Cold Elm Coders Posts Rants Articles Stories About
Published: 06-Feb-2021
Updated: 06-Feb-2021
Author: Shaun Curtis

Table of Contents

Async Programming in DotNetCore

A Practical Examination of Async Programming in DotNetCore..

My first article on this subject provided an overview of async programming in DotNetCore and explained some of the key concepts. You'll find the article here. This article takes a practical approach to demonstrating some of of those key concepts, and introducing more complex coding patterns. The article is based around a DotNetCore console application.

You'll need a DotNetCore compatible development environment, normally either either Visual Studio or Visual Code, and a copy of the Repo associated with this project to run the code.

DISCLAIMER - The code is Experimental, not Production. Designed to be concise with minimal error trapping and handling to keep it easy to read and understand. Classes are kept simple for the same reason.

Code Repository

The code in available in a GitHub Repo here. The code for this project is in Async-Demo. Ignore any other projects - they are for a further Async Programming article.

Library Classes

Before we start your need to be aware of two helper classses

  1. LongRunningTasks - emulates work.
    1. RunLongProcessorTaskAsync and RunLongProcessorTask use prime number calculations to emulate a processor heavy task.
    2. RunYieldingLongProcessorTaskAsync is a version that yields every 100 calculations.
    3. RunLongIOTaskAsync uses Task.Delay to emulate a slow I/O operations.
  2. UILogger provides an abstraction layer for logging information to the UI. You pass a delegate Action to the methods. UILogger builds the message, and then calls the Action to actually write it to wherever the Action is configured to write to. In our case LogToConsole in Program, which runs Console.WriteLine. It could just as easily write to a text file.

Getting Started

Our first challenge is the switch from sync to async.

Make sure you're running the correct framework and latest language version. (C# 7.1 onwards supports a Task based Main).

  <PropertyGroup>
    <OutputType>Exe</OutputType>
    <TargetFramework>net5</TargetFramework>
    <LangVersion>latest</LangVersion>
    <RootNamespace>Async_Demo</RootNamespace>
  </PropertyGroup>

Pre #7.1, Main could only run synchronously, and you needed a "NONO", using Wait, to prevent Main dropping out the bottom and closing the program. Post #7.1, declare Main to return a Task.

The async Main pattern is shown below. Declaring async depends on whether on not there's an await in the code

// With await
static async Task Main(string[] args)
{
    // code
    // await somewhere in here
}

// No awaits
static Task Main(string[] args)
{
    // code
    // no awaits
    return Task.CompletedTask;
}

Note:

  1. If you use the async keyword but don't have an await, the compiler warns, but then compiles anyway, treating the method as synchronous code.
  2. You can't declare a method as async and return a Task. You simply return the correct value and the compiler will do all the donkey work.

So let's run some code. Our first run:

static Task Main(string[] args)
{
    var watch = new Stopwatch();
    watch.Start();
    UILogger.LogThreadType(LogToConsole, "Main");
    var millisecs = LongRunningTasks.RunLongProcessorTask(5);
    watch.Stop();
    UILogger.LogToUI(LogToConsole, $"Main ==> Completed in { watch.ElapsedMilliseconds} milliseconds", "Main");
    return Task.CompletedTask;
}

The Task ran synchronously as expected. A bunch of synchronous code inside a Task. No yielding.

[11:35:32][Main Thread][Main] >  running on Application Thread
[11:35:32][Main Thread][LongRunningTasks] > ProcessorTask started
[11:35:36][Main Thread][LongRunningTasks] > ProcessorTask completed in 3399 millisecs
[11:35:36][Main Thread][Main] > Main ==> Completed in 3523 milliseconds
Press any key to close this window . . .

Our second run:

static async Task Main(string[] args)
{
    var watch = new Stopwatch();
    watch.Start();
    UILogger.LogThreadType(LogToConsole, "Main");
    var millisecs = await LongRunningTasks.RunLongProcessorTaskAsync(5, LogToConsole);
    UILogger.LogToUI(LogToConsole, $"Yielded to Main", "Main");
    watch.Stop();
    UILogger.LogToUI(LogToConsole, $"Main ==> Completed in { watch.ElapsedMilliseconds} milliseconds", "Main");
}

The Task ran synchronously - no yielding. Logical because there was no reason to yield. RunLongProcessorTaskAsync is a synchronous bunch of code inside a Task - calculating prime numbers - so it ran to completion. The await is redundant, it may be a Task but it doesn't yield, so never gives up the thread until complete.

[11:42:43][Main Thread][Main] >  running on Application Thread
[11:42:43][Main Thread][LongRunningTasks] > ProcessorTask started
[11:42:46][Main Thread][LongRunningTasks] > ProcessorTask completed in 3434 millisecs
[11:42:46][Main Thread][Main] > Yielded
[11:42:46][Main Thread][Main] > Main ==> Completed in 3593 milliseconds

Our third run:

static async Task Main(string[] args)
{
    var watch = new Stopwatch();
    watch.Start();
    UILogger.LogThreadType(LogToConsole, "Main");
    var millisecs = LongRunningTasks.RunYieldingLongProcessorTaskAsync(5, LogToConsole);
    UILogger.LogToUI(LogToConsole, $"Yielded to Main", "Main");
    watch.Stop();
    UILogger.LogToUI(LogToConsole, $"Main ==> Completed in { watch.ElapsedMilliseconds} milliseconds", "Main");
}

Before we look at the result, let's look at the difference between RunLongProcessorTaskAsync and RunYieldingLongProcessorTaskAsync. We've added a Task.Yield() to yield control every 100 primes.

if (isPrime)
{
    counter++;
    // only present in Yielding version
    if (counter > 100)
    {
        await Task.Yield();
        counter = 0;
    }
}

The long running task didn't complete. RunYieldingLongProcessorTaskAsync yielded back to Main after the first 100 primes had been calculated - a little short of 173 millisecs - and Main ran to completion during the yield.

[12:13:56][Main Thread][Main] >  running on Application Thread
[12:13:56][Main Thread][LongRunningTasks] > ProcessorTask started
[12:13:57][Main Thread][Main] > Yielded to Main
[12:13:57][Main Thread][Main] > Main ==> Completed in 173 milliseconds

If we update Main to await the long processor task

    var millisecs = await LongRunningTasks.RunYieldingLongProcessorTaskAsync(5, LogToConsole);

It runs to completion. Although it yields, we await on the RunYieldingLongProcessorTaskAsync Task to complete, before moving on in Main. There's another important point to note here. Look at which thread the long running task ran on, and compare it to previous runs. It jumped to a new thread [LongRunningTasks Thread] after starting on [Main Thread].

[12:45:10][Main Thread:1][Main] >  running on Application Thread
[12:45:11][Main Thread:1][LongRunningTasks] > ProcessorTask started
[12:45:14][LongRunningTasks Thread:7][LongRunningTasks] > ProcessorTask completed in 3892 millisecs
[12:45:14][LongRunningTasks Thread:7][Main] > Yielded to Main
[12:45:14][LongRunningTasks Thread:7][Main] > Main ==> Completed in 4037 milliseconds

Add a quick Console.Write in RunYieldingLongProcessorTaskAsync to see which thread each yielded iteration runs on - writing the ManagedThreadId.

counter++;
if (counter > 100)
{
    Console.WriteLine($"Thread ID:{Thread.CurrentThread.ManagedThreadId}");
    await Task.Yield();
    counter = 0;
}

The result is shown below. Notice the regular thread jumping. Yield creates a new continuation Task, and schedules it to run asynchronously. On the first Task.Yield the application thread scheduler passes the new Task to the application pool, and for then on the application pool Scheduler makes decisions on where to run Tasks.

Task.Yield(), to quote Microsoft "Creates an awaitable task that asynchronously yields back to the current context when awaited." I translate that to mean it's syntactic sugar for yielding control up the tree and creating a continuation Task that gets posted back to the Scheduler to run when it schedules it. To quote further "A context that, when awaited, will asynchronously transition back into the current context at the time of the await." In other words, it doesn't await unless you tell it to. Hit the first yield in the continuation and processing trucks on through to the code below Task.Yield(). I've tested it.

[12:38:16][Main Thread:1][Main] >  running on Application Thread
[12:38:16][Main Thread:1][LongRunningTasks] > ProcessorTask started
Thread ID:1
Thread ID:4
Thread ID:4
Thread ID:6
Thread ID:6
Thread ID:7

Finally, change over to the RunLongIOTaskAsync long running task.

    var millisecs = await LongRunningTasks.RunLongIOTaskAsync(5, LogToConsole);

If you don't await, the same as before:

[14:26:46][Main Thread:1][Main] >  running on Application Thread
[14:26:47][Main Thread:1][LongRunningTasks] > IOTask started
[14:26:47][Main Thread:1][Main] > Yielded to Main
[14:26:47][Main Thread:1][Main] > Main ==> Completed in 322 milliseconds

And if you await it runs to completion, again with the thread switch.

[14:27:16][Main Thread:1][Main] >  running on Application Thread
[14:27:16][Main Thread:1][LongRunningTasks] > IOTask started
[14:27:21][LongRunningTasks Thread:4][LongRunningTasks] > IOTask completed in 5092 millisecs
[14:27:21][LongRunningTasks Thread:4][Main] > Yielded to Main
[14:27:21][LongRunningTasks Thread:4][Main] > Main ==> Completed in 5274 milliseconds

More Complexity

Ok, now to move closer to reality and code doing something.

JobRunner

JobRunner is a simple class to run and control asynchronous jobs. For our purposes, it runs one of the long running tasks to simulate work, but you can use the basic pattern for real world situations.

It's self-evident what most of the code does, but I'll introduce TaskCompletionSource.

To quote MS "Represents the producer side of a Task<TResult> unbound to a delegate, providing access to the consumer side through the Task property." You get a Task exposed by TaskCompletionSource.Task that you control through the TaskCompletionSource instance - in other words, a manually controlled Task uncoupled from the method.

The Task that represents the state of the JobRunner is exposed as the JobTask property. If the underlying TaskCompletionSource isn't set it returns a simple Task.CompletedTask object, otherwise it returns the Task of JobTaskController. The Run method uses the async event pattern - we need a block of code that runs asynchronously, yielding control with await. Run controls the Task state, but the Task itself is independant of Run. IsRunning ensures you can't start the job once it's running.

class JobRunner
{
    public enum JobType { IO, Processor, YieldingProcessor } 

    public JobRunner(string name, int secs, JobType type = JobType.IO)
    {
        this.Name = name;
        this.Seconds = secs;
        this.Type = type;
    }

    public string Name { get; private set; }
    public int Seconds { get; private set; }
    public JobType Type { get; set; }
    private bool IsRunning;

    public Task JobTask => this.JobTaskController == null ? Task.CompletedTask : this.JobTaskController.Task;
    private TaskCompletionSource JobTaskController { get; set; } = new TaskCompletionSource();

    public async void Run()
    {
        if (!this.IsRunning) {
            this.IsRunning = true;
            this.JobTaskController = new TaskCompletionSource();
            switch (this.Type)
            {
                case JobType.Processor:
                    await LongRunningTasks.RunLongProcessorTaskAsync(Seconds, Program.LogToConsole, Name);
                    break;
                    
                case JobType.YieldingProcessor:
                    await LongRunningTasks.RunYieldingLongProcessorTaskAsync(Seconds, Program.LogToConsole, Name);
                    break;

                default:
                    await LongRunningTasks.RunLongIOTaskAsync(Seconds, Program.LogToConsole, Name);
                    break;
            }

            this.JobTaskController.TrySetResult();
            this.IsRunning = false;
        }
    }
}

JobScheduler

JobScheduler is the method used to actually schedule the jobs. It's separated from Main to demonstrate some key behaviours of async programming.

  1. Stopwatch provides timing.
  2. Creates four different IO jobs.
  3. Starts the four jobs.
  4. Uses Task.WhenAll to wait on certain tasks before continuing. Note the Tasks are the JobTasks exposed by the JobRunnner instances.

WhenAll is one of several static Task methods. WhenAll creates a single Task which awaits all the Tasks in the submitted array. It's status will change to Complete when all the Tasks complete. WhenAny is similar, but will be set to Complete when any are complete. They could be named AwaitAll and AwaitAny. WaitAll and WaitAny are blocking versions and similar to Wait. Not sure about the reasons for the slightly confusing naming conversion - I'm sure there was one.

static async Task JobScheduler()
{
    var watch = new Stopwatch();
    watch.Start();
    var name = "Job Scheduler";
    var quickjob = new JobRunner("Quick Job", 3);
    var veryslowjob = new JobRunner("Very Slow Job", 7);
    var slowjob = new JobRunner("Slow Job", 5);
    var veryquickjob = new JobRunner("Very Quick Job", 2);
    quickjob.Run();
    veryslowjob.Run();
    slowjob.Run();
    veryquickjob.Run();
    UILogger.LogToUI(LogToConsole, $"All Jobs Scheduled", name);
    await Task.WhenAll(new Task[] { quickjob.JobTask, veryquickjob.JobTask }); ;
    UILogger.LogToUI(LogToConsole, $"Quick Jobs completed in {watch.ElapsedMilliseconds} milliseconds", name);
    await Task.WhenAll(new Task[] { slowjob.JobTask, quickjob.JobTask, veryquickjob.JobTask, veryslowjob.JobTask }); ;
    UILogger.LogToUI(LogToConsole, $"All Jobs completed in {watch.ElapsedMilliseconds} milliseconds", name);
    watch.Stop();
}

We now need to make some changes to Main:

static async Task Main(string[] args)
{
    var watch = new Stopwatch();
    watch.Start();
    UILogger.LogThreadType(LogToConsole, "Main");
    var task = JobScheduler();
    UILogger.LogToUI(LogToConsole, $"Job Scheduler yielded to Main", "Main");
    await task;
    UILogger.LogToUI(LogToConsole, $"final yield to Main", "Main");
    watch.Stop();
    UILogger.LogToUI(LogToConsole, $"Main ==> Completed in { watch.ElapsedMilliseconds} milliseconds", "Main");

    //return Task.CompletedTask;
}

When you run this you get the output below. The interesting bits to note are:

  1. Each of the jobs start, and then yield at their first await, passing control back to the caller - in this case JobSchedular.
  2. JobScheduler runs to it's first await and yields back to Main.
  3. When the first two jobs finish their JobTask is set to complete and JobScheduler continues to the next await.
  4. JobScheduler completes in a little over the time needed to run the longest Job.
[16:58:52][Main Thread:1][Main] >  running on Application Thread
[16:58:52][Main Thread:1][LongRunningTasks] > Quick Job started
[16:58:52][Main Thread:1][LongRunningTasks] > Very Slow Job started
[16:58:52][Main Thread:1][LongRunningTasks] > Slow Job started
[16:58:52][Main Thread:1][LongRunningTasks] > Very Quick Job started
[16:58:52][Main Thread:1][Job Scheduler] > All Jobs Scheduled
[16:58:52][Main Thread:1][Main] > Job Scheduler yielded to Main
[16:58:54][LongRunningTasks Thread:4][LongRunningTasks] > Very Quick Job completed in 2022 millisecs
[16:58:55][LongRunningTasks Thread:4][LongRunningTasks] > Quick Job completed in 3073 millisecs
[16:58:55][LongRunningTasks Thread:4][Job Scheduler] > Quick Jobs completed in 3090 milliseconds
[16:58:57][LongRunningTasks Thread:4][LongRunningTasks] > Slow Job completed in 5003 millisecs
[16:58:59][LongRunningTasks Thread:6][LongRunningTasks] > Very Slow Job completed in 7014 millisecs
[16:58:59][LongRunningTasks Thread:6][Job Scheduler] > All Jobs completed in 7111 milliseconds
[16:58:59][LongRunningTasks Thread:6][Main] > final yield to Main
[16:58:59][LongRunningTasks Thread:6][Main] > Main ==> Completed in 7262 milliseconds

Now change the job type over to Processor as below:

var quickjob = new JobRunner("Quick Job", 3, JobRunner.JobType.Processor);
var veryslowjob = new JobRunner("Very Slow Job", 7, JobRunner.JobType.Processor);
var slowjob = new JobRunner("Slow Job", 5, JobRunner.JobType.Processor);
var veryquickjob = new JobRunner("Very Quick Job", 2, JobRunner.JobType.Processor);

When you run this, you'll see everything is run sequentially on the Main Thread. At first you think why? We have more than one thread available and the Scheduler has demonstrated it's ability to switch tasks between threads. Why isn't it switching?

The answer is very simple. Once we initialise the JobRunnner object we run them in to the Scheduler one at a time. As the code we run is sequential - calculating primes without breaks - we don't execute the next line of code (feeding in the second job) until the first job completes.

[17:59:48][Main Thread:1][Main] >  running on Application Thread
[17:59:48][Main Thread:1][LongRunningTasks] > Quick Job started
[17:59:53][Main Thread:1][LongRunningTasks] > Quick Job completed in 4355 millisecs
[17:59:53][Main Thread:1][LongRunningTasks] > Very Slow Job started
[17:59:59][Main Thread:1][LongRunningTasks] > Very Slow Job completed in 6057 millisecs
[17:59:59][Main Thread:1][LongRunningTasks] > Slow Job started
[18:00:03][Main Thread:1][LongRunningTasks] > Slow Job completed in 4209 millisecs
[18:00:03][Main Thread:1][LongRunningTasks] > Very Quick Job started
[18:00:05][Main Thread:1][LongRunningTasks] > Very Quick Job completed in 1737 millisecs
[18:00:05][Main Thread:1][Job Scheduler] > All Jobs Scheduled
[18:00:05][Main Thread:1][Job Scheduler] > Quick Jobs completed in 16441 milliseconds
[18:00:05][Main Thread:1][Job Scheduler] > All Jobs completed in 16441 milliseconds
[18:00:05][Main Thread:1][Main] > Job Scheduler yielded to Main
[18:00:05][Main Thread:1][Main] > final yield to Main
[18:00:05][Main Thread:1][Main] > Main ==> Completed in 16591 milliseconds

Now, change the jobs over to run YieldingProcessor.

var quickjob = new JobRunner("Quick Job", 3, JobRunner.JobType.YieldingProcessor);
var veryslowjob = new JobRunner("Very Slow Job", 7, JobRunner.JobType.YieldingProcessor);
var slowjob = new JobRunner("Slow Job", 5, JobRunner.JobType.YieldingProcessor);
var veryquickjob = new JobRunner("Very Quick Job", 2, JobRunner.JobType.YieldingProcessor);

The result is very different. The time taken will depend on the number of processor cores and threads on your computer. You can see all the jobs start quickly and completion in 11 seconds, with the slowest job taking 9 seconds. The key difference here is that the processor long running job yields regularly. This gives the Scheduler a chance to divy out out the work to other threads.

Yielding Processor code

[17:50:12][Main Thread:1][Main] >  running on Application Thread
[17:50:12][Main Thread:1][LongRunningTasks] > Quick Job started
[17:50:12][Main Thread:1][LongRunningTasks] > Very Slow Job started
[17:50:12][Main Thread:1][LongRunningTasks] > Slow Job started
[17:50:12][Main Thread:1][LongRunningTasks] > Very Quick Job started
[17:50:12][Main Thread:1][Job Scheduler] > All Jobs Scheduled
[17:50:12][Main Thread:1][Main] > Job Scheduler yielded to Main
[17:50:16][LongRunningTasks Thread:7][LongRunningTasks] > Very Quick Job completed in 4131 millisecs
[17:50:18][LongRunningTasks Thread:7][LongRunningTasks] > Quick Job completed in 6063 millisecs
[17:50:18][LongRunningTasks Thread:7][Job Scheduler] > Quick Jobs completed in 6158 milliseconds
[17:50:21][LongRunningTasks Thread:6][LongRunningTasks] > Slow Job completed in 9240 millisecs
[17:50:23][LongRunningTasks Thread:9][LongRunningTasks] > Very Slow Job completed in 11313 millisecs
[17:50:23][LongRunningTasks Thread:9][Job Scheduler] > All Jobs completed in 11411 milliseconds
[17:50:23][LongRunningTasks Thread:9][Main] > final yield to Main
[17:50:23][LongRunningTasks Thread:9][Main] > Main ==> Completed in 11534 milliseconds

Conclusions and Wrap Up.

Hopefully helpful/informative? Some of the key points that I've learned in my voyage down the async road, and are demonstrated here are:

  1. Async and Await All The Way. Don't mix synchronous and asynchronous methods. Start at the bottom - the data or process interface - and code async all the way up though the data and business/logic layers to the UI.
  2. You can't run asynchronously if you don't yield. You've got to give the task schedulers a chance! Wrapping a few synchronous routines in Task is talking-the-talk not walking-the-walk.
  3. Fire and forget void return methods need to yield to pass control back to the caller. They are no different to Task returning methods in their behaviour. They just don't return a Task for you to await or monitor progress.
  4. If you're writing processor intensive activities - modelling, big numbercrunching,.. make sure to make them async with plenty of yielding at appropriate places.
  5. ONLY use Task.Run in the UI, right up at the top of the call stack. NEVER EVER use it in libraries. And don't use it at all unless you have a solid reason.
  6. Use logging and breakpoints on awaits to see when you hit them. How quickly your code falls back to the outside await is a very good indicator of responsiveness. Take out your outside await and see how quickly you drop out the bottom!
  7. You may have noticed no ContinueWith. I don't often use it. Normally a simple await followed by continuation code achieves the same result. I've read commentary that it's heavier on processing, because it creates a new task whereas await/continuation reuses the same Task. I haven't delved deeply enough into the code yet to check.
  8. Always use Async and Await, don't get fancy.
  9. If your library provides both async and sync calls, code them separately. "Code it once" best practice doesn't apply here. NEVER call one from the other if you don't want to shoot yourself in the foot at some point!