed up, causing the control to crash. This doesn't really seem like a bug in the control, since it was no doubt written with the assumption that these paint methods are running only on the UI thread. What we see here makes it look like the UI thread is acting like two threads.
It would seem possible that this is merely a Visual Studio bug in the presentation of the callstack, except that this whole endeavor is resulting from an effort to troubleshoot the occasional crash of the control in the released app (in which case it shows as a big red X in the UI), so it seems the problem is not isolated to the debug environment.
Alright, that was complicated, but hopefully made sense. Any ideas?
I would strongly recommend against locking the UI to wait for background processing. Consider something like multiple buffering. You can probably get this behavior fairly easily by utilizing thread-safe collections in .NET 4, but if that's not an option there are versions of those in the Parallel Extensions released prior to v4.
What about altering the synchronization scheme so that you don't need to acquire exclusive locks to read data?
In situations where you can be sure that a read will always produce consistent data even when it happens while the data is also being written, you might be able to get away with having no lock statements for getters. Otherwise there's ReaderWriterLockSlim, which permits multiple concurrent readers but still allows you to stop the presses for write operations.
It doesn't fix everything, but it does at least reduce the number of opportunities for deadlocks.
We see something similar in our project. The stack trace looks like the pump's event loop is called while the UI thread is waiting on a lock. This could happen if Monitor.enter has some special behavior when called on the UI thread. I believe this is what's happening, but I haven't found any documentation to back it up yet.
Probably has something to do with synchronization contexts :)
I was writing up a long, detailed question, but just scrapped it in favor of a simpler question that I didn't find an answer to here.
Brief app description:
I have a WPF app that spawns several threads, and each thread executes its own WF. What are some of the best ways to handle errors in the threads and WF that will allow user interaction from the GUI side? I definitely plan to handle any low level exceptions in the thread, because I don't want the thread to exit.
Summary of questions:
How have you implemented communication between WF and the thread that starts it? There is WorkflowTerminated, but I don't want the workflow to exit -- I need to fix the problem and let it continue. I assume the only option is using a FaultHandler, but was wondering if there's another way to do it without using an activity block. I am hoping there's a framework out there that I just haven't found yet.
The error from WF needs to get caught by the thread, which then needs to display the error in the GUI. The user will then make a logical choice for recovery, which should then be sent back to the thread, and then to WF. Again, is there something existing out there that I should take a look at?
Even buzzwords / keywords that accomplish what I am describing would be really helpful, and I can do the legwork on researching each of them. However, any additional insight is always welcome. :)
What's worked for me in multi-threaded WPF apps is to have the errant thread invoke a callback method that passes the exception and other info back to the UI thread. Callbacks can have return values, so if your thread can block while waiting for the user to respond, then that can work for you. Remember that the callback will run on the thread that calls it, so any UI updates have to be done via the control's dispatcher. You will have to decide whether all of the threads use the same callback and what kind of synchronization you'll need if there's a chance that multiple threads can throw exceptions simultaneously.
Here's how I ended up solving this problem. But first a little background info:
User clicks a button in the GUI that causes the candy packager to start running. This is done via a command binding in the ViewModel, which then calls a low-level function in the Model. The function in the model launches a thread and executes a state machine.
At some point, the machine will fail. When it does, I compile information about the error and possible (known) recovery methods. I put this into an object and then pass it to the GUI via a callback interface. In the meantime, the worker thread is stuck waiting for an Event to get set.
Eventually, the candy worker will notice the error and will click a button telling the system what to do. This results in two things: 1) it flags one of the recovery methods as the preferred one, and 2) sets the event. Now the worker thread continues on, checks for the preferred error recovery method and transitions into the respective state in the state machine.
This works very well (so far). The part I know is totally lame is the manner in which it checks for the preferred error recovery method. I am essentially setting a string variable, and then comparing this string to a list of known strings. Ultra lame, but I'm not sure of a better way to do this, other than using an enum. Does anyone have recommendations for me?