Tasks - Branch Execution

Date: 04/26/2021

The Users Challenge

I have to await two (maybe more) Tasks that call 3rd party services and those calls are really slow. They don't depend on each other... is there anything I can do to speed things up from my side?

Task-based Branch Execution

Eventually most C# developers learn about async and the corresponding keyword await. It is a rite of passage!

The use case 99% of the time is that there is an async Function returning a Task (maybe Task<T>) that you know you have to await. I admit that this is not as clean as golang and goroutines but C# Tasks and Task<T> are still incredibly powerful concepts that when used right, can really help with performance not just responsiveness.

Let's take the mental handcuffs off of how you typically see the standard Task usage for a second. If we are allowed to modify the following bit of code we can improve the overall performance without knowing much - if anything - of what's going on beneath the hood.

using System;
using System.Collections.Generic;
using System.Threading.Tasks;

namespace Tasks
{
    public static class Program
    {
        public static async Task Main()
        {
            Console.WriteLine("Start!");

            // Executing an async Task and awaiting.
            await ProcessMessageAsync("test0"); // this finishes first, we wait here till that happens.
            await ProcessMessageAsync("test1"); // this finishes second, we wait here till that happens.
        } // we make it here when everything else finishes.

        public static async Task ProcessMessageAsync(string input)
        {
            await Task.Yield(); // prevents this from executing synchronously
            await Console.Out.WriteLineAsync($"Processing Message: {input}");
        }
    }
}

We normally use async and await to execute an expensive call without blocking the calling thread. You see above, that we have the two async operations (Tasks) being properly awaited by the developer and we are assuming they are slow. There is nothing wrong with the internal code (another assumption) and we have to await on each one to finish before leaving the method.

They are written now as non-blocking, but they will execute in order sequentially.

What makes everything proceed in an orderly fashion is the use of await. The developer achieved the good design of non-blocking calling threads but there is no concurrency due to the same mechanism: the await. Which leads me to: there is no rule you have to await code that is async.

WARNING: That also doesn't mean go batshit crazy not using await in your code.

What I really should clarify is, you don't always have to await here like where we did in the above example. await ensures that the execution finishes and that it is also not lost to the GC ether. In other words, we want to use await and not using it generally causes unintended nasty side-effects (like code not even executing)!

Here though, let us imagine if we did not use await? What would happen assuming ProcessMessageAsync was a schedulable operation? As soon as the code finished invoking the first ProcessMessageAsync, it would begin to invoke the next line of code without stopping or awaiting.

If using await ensures the integrity of our executions, where do we put await?  
I still need it right? 

Yes!

By altering the example, we can invoke (start) both Tasks (that are independent of each other) concurrently, store a reference to these operations into a local variable (called task1 etc.), then use those references as inputs to Task.WhenAll() allowing us to await till both are finished.

TL;DR

1. Branch out the execution of our two (or more) methods.
2. Then "rejoin" to this execution context with Task.WhenAll().

using System;
using System.Collections.Generic;
using System.Threading.Tasks;

namespace Tasks
{
    public static class Program
    {
        public static async Task Main()
        {
            Console.WriteLine("Start!");

            // Tasks holding the operation and future (result). 
            var task0 = ProcessMessageAsync("test0"); // begin executing (branch #0)
            var task1 = ProcessMessageAsync("test1"); // begin executing (branch #1)
            
            // Non-blocking await - we wait here with out blocking the thread till all the input Tasks have Completed.
            await Task.WhenAll(task1, task2);
        }

        public static async Task ProcessMessageAsync(string input)
        {
            await Task.Yield(); // prevents this from executing synchronously
            await Console.Out.WriteLineAsync($"Processing Message: {input}");
        }
    }
}

If task0 usually takes 30 seconds and task1 is usually 30 seconds, our starting example code would take a total time of 60 seconds.

By re-arranging when we call the await (till after they have both started executing) we now have a task0 taking 30 seconds and task1 taking 30 seconds concurrently. Our total execution time is now only 30 seconds (or which ever of the tasks was longest) as they happened concurrently.

This doesn't always occur...

This is the general use case, but this isn't a guarantee of execution. This code operates more like an instruction/suggestion. The execution concurrently is not fully guaranteed. For a lot of use cases this is exactly how it works, but it is based on how busy the TaskScheduler/ThreadPool is or if it determines that this thing should execute immediately.

That is a rather complex concept and worth a whole separate and detailed article. As long as the Task is properly async, able to be scheduled, then these can execute concurrently. Some examples would be a call out to a Web.Api, a save to a database etc.

Note: To demonstrate sync execution immediately, you can remove the await Task.Yield(); from ProcessMessageAsync. You will then see it execute in order.

What happens when you have more than two tasks?

Well you can use the same concept of await Task.WhenAll(). Here is what that could look like.

using System;
using System.Collections.Generic;
using System.Threading.Tasks;

namespace Tasks
{
    public static class Program
    {
        public static async Task Main()
        {
            Console.WriteLine("Start!");

            // All my messages
            var myStrings = new List<string>
            {
                "test0",
                "test1",
                "test2",
                "test3",
                "test4",
                "test5",
                "test6",
                "test7",
                "test8",
                "test9"
            };

            // All my messages, assigned to a Task assigned inside an array.
            // Execution of each task begins when created but there is no blocking here.
            var tasks = new Task[myStrings.Count];
            for (int i = 0; i < myStrings.Count; i++)
            {
                tasks[i] = ProcessMessageAsync(myStrings[i]);
            }

            // Non-blocking await till all tasks are finished.
            await Task.WhenAll(tasks);
        }

        public static async Task ProcessMessageAsync(string input)
        {
            await Task.Yield(); // prevents this from executing synchronously
            await Console.Out.WriteLineAsync($"Processing Message: {input}");
        }
    }
}

The same possible decrease on execution time is possible, leading to significantly reduced total execution time. This is really only true though when are able to create indepenent execution "branches" and the code is not dependent on the previous task having to finish.

Conclusion

Sometimes, it just takes being mindful of how you use async and await to greatly increase execution performance. Other times, it requires heavy refactoring. Production scenarios are rarely ever as easy as the above scenario demonstrates.

Some weaknesses to this strategy are:

1. Not all code will execute concurrently/parallely.
   1. This is due to advanced scheduling algorithm or for a variety of reasons like a hot for loop preventing scheduling.
      1. There are mechanisms that force scheduling to occur such as Task.Run or Task.Yield and will make the execution occur on a background thread.
2. If your workload is not even, you will be prone to burst resource utilization/traffic.
   1. This means that there can overhall hiccups in performance, or bottlenecks on unrelated portions of the application.
      1. The execution resources are shared application wide unless you have created a custom and independent TaskScheduler.
3. Exceptions can stop the execution of remaining tasks.
   1. This may be desireable though.

Example #3 would be best handled with a different approach because the context of the situation changes having a lot more work to await.

Consider something like ParallelForEachAsync instead.

Links

• Microsoft - Task Class
• Microsoft - Task.WhenAll Method
• Microsoft - IAsyncResult
• Microsoft - Task-based asynchronous programming

Have an issue? Make an Edit and Pull Request here :)