Async/Await simplifies async code, use it everywhere and life becomes so simple right?
While this is true I’ve seen situations where users either chose to, or had to, mix async and non async code and got themselves into a world of problems.
One problem I’ve seen time over time is with Windows Desktop applications where a simple blocking call on a Task can deadlock an application entirely, here I demonstrate the problem with a contrived Windows Forms example.
Application simply downloads some html asynchronously and displays it in a web browser
Implementation of async function is:
|
|
(Let’s ignore the urge to make the click handler async, imagine the async call was in the form constructor if you must)
How can such a simple bit of code deadlock the windows application?
Well the problem occurs because of how Async/Await state machines work.
I’m really going to simplify this explanation as I want people to grasp it (so grit ur teeth if you already know the detail )
The async keyworks is simply a compile instruction that doesn’t do much so lets ingore that and focus on the await call
The await calls an async function then waits on a call back, when the call back occurs the code resumes to the next step…
ok so far so good, this is what we expect… simple right…wrong!
A quick recap of windows UI threads and messages
Before we continue let’s have a quick recap of windows UI threads and message loops.
A message loop is an obligatory section of code in every program that uses a graphical user interface under Microsoft Windows. Windows programs that have a GUI are event-driven. Windows maintains an individual message queue for each thread that has created a window.
Now as anyone working on a windows application knows you always call any code that updates the UX on the GUI thread, try with any other threads and you’ll get presented with a cross thread exception.
Windows forms application code can call Invoke/BeginInvoke on a windows control and execute the code back on the GUI Thread, in WPF we would use a dispatcher, in UWP/Wintr it’s something else.
Another approach is to use the SyncronizationContext contruct, this is aware if it should call Invoke or Dispatch or something else on our behalf.
Back to await
The callback I mentioned above is smart in that it tries to use the existing Syncronization context if it exists, so when that await finally returns we’re back on the GUI thread can can update the UX without those pesky errors.
In our calling code above we never left the GUI thread as we were blocked on the Result of the task.
The crux of the problem is that the await call back puts a message on the windows message queue to tell it to continue, but the message queue is bocked in that call to .Result on the task, so we’re well and truly deadlocked.
Solutions
I’ll avoid telling you to embrace async/await everywhere! and offer some alternate solutions..
1) ConfigureAwait(false)
This works as it tells the task not to continue on the current syncroniztion context (which let’s remember is the GUI thread), another context is used to complete the async await state machine call back and allow the task to be completed.
2) Run on another syncronization context you create
3) Use Async/Await everywhere
Summary
This is a very dumbed down explanation of how you might encounter a deadlock with async await.
If it happens then don’t panic it can be easily fixed once you know what’s happening, best practise is nearly never to work around the problem and always use async await entirely.
Standard documentation is great there are lots for really good articles floating about: e.g. https://devblogs.microsoft.com/dotnet/configureawait-faq/