I have a confession to make: I enjoy etiquette. My wife and I own a good dozen or two etiquette books, ranging from the classic Post to a more playful Austen-themed guide as well as several historical books (including an uncut copy of Social Life of Virginia in the Seventeenth Century). There's a certain nerdy appreciation of knowing how to act (at least, according to somebody...).
Many developers are unsure how to properly use
Task.Run. This confusion is perfectly normal; when you first start using
async, it's kind of like sitting down at a formal dinner and seeing three forks when you've spent your entire life only using spoons. "When should I use the
Task.Run fork? Is this for salad or dessert???"
The key central theme of all etiquette is to treat others well. In this case, the "others" are other developers. "Others" will come along and maintain your code; "others" will try to reuse your code in different contexts. As we'll see, treating those "others" well is the key to properly using
First, let's think of what
Task.Run is really for (you wouldn't want to use a fork to eat soup!). The purpose of
Task.Run is to execute CPU-bound code in an asynchronous way.
Task.Run does this by executing the method on a thread pool thread and returning a
Task representing the completion of that method.
That sounds so simple, but we've already eliminated a whole slew of poor examples. Many
async newbies start off by trying to treat asynchronous tasks the same as parallel (TPL) tasks, and this is a major misstep. True, asynchronous tasks and parallel tasks are the same type (
Task, of course), but their purpose is completely different and thus their proper usage is also completely different. Developers new to
async begin reading some examples, see these newfangled
async methods returning tasks, and (incorrectly) assume that
async is all about background threads and whatnot. With that initial misunderstanding, the next logical step is to attempt to implement all asynchronous methods using
Task.Run (or even worse,
I've seen many, many intelligent developers fall into that same mistake.
So, the question remains: Where should I use
Task.Run to call CPU-bound methods. That is all.
One common mistake is to try to make asynchronous "wrappers" around existing synchronous methods. Stephen Toub has a detailed blog post describing why this is a bad idea. I call such methods "fake-asynchronous methods" because they look asynchronous but are really just faking it by doing synchronous work on a background thread. In general, do not use
Task.Run in the implementation of the method; instead, use
Task.Run to call the method. There are two reasons for this guideline:
- Consumers of your code assume that if a method has an asynchronous signature, then it will act truly asynchronously. Faking asynchronicity by just doing synchronous work on a background thread is surprising behavior.
- If your code is ever used on ASP.NET, a fake-asynchronous method leads developers down the wrong path. The goal of
asyncon the server side is scalability, and fake-asynchronous methods are less scalable than just using synchronous methods.
So, any code that you want to be reusable should not use
Task.Run in its implementation. Consider the developers (including yourself) who will need to consume that code.
OK, so let's complicate the question a bit. What if I have a reusable method that uses significant amounts of both I/O and CPU? Should I use
Task.Run for the CPU-bound parts?
The answer is still no.
However, in this (uncommon) situation, you do end up with a bit of an awkward solution: an asynchronous method that also does CPU-bound work. In this case, you should document clearly that the method is not fully asynchronous, so that callers know to wrap it in a
Task.Run if necessary. (Remember, it's necessary if it's being called from the UI thread, but not necessary if called from a background thread or ASP.NET).
To conclude, synchronous methods should have a synchronous signature:
// Documentation: This method is CPU-bound (use Task.Run to call from a UI thread). void DoWork();
Asynchronous methods should have an asynchronous signature:
And methods that are a mixture of synchronous and asynchronous work should have an asynchronous signature with documentation point out their partially-synchronous nature:
// Documentation: This method is CPU-bound (use Task.Run to call from a UI thread). Task DoWorkAsync();