I'm still learning the async/await, so please excuse me if I'm asking something obvious. Consider the following example:
class Program {
static void Main(string[] args) {
var result = FooAsync().Result;
Console.WriteLine(result);
}
static async Task<int> FooAsync() {
var t1 = Method1Async();
var t2 = Method2Async();
var result1 = await t1;
var result2 = await t2;
return result1 + result2;
}
static Task<int> Method1Async() {
return Task.Run(
() => {
Thread.Sleep(1000);
return 11;
}
);
}
static Task<int> Method2Async() {
return Task.Run(
() => {
Thread.Sleep(1000);
return 22;
}
);
}
}
This behaves as expected and prints "33" in the console.
If I replace the second await with an explicit wait...
static async Task<int> FooAsync() {
var t1 = Method1Async();
var t2 = Method2Async();
var result1 = await t1;
var result2 = t2.Result;
return result1 + result2;
}
...I seem to get the same behavior.
Are these two examples completely equivalent?
And if they are equivalent in this case, are there any other cases where replacing the last await by an explicit wait would make a difference?
Your replacement version blocks the calling thread waiting for the task to finish. It's hard to see a visible difference in a console app like that since you're intentionally blocking in Main, but they're definitely not equivalent.
They are not equivalent.
Task.Result blocks until the result is available. As I explain on my blog, this can cause deadlocks if you have an async context that requires exclusive access (e.g., a UI or ASP.NET app).
Also, Task.Result will wrap any exceptions in AggregateException, so error handling is harder if you synchronously block.
OK, I think I figured this out so let me sum it up, in what will hopefully be a more complete explanation than the answers provided so far...
Short Answer
Replacing the second await with an explicit wait will have no appreciable effect on a console application, but will block the UI thread of a WPF or WinForms application for the duration of the wait.
Also, the exception handling is slightly different (as noted by Stephen Cleary).
Long Answer
In a nutshell, the await does this:
If the awaited task has already finished, it just retrieves its result and continues.
If it hasn't, it posts the continuation (the rest of the method after the await) to the current synchronization context, if there is one. Essentially, await is trying to return us where we started from.
If there isn't a current context, it just uses the original TaskScheduler, which is usually thread pool.
The second (and third and so on...) await does the same.
Since the console applications typically have no synchronization context, continuations will typically be handled by the thread pool, so there is no issue if we block within the continuation.
WinForms or WPF, on the other hand, have synchronization context implemented on top of their message loop. Therefore, await executed on a UI thread will (eventually) execute its continuation on the UI thread as well. If we happen to block in the continuation, it will block the message loop and make the UI non-responsive until we unblock. OTOH, if we just await, it will neatly post continuations to be eventually executed on the UI thread, without ever blocking the UI thread.
In the following WinForms form, containing one button and one label, using await keeps the UI responsive at all times (note the async in front of the click handler):
public partial class Form1 : Form {
public Form1() {
InitializeComponent();
}
private async void button1_Click(object sender, EventArgs e) {
var result = await FooAsync();
label1.Text = result.ToString();
}
static async Task<int> FooAsync() {
var t1 = Method1Async();
var t2 = Method2Async();
var result1 = await t1;
var result2 = await t2;
return result1 + result2;
}
static Task<int> Method1Async() {
return Task.Run(
() => {
Thread.Sleep(3000);
return 11;
}
);
}
static Task<int> Method2Async() {
return Task.Run(
() => {
Thread.Sleep(5000);
return 22;
}
);
}
}
If we replaced the second await in FooAsync with t2.Result, it would continue to be responsive for about 3 seconds after the button click, and then freeze for about 2 seconds:
The continuation after the first await will politely wait its turn to be scheduled on the UI thread, which would happen after Method1Async() task finishes, i.e. after about 3 seconds,
at which point the t2.Result will rudely block the UI thread until the Method2Async() task finishes, about 2 seconds later.
If we removed the async in front of the button1_Click and replaced its await with FooAsync().Result it would deadlock:
The UI thread would wait on FooAsync() task to finish,
which would wait on its continuation to finish,
which would wait on the UI thread to become available,
which it isn't, since it is blocked by the FooAsync().Result.
The article "Await, SynchronizationContext, and Console Apps" by Stephen Toub was invaluable to me in understanding this subject.
Related
Related to this question: Does await completely blocks the thread?
[...] it will first check to see if the called method completed, and if not will register the continuation and return from that method call. Later, once that method completes, it will re-enter the state-machine in order to complete the method
And to this question also: When is the best place to use Task.Result instead of awaiting Task
await simply means "this workflow cannot progress further until this task is completed, so if it is not complete, find more work to do and come back later"
And finally to this post: https://blog.stephencleary.com/2012/02/async-and-await.html
If “await” sees that the awaitable has not completed, then it acts asynchronously. It tells the awaitable to run the remainder of the method when it completes, and then returns from the async method. Later on, when the awaitable completes, it will execute the remainder of the async method. If you’re awaiting a built-in awaitable (such as a task), then the remainder of the async method will execute on a “context” that was captured before the “await” returned.
So from these posts I get that the await operator does indeed not block, but when I've tried to test it i just cannot get this principle to work the way it states to work. Obviously I'm missing something:
//This will take 10 seconds
[HttpGet("test1")]
public async Task<TimeSpan> test()
{
var t1 = DateTime.Now;
var wait1 = DoAsyncEcho("The first!", 10000);
var wait2 = DoAsyncEcho("The second!", 10000);
_logger.LogInformation(await wait1);
_logger.LogInformation(await wait2);
_logger.LogInformation("DONE!");
var t2 = DateTime.Now;
return t2 - t1;
}
//This will take 10 seconds too
[HttpGet("test2")]
public async Task<TimeSpan> test2()
{
var t1 = DateTime.Now;
var wait1 = DoAsyncEcho("The first!", 10000);
var wait2 = DoAsyncEcho("The second!", 10000);
Thread.Sleep(10000);
_logger.LogInformation(await wait1);
_logger.LogInformation(await wait2);
_logger.LogInformation("DONE!");
var t2 = DateTime.Now;
return t2 - t1;
}
//This will take 20
[HttpGet("test3")]
public async Task<TimeSpan> test3()
{
var t1 = DateTime.Now;
var wait1 = await DoAsyncEcho("The first!", 10000);
var wait2 = await DoAsyncEcho("The second!", 10000);
_logger.LogInformation(wait1);
_logger.LogInformation(wait2);
_logger.LogInformation("DONE!");
var t2 = DateTime.Now;
return t2 - t1;
}
//This will take 30
[HttpGet("test4")]
public async Task<TimeSpan> test4()
{
var t1 = DateTime.Now;
var wait1 = await DoAsyncEcho("The first!", 10000);
var wait2 = await DoAsyncEcho("The second!", 10000);
Thread.Sleep(10000);
_logger.LogInformation(wait1);
_logger.LogInformation(wait2);
_logger.LogInformation("DONE!");
var t2 = DateTime.Now;
return t2 - t1;
}
private Task<string> DoAsyncEcho(string v, int t)
{
return Task<string>.Factory.StartNew(() =>
{
Thread.Sleep(t);
return v;
}
);
}
As I see from the methods test3 and test4, await does indeed wait, it does not enter into a state-machine and does a callback later on because it waits the full 10 seconds of the first DoAsyncEcho and then another 10s on the second call. On the methods test1 and test2 execution time lasts for 10s as code does not waits for the return of the DoAsyncEcho but only it awaits for the result later. Particulary test2 method sleeps the 3 calls of 10 seconds in parallel so after all it's just a 10s run.
What do I'm missing here?
I think the best way to demonstrate this is through a simple Windows Forms app.
Create a default Windows Forms app and drop 3 buttons onto it (called button1, button2 and button3.
Then add the following code:
async void button1_Click(object sender, EventArgs e)
{
this.Text = "[button1_Click] About to await slowMethodAsync()";
int result = await slowMethodAsync();
this.Text = "[button1_Click] slowMethodAsync() returned " + result;
}
void button2_Click(object sender, EventArgs e)
{
this.Text = "[button2_Click] About to start task to call slowMethod()";
int result = 0;
Task.Run(() =>
{
result = slowMethod();
}).ContinueWith(_ =>
{
this.Invoke(new Action(() =>
{
this.Text = "[button2_Click] slowMethod() returned " + result;
}));
});
}
void button3_Click(object sender, EventArgs e)
{
this.Text = "[button3_Click] About to call slowMethod()";
int result = slowMethod();
this.Text = "[button3_Click] slowMethod() returned " + result;
}
static async Task<int> slowMethodAsync()
{
await Task.Delay(5000);
return 42;
}
static int slowMethod()
{
Thread.Sleep(5000);
return 42;
}
If you try this code out, you will notice the following:
Pressing button1 will immediately change the title to [button1_Click] About to await Task.Delay(5000), and you can resize the dialog while waiting for 5 seconds, after which the title will change to [button1_Click] Awaited Task.Delay(5000).
The code for handling button2 is very roughly equivalent to the state machine that is generated from the await code for button1. If you press button2, you will see similar effects to pressing button1.
(The actual code for await is in reality quite different, but the underlying mechanism of using a continuation - i.e., ContinueWith() and Invoke(), to continue executing the code after the await on the UI thread illustrates its approach.)
The code for button3 completely blocks during the Thread.Sleep(), and if you press button3 the UI locks up completely for 5 seconds,.
To illustrate what happens with a non-UI example, consider the following console application:
using System;
using System.Threading;
using System.Threading.Tasks;
namespace Demo
{
static class Program
{
static async Task Main()
{
Console.WriteLine("Main thread ID = " + Thread.CurrentThread.ManagedThreadId);
int result = slowMethod();
Console.WriteLine("result = " + result);
Console.WriteLine("After calling slowMethod(), thread ID = " + Thread.CurrentThread.ManagedThreadId);
result = await slowMethodAsync();
Console.WriteLine("result = " + result);
Console.WriteLine("After calling slowMethodAsync(), thread ID = " + Thread.CurrentThread.ManagedThreadId);
}
static async Task<int> slowMethodAsync()
{
await Task.Delay(5000);
return 42;
}
static int slowMethod()
{
Thread.Sleep(5000);
return 42;
}
}
}
If you run that, you will see output similar to the following:
Main thread ID = 1
result = 42
After calling slowMethod(), thread ID = 1
result = 42
After calling slowMethodAsync(), thread ID = 4
Note how the code has resumed on a different thread after the await.
The key thing to realise is that as far as calling code is concerned, y = await X(); does not return until it has a value to return, and the code that runs afterwards may be running on a different thread.
The effect of this in terms of blocking THREADS is that the calling thread is freed up to go off and execute some other code, and another thread is only required when the async method returns.
In many cases, this means that no additional thread is required (for the continuation), and in all cases it means that the original calling thread is not blocked and can be freed up to the thread pool for use for another task.
This is the "non blocking" part of all this.
For a good, detailed explanation of why sometimes no additional thread is needed, read Stephen Cleary's excellent "There is no thread".
It seems that you are confusing two different interpretations of wait and block. The objective of asynchronous code is to block your code, while the thread remains unblocked. If you don't want to block your code, then the solution is easy: don't use await. But if you don't block your code, then you can't use the result of the asynchronous operation, because the asynchronous operation is running concurrently with your code.
What has not happened yet belongs to the future, and the future is unknown. Not only do you not know the result, you don't even know if the operation succeeded or failed. In most cases this is problematic. You need the result of the operation before continuing with processing this result. So you must block your code. And this is why await was invented, to block your code without having to block a thread too.
You need the thread to remain unblocked, so that it continues running the UI message pump that keeps your application responsive. Just because your code is blocked, your _application) need not to be blocked also. For ASP.NET applications, you need the thread to remain unblocked so that it can serve other incoming web requests. The fewer threads you block, the more requests you can serve. In this case async/await becomes a booster of scalability.
I'm experimenting with Tasks.
I have
private async Task<string> GetStringWithInnerCallConfigureAwaitFalseAsync()
{
await Task.Delay(3000).ConfigureAwait(false);
return "Finished!";
}
private async Task<string> GetStringAsync()
{
await Task.Delay(3000);
return "Finished!";
}
What's strange to me:
private void Button10_Click(object sender, RoutedEventArgs e)
{
Button10.Content = "GetAwaiter() GetResult() + deadlock";
var task = GetStringAsync().ConfigureAwait(false).GetAwaiter();
var result = task.GetResult(); // deadlock
Button10.Content = result;
}
I expected NO deadlock and crash on the last line because of non-UI context as it would be didn't I used GetAwaiter(), but I experience a deadlock
Next:
private void Button11_Click(object sender, RoutedEventArgs e)
{
Button11.Content = "GetAwaiter() GetResult() No deadlock";
var task = GetStringWithInnerCallConfigureAwaitFalseAsync().ConfigureAwait(false).GetAwaiter();
var result = task.GetResult(); // No deadlock
Button11.Content = result; // No crash
}
Here I expected crash on the last line because of non-UI context, but it works without issues.
Doesn't ConfigureAwait(false) make sense when we use GetAwaiter()?
ConfigureAwait configures the behaviour of the await keyword in the same expression. It does not affect other await statements in other methods, and it has no effect if you do not use await.
It controls whether the await statement captures the current SynchronizationContext (if there is one). In practice, if you run an await statement on the UI thread, then await task; will run the code after the await on the UI thread as well, whereas await task.ConfigureAwait(false) will run the code after the await on a ThreadPool thread.
In your first example:
private async Task<string> GetStringWithInnerCallConfigureAwaitFalseAsync()
{
await Task.Delay(3000).ConfigureAwait(false);
return "Finished!"; // <-- Run on the thread pool
}
private async Task<string> GetStringAsync()
{
await Task.Delay(3000);
return "Finished!"; // <-- Run on the captured SynchronizationContext (if any)
}
There is a difference in behaviour here, and that is which thread the return statement is run on.
In the first method, when the Task being awaited completes, a message is posted to the thread pool, which runs the return statement.
In the second method, the await statement captures the current SynchronizationContext (which refers to the UI thread), and uses this to run the return statement on. This means that a message is posted to the UI thread when 3 seconds have elapsed, telling it to run that return statement.
In your first snippet which calls GetStringAsync:
var task = GetStringAsync().ConfigureAwait(false).GetAwaiter();
The call to ConfigureAwait does nothing here, because you're not awaiting the result. You can remove it with no change.
var result = task.GetResult(); // deadlock
The UI thread is needed to run the return statement in GetStringAsync. Since you've blocked it in the call to GetResult(), it can't finish the GetStringAsync method, and so you've got a deadlock.
In your second snippet which calls GetStringWithInnerCallConfigureAwaitFalseAsync:
var task = GetStringWithInnerCallConfigureAwaitFalseAsync().ConfigureAwait(false).GetAwaiter();
Again, the call to ConfigureAwait(false) does nothing, because you're not awaiting the result.
var result = task.GetResult(); // No deadlock
This time, GetStringWithInnerCallConfigureAwaitFalseAsync does await ...ConfigureAwait(false), and so the code after the await is run on the thread pool and not the UI thread. Therefore the UI thread isn't needed to complete this method, and so you can safely (!) block it.
Button11.Content = result; // No crash
You called this method on the UI thread, and you've never moved off it - you call everything synchronously, there are no awaits, etc. Therefore you're still on the UI thread at this point
I'm still getting up to speed with async & multi threading. I'm trying to monitor when the Task I Start is still running (to show in a UI). However it's indicating that it is RanToCompletion earlier than I want, when it hits an await, even when I consider its Status as still Running.
Here is the sample I'm doing. It all seems to be centred around the await's. When it hits an await, it is then marked as RanToCompletion.
I want to keep track of the main Task which starts it all, in a way which indicates to me that it is still running all the way to the end and only RanToCompletion when it is all done, including the repo call and the WhenAll.
How can I change this to get the feedback I want about the tskProdSeeding task status?
My Console application Main method calls this:
Task tskProdSeeding;
tskProdSeeding = Task.Factory.StartNew(SeedingProd, _cts.Token);
Which the runs this:
private async void SeedingProd(object state)
{
var token = (CancellationToken)state;
while (!token.IsCancellationRequested)
{
int totalSeeded = 0;
var codesToSeed = await _myRepository.All().ToListAsync(token);
await Task.WhenAll(Task.Run(async () =>
{
foreach (var code in codesToSeed)
{
if (!token.IsCancellationRequested)
{
try
{
int seedCountByCode = await _myManager.SeedDataFromLive(code);
totalSeeded += seedCountByCode;
}
catch (Exception ex)
{
_logger.InfoFormat(ex.ToString());
}
}
}
}, token));
Thread.Sleep(30000);
}
}
If you use async void the outer task can't tell when the task is finished, you need to use async Task instead.
Second, once you do switch to async Task, Task.Factory.StartNew can't handle functions that return a Task, you need to switch to Task.Run(
tskProdSeeding = Task.Run(() => SeedingProd(_cts.Token), _cts.Token);
Once you do both of those changes you will be able to await or do a .Wait() on tskProdSeeding and it will properly wait till all the work is done before continuing.
Please read "Async/Await - Best Practices in Asynchronous Programming" to learn more about not doing async void.
Please read "StartNew is Dangerous" to learn more about why you should not be using StartNew the way you are using it.
P.S. In SeedingProd you should switch it to use await Task.Delay(30000); insetad of Thread.Sleep(30000);, you will then not tie up a thread while it waits. If you do this you likely could drop the
tskProdSeeding = Task.Run(() => SeedingProd(_cts.Token), _cts.Token);
and just make it
tskProdSeeding = SeedingProd(_cts.Token);
because the function no-longer has a blocking call inside of it.
I'm not convinced that you need a second thread (Task.Run or StartNew) at all. It looks like the bulk of the work is I/O-bound and if you're doing it asynchronously and using Task.Delay instead of Thread.Sleep, then there is no thread consumed by those operations and your UI shouldn't freeze. The first thing anyone new to async needs to understand is that it's not the same thing as multithreading. The latter is all about consuming more threads, the former is all about consuming fewer. Focus on eliminating the blocking and you shouldn't need a second thread.
As others have noted, SeedingProd needs to return a Task, not void, so you can observe its completion. I believe your method can be reduced to this:
private async Task SeedingProd(CancellationToken token)
{
while (!token.IsCancellationRequested)
{
int totalSeeded = 0;
var codesToSeed = await _myRepository.All().ToListAsync(token);
foreach (var code in codesToSeed)
{
if (token.IsCancellationRequested)
return;
try
{
int seedCountByCode = await _myManager.SeedDataFromLive(code);
totalSeeded += seedCountByCode;
}
catch (Exception ex)
{
_logger.InfoFormat(ex.ToString());
}
}
await Task.Dealy(30000);
}
}
Then simply call the method, without awaiting it, and you'll have your task.
Task mainTask = SeedingProd(token);
When you specify async on a method, it compiles into a state machine with a Task, so SeedingProd does not run synchronously, but acts as a Task even if returns void. So when you call Task.Factory.StartNew(SeedingProd) you start a task that kick off another task - that's why the first one finishes immediately before the second one. All you have to do is add the Task return parameter instead of void:
private async Task SeedingProdAsync(CancellationToken ct)
{
...
}
and call it as simply as this:
Task tskProdSeeding = SeedingProdAsync(_cts.Token);
Task.Run(()=>{}) puts the action delegate into the queue and returns the task .
Is there any benefit of having async/await within the Task.Run()?
I understand that Task.Run() is required since if we want to use await directly, then the calling method will need to be made Async and will affect the calling places.
Here is the sample code which has async await within Task.Run(). The full sample is provided here: Create pre-computed tasks.
Task.Run(async () => { await new WebClient().DownloadStringTaskAsync("");});
Alternatively this could have been done:
Task.Run(() => new WebClient().DownloadStringTaskAsync("").Result;);
Since both, Task.Run() and Await will queue the work and will be picked by the thread pool, could the async/await within the Task.Run() be a bit redundant?
Is there any benefit of having async/await within the Task.Run() ?
Yes. Task.Run runs some action on a thread-pool thread. If such action does some IO work and asynchronously waits for the IO operation to complete via await, then this thread-pool thread can be used by the system for other work while the IO operation is still running.
Example:
Task.Run( async () =>
{
DoSomeCPUIntensiveWork();
// While asynchronously waiting for this to complete,
// the thread is given back to the thread-pool
var io_result = await DoSomeIOOperation();
DoSomeOtherCPUIntensiveWork(io_result);
});
Is there any benefit of having async/await within the Task.Run()
An async method returns to the caller as soon as the first await is hit (that operates on a non-completed task). So if that first execution "streak" of an async method takes a long time Task.Run will alter behavior: It will cause the method to immediately return and execute that first "streak" on the thread-pool.
This is useful in UI scenarios because that way you can make 100% sure that you are not blocking the UI. Example: HttpWebRequestdoes DNS resolution synchronously even when you use one of the async methods (this is basically a library bug/design error). This can pause the UI thread. So you can use Task.Run to be 100% sure that the UI is never blocked for longer than a few microseconds.
So back to the original question: Why await inside a Task.Run body? For the same reason you normally await: To unblock the thread.
In the example that you linked the main thread is being blocked until the asynchronous operation is done. It's being blocked by calling Wait() (which by the way is generally a bad idea).
Let's have a look at the return from the DownloadStringAsync in the linked sample:
return Task.Run(async () =>
{
content = await new WebClient().DownloadStringTaskAsync(address);
cachedDownloads.TryAdd(address, content);
return content;
});
Why would you wrap this in a Task? Think about your options for a second. If you don't want to wrap this in a Task, how would you make sure the method returns a Task<string> and still have it work? You'd mark the method as async of course! However, if you mark your method as async and you call Wait on it, you'll most likely end up with a deadlock, since the main thread is waiting for the work to finish, and your blocking the main thread so it can't let you know it's done.
When marking a method as async, the state machine will run on the calling thread, in your example however, the state machine runs on a separate thread, meaning there is little to no work being done on the main thread.
Calling Async from Non-Async Methods
We do some stuff like that when we are trying to call an async method inside of a non-async method. Especially if the async method is a known quantity. We use more of a TaskFactory though ... fits a pattern, makes it easier to debug, makes sure everyone takes the same approach (and -- gives us one throat to choke if async-->sync starts acting buggy).
So, Your Example
Imagine, in your example, that you have a non-async function. And, within that function, you need to call await webClient.DoSomethingAsync(). You can't call await inside of a function that's not async -- the compiler won't let you.
Option 1: Zombie Infestation
Your first option is to crawl all the way back up your call stack, marking every da*n method along the way as async and adding awaits everywhere. Everywhere. Like, everywhere. Then -- since all those methods are now async, you need to make the methods that reference them all async.
This, btw, is probably the approach many of the SO enthusiasts are going to advocate. Because, "blocking a thread is a bad idea."
So. Yeah. This "let async be async" approach means your little library routine to get a json object just reached out and touched 80% of the code. Who's going to call the CTO and let him know?
Option 2: Just Go with the One Zombie
OR, you can encapsulate your async inside of some function like yours...
return Task.Run(async () => {
content = await new WebClient().DoSomethingAsync();
cachedDownloads.TryAdd(address, content);
return content;
});
Presto... the zombie infestation has been contained to a single section of code. I'll leave it to the bit-mechanics to argue over how/why that gets executed at the CPU-level. I don't really care. I care that nobody has to explain to the CTO why the entire library should now be 100% async (or something like that).
Confirmed, wrapping await with Task.Run use 2 threads instead of one.
Task.Run(async () => { //thread #1
await new WebClient().DownloadStringTaskAsync(""); //thread #2
});
Say you have four calls wrapped like this, it will use 4 x 2 = 8 threads.
It would be better to just call these with simple await instead. For example:
Task<byte[]> t1 = new WebClient().DownloadStringTaskAsync("");
Task<byte[]> t2 = new WebClient().DownloadStringTaskAsync("");
byte[] t1Result = await t1;
byte[] t2Result = await t2;
Here is the proof that wrapped Task.Run are using extra threads. (Not using WebClient to prove the point)
private static async Task wrapped()
{
List<Task> tasks = new List<Task>();
tasks.AddRange(new []
{
Task.Run(async() => await new MyThread().RunMe()),
Task.Run(async() => await new MyThread().RunMe()),
Task.Run(async() => await new MyThread().RunMe()),
Task.Run(async() => await new MyThread().RunMe()),
});
Thread.Sleep(1000);
int number = Process.GetCurrentProcess().Threads.Count;
Console.WriteLine($"While running thread count: {number}");
await Task.WhenAll(tasks);
}
Unwrapped
private static async Task unwrapped()
{
List<Task> tasks = new List<Task>();
Task<int> t1 = new MyThread().RunMe();
Task<int> t2 = new MyThread().RunMe();
Task<int> t3 = new MyThread().RunMe();
Task<int> t4 = new MyThread().RunMe();
tasks.AddRange(new[] {t1, t2, t3, t4});
Thread.Sleep(1000);
int number = Process.GetCurrentProcess().Threads.Count;
Console.WriteLine($"While running thread count: {number}");
int i1 = await t1;
int i2 = await t2;
int i3 = await t3;
int i4 = await t4;
}
Full POC code here
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Threading;
using System.Threading.Tasks;
namespace AsyncThreadDemo
{
class Program
{
static async Task Main(string[] args)
{
int number = Process.GetCurrentProcess().Threads.Count;
Console.WriteLine($"Init thread count: {number}");
//await wrapped();
await unwrapped();
number = Process.GetCurrentProcess().Threads.Count;
Console.WriteLine($"Done thread count: {number}");
Console.ReadLine();
}
private static async Task wrapped()
{
List<Task> tasks = new List<Task>();
tasks.AddRange(new []
{
Task.Run(async() => await new MyThread().RunMe()),
Task.Run(async() => await new MyThread().RunMe()),
Task.Run(async() => await new MyThread().RunMe()),
Task.Run(async() => await new MyThread().RunMe()),
});
Thread.Sleep(1000);
int number = Process.GetCurrentProcess().Threads.Count;
Console.WriteLine($"While running thread count: {number}");
await Task.WhenAll(tasks);
}
private static async Task unwrapped()
{
List<Task> tasks = new List<Task>();
Task<int> t1 = new MyThread().RunMe();
Task<int> t2 = new MyThread().RunMe();
Task<int> t3 = new MyThread().RunMe();
Task<int> t4 = new MyThread().RunMe();
tasks.AddRange(new[] {t1, t2, t3, t4});
Thread.Sleep(1000);
int number = Process.GetCurrentProcess().Threads.Count;
Console.WriteLine($"While running thread count: {number}");
int i1 = await t1;
int i2 = await t2;
int i3 = await t3;
int i4 = await t4;
}
}
public class MyThread
{
public static int _counter;
public async Task<int> RunMe()
{
await Task.Run(() =>
{
for (int i = 0; i < 2; ++i)
{
Thread.Sleep(1000);
Console.WriteLine($"T{Thread.CurrentThread.ManagedThreadId} {i}");
}
Console.WriteLine($"T{Thread.CurrentThread.ManagedThreadId} done");
});
return _counter++;
}
}
}
Apologies in advance if this question is opinion-based. The lack of Task.Yield version which wouldn't capture the execution context was already discussed here. Apparently, this feature was present in some form in early versions of Async CTP but was removed because it could easily be misused.
IMO, such feature could be as easily misused as Task.Run itself. Here's what I mean. Imagine there's an awaitable SwitchContext.Yield API which schedules the continuation on ThreadPool, so the execution will always continues on a thread different from the calling thread. I could have used it in the following code, which starts some CPU-bound work from a UI thread. I would consider it a convenient way of continuing the CPU-bound work on a pool thread:
class Worker
{
static void Log(string format, params object[] args)
{
Debug.WriteLine("{0}: {1}", Thread.CurrentThread.ManagedThreadId, String.Format(format, args));
}
public async Task UIAction()
{
// UI Thread
Log("UIAction");
// start the CPU-bound work
var cts = new CancellationTokenSource(5000);
var workTask = DoWorkAsync(cts.Token);
// possibly await for some IO-bound work
await Task.Delay(1000);
Log("after Task.Delay");
// finally, get the result of the CPU-bound work
int c = await workTask;
Log("Result: {0}", c);
}
async Task<int> DoWorkAsync(CancellationToken ct)
{
// start on the UI thread
Log("DoWorkAsync");
// switch to a pool thread and yield back to the UI thread
await SwitchContext.Yield();
Log("after SwitchContext.Yield");
// continue on a pool thread
int c = 0;
while (!ct.IsCancellationRequested)
{
// do some CPU-bound work on a pool thread: counting cycles :)
c++;
// and use async/await too
await Task.Delay(50);
}
return c;
}
}
Now, without SwitchContext.Yield, DoWorkAsync would look like below. It adds some extra level of complexity in form of async delegate and task nesting:
async Task<int> DoWorkAsync(CancellationToken ct)
{
// start on the UI thread
Log("DoWorkAsync");
// Have to use async delegate
// Task.Run uwraps the inner Task<int> task
return await Task.Run(async () =>
{
// continue on a pool thread
Log("after Task.Yield");
int c = 0;
while (!ct.IsCancellationRequested)
{
// do some CPU-bound work on a pool thread: counting cycles :)
c++;
// and use async/await too
await Task.Delay(50);
}
return c;
});
}
That said, implementing SwitchContext.Yield may actually be quite simple and (I dare to say) efficient:
public static class SwitchContext
{
public static Awaiter Yield() { return new Awaiter(); }
public struct Awaiter : System.Runtime.CompilerServices.INotifyCompletion
{
public Awaiter GetAwaiter() { return this; }
public bool IsCompleted { get { return false; } }
public void OnCompleted(Action continuation)
{
ThreadPool.QueueUserWorkItem((state) => ((Action)state)(), continuation);
}
public void GetResult() { }
}
}
So, my question is, why should I prefer the second version of DoWorkAsync over the first one, and why would using SwitchContext.Yield be considered a bad practice?
You don't have to put the Task.Run in DoWorkAsync. Consider this option:
public async Task UIAction()
{
// UI Thread
Log("UIAction");
// start the CPU-bound work
var cts = new CancellationTokenSource(5000);
var workTask = Task.Run(() => DoWorkAsync(cts.Token));
// possibly await for some IO-bound work
await Task.Delay(1000);
Log("after Task.Delay");
// finally, get the result of the CPU-bound work
int c = await workTask;
Log("Result: {0}", c);
}
This results in code with much clearer intent. DoWorkAsync is a naturally synchronous method, so it has a synchronous signature. DoWorkAsync neither knows nor cares about the UI. The UIAction, which does care about the UI thread, pushes off the work onto a background thread using Task.Run.
As a general rule, try to "push" any Task.Run calls up out of your library methods as much as possible.