DevTut

# Task Parallel Library (TPL)

# Basic producer-consumer loop (BlockingCollection)

var collection = new BlockingCollection<int>(5);
var random = new Random();

var producerTask = Task.Run(() => {
    for(int item=1; item<=10; item++) 
    {
        collection.Add(item);
        Console.WriteLine("Produced: " + item);
        Thread.Sleep(random.Next(10,1000));
    }
    collection.CompleteAdding();
    Console.WriteLine("Producer completed!");
});

It is worth noting that if you do not call collection.CompleteAdding();, you are able to keep adding to the collection even if your consumer task is running. Just call collection.CompleteAdding(); when you are sure there are no more additions. This functionality can be used to make a Multiple Producer to a Single Consumer pattern where you have multiple sources feeding items into the BlockingCollection and a single consumer pulling items out and doing something with them. If your BlockingCollection is empty before you call complete adding, the Enumerable from collection.GetConsumingEnumerable() will block until a new item is added to the collection or BlockingCollection.CompleteAdding(); is called and the queue is empty.

var consumerTask = Task.Run(() => {
    foreach(var item in collection.GetConsumingEnumerable())
    {
        Console.WriteLine("Consumed: " + item);
        Thread.Sleep(random.Next(10,1000));
    }
    Console.WriteLine("Consumer completed!");
});
  
Task.WaitAll(producerTask, consumerTask);
       
Console.WriteLine("Everything completed!");

# Parallel.Invoke

var actions = Enumerable.Range(1, 10).Select(n => new Action(() =>
{
    Console.WriteLine("I'm task " + n);
    if((n & 1) == 0)
        throw new Exception("Exception from task " + n);
})).ToArray();

try
{
    Parallel.Invoke(actions);
}
catch(AggregateException ex)
{
    foreach(var inner in ex.InnerExceptions)
        Console.WriteLine("Task failed: " + inner.Message);
}

# Task: basic instantiation and Wait

A task can be created by directly instantiating the Task class...

var task = new Task(() =>
{
    Console.WriteLine("Task code starting...");
    Thread.Sleep(2000);
    Console.WriteLine("...task code ending!");
});

Console.WriteLine("Starting task...");
task.Start();
task.Wait();
Console.WriteLine("Task completed!");

...or by using the static Task.Run method:

Console.WriteLine("Starting task...");
var task = Task.Run(() =>
{
    Console.WriteLine("Task code starting...");
    Thread.Sleep(2000);
    Console.WriteLine("...task code ending!");
});
task.Wait();
Console.WriteLine("Task completed!");

Note that only in the first case it is necessary to explicitly invoke Start.

# Task.WhenAll

var random = new Random();
IEnumerable<Task<int>> tasks = Enumerable.Range(1, 5).Select(n => Task.Run(() =>
{
    Console.WriteLine("I'm task " + n);
    return n;
}));

Task<int[]> task = Task.WhenAll(tasks);
int[] results = await task;

Console.WriteLine(string.Join(",", results.Select(n => n.ToString())));
// Output: 1,2,3,4,5

# Parallel.ForEach

This example uses Parallel.ForEach to calculate the sum of the numbers between 1 and 10000 by using multiple threads. To achieve thread-safety, Interlocked.Add is used to sum the numbers.

using System.Threading;

int Foo()
{
    int total = 0;
    var numbers = Enumerable.Range(1, 10000).ToList();
    Parallel.ForEach(numbers, 
        () => 0, // initial value,
        (num, state, localSum) => num + localSum,
        localSum => Interlocked.Add(ref total, localSum));
    return total; // total = 50005000
}

# Parallel.For

This example uses Parallel.For to calculate the sum of the numbers between 1 and 10000 by using multiple threads. To achieve thread-safety, Interlocked.Add is used to sum the numbers.

using System.Threading;

int Foo()
{
    int total = 0;
    Parallel.For(1, 10001, 
        () => 0, // initial value,
        (num, state, localSum) => num + localSum,
        localSum => Interlocked.Add(ref total, localSum));
    return total; // total = 50005000
}

# Task: Returning a value

Task that return a value has return type of Task< TResult > where TResult is the type of value that needs to be returned. You can query the outcome of a Task by its Result property.

Task<int> t = Task.Run(() => 
    {
        int sum = 0;

        for(int i = 0; i < 500; i++)
            sum += i;

        return sum;
    });

Console.WriteLine(t.Result); // Outuput 124750

If the Task execute asynchronously than awaiting the Task returns it's result.

public async Task DoSomeWork()
{
    WebClient client = new WebClient();
    // Because the task is awaited, result of the task is assigned to response
    string response = await client.DownloadStringTaskAsync("http://somedomain.com");
}

# Task: WaitAll and variable capturing

var tasks = Enumerable.Range(1, 5).Select(n => new Task<int>(() =>
{
    Console.WriteLine("I'm task " + n);
    return n;
})).ToArray();

foreach(var task in tasks) task.Start();
Task.WaitAll(tasks);

foreach(var task in tasks)
    Console.WriteLine(task.Result);

# Task: WaitAny

var allTasks = Enumerable.Range(1, 5).Select(n => new Task<int>(() => n)).ToArray();
var pendingTasks = allTasks.ToArray();

foreach(var task in allTasks) task.Start();

while(pendingTasks.Length > 0)
{
    var finishedTask = pendingTasks[Task.WaitAny(pendingTasks)];
    Console.WriteLine("Task {0} finished", finishedTask.Result);
    pendingTasks = pendingTasks.Except(new[] {finishedTask}).ToArray();
}

Task.WaitAll(allTasks);

Note: The final WaitAll is necessary becasue WaitAny does not cause exceptions to be observed.

# Task: handling exceptions (using Wait)

var task1 = Task.Run(() =>
{
    Console.WriteLine("Task 1 code starting...");
    throw new Exception("Oh no, exception from task 1!!");
});

var task2 = Task.Run(() =>
{
    Console.WriteLine("Task 2 code starting...");
    throw new Exception("Oh no, exception from task 2!!");
});

Console.WriteLine("Starting tasks...");
try
{
    Task.WaitAll(task1, task2);
}
catch(AggregateException ex)
{
    Console.WriteLine("Task(s) failed!");
    foreach(var inner in ex.InnerExceptions)
        Console.WriteLine(inner.Message);
}

Console.WriteLine("Task 1 status is: " + task1.Status); //Faulted
Console.WriteLine("Task 2 status is: " + task2.Status); //Faulted

# Task: handling exceptions (without using Wait)

var task1 = Task.Run(() =>
{
    Console.WriteLine("Task 1 code starting...");
    throw new Exception("Oh no, exception from task 1!!");
});

var task2 = Task.Run(() =>
{
    Console.WriteLine("Task 2 code starting...");
    throw new Exception("Oh no, exception from task 2!!");
});

var tasks = new[] {task1, task2};

Console.WriteLine("Starting tasks...");
while(tasks.All(task => !task.IsCompleted));

foreach(var task in tasks)
{
    if(task.IsFaulted)
        Console.WriteLine("Task failed: " +
            task.Exception.InnerExceptions.First().Message);
}

Console.WriteLine("Task 1 status is: " + task1.Status); //Faulted
Console.WriteLine("Task 2 status is: " + task2.Status); //Faulted

# Task: cancelling using CancellationToken

var cancellationTokenSource = new CancellationTokenSource();
var cancellationToken = cancellationTokenSource.Token;

var task = new Task((state) =>
    {
        int i = 1;
        var myCancellationToken = (CancellationToken)state;
        while(true)
        {
            Console.Write("{0} ", i++);
            Thread.Sleep(1000);
            myCancellationToken.ThrowIfCancellationRequested();
        }
    },
    cancellationToken: cancellationToken,
    state: cancellationToken);

Console.WriteLine("Counting to infinity. Press any key to cancel!");
task.Start();
Console.ReadKey();

cancellationTokenSource.Cancel();
try
{
    task.Wait();
}
catch(AggregateException ex)
{
    ex.Handle(inner => inner is OperationCanceledException);
}

Console.WriteLine($"{Environment.NewLine}You have cancelled! Task status is: {task.Status}");
//Canceled

As an alternative to ThrowIfCancellationRequested, the cancellation request can be detected with IsCancellationRequested and a OperationCanceledException can be thrown manually:

//New task delegate
int i = 1;
var myCancellationToken = (CancellationToken)state;
while(!myCancellationToken.IsCancellationRequested)
{
    Console.Write("{0} ", i++);
    Thread.Sleep(1000);
}
Console.WriteLine($"{Environment.NewLine}Ouch, I have been cancelled!!");
throw new OperationCanceledException(myCancellationToken);

Note how the cancellation token is passed to the task constructor in the cancellationToken parameter. This is needed so that the task transitions to the Canceled state, not to the Faulted state, when ThrowIfCancellationRequested is invoked. Also, for the same reason, the cancellation token is explicitly supplied in the constructor of OperationCanceledException in the second case.

# Task.WhenAny

var random = new Random();
IEnumerable<Task<int>> tasks = Enumerable.Range(1, 5).Select(n => Task.Run(async() =>
{
    Console.WriteLine("I'm task " + n);
    await Task.Delay(random.Next(10,1000));
    return n;
}));

Task<Task<int>> whenAnyTask = Task.WhenAny(tasks);
Task<int> completedTask = await whenAnyTask;
Console.WriteLine("The winner is: task " + await completedTask);

await Task.WhenAll(tasks);
Console.WriteLine("All tasks finished!");

# Flowing execution context with AsyncLocal

When you need to pass some data from the parent task to its children tasks, so it logically flows with the execution, use AsyncLocal class (opens new window):

void Main()
{
    AsyncLocal<string> user = new AsyncLocal<string>();
    user.Value = "initial user";
    
    // this does not affect other tasks - values are local relative to the branches of execution flow
    Task.Run(() => user.Value = "user from another task"); 
    
    var task1 = Task.Run(() =>
    {
        Console.WriteLine(user.Value); // outputs "initial user"
        Task.Run(() =>
        {
            // outputs "initial user" - value has flown from main method to this task without being changed
            Console.WriteLine(user.Value);
        }).Wait();

        user.Value = "user from task1";

        Task.Run(() =>
        {
            // outputs "user from task1" - value has flown from main method to task1
            // than value was changed and flown to this task.
            Console.WriteLine(user.Value);
        }).Wait();
    });
    
    task1.Wait();
    
    // ouputs "initial user" - changes do not propagate back upstream the execution flow    
    Console.WriteLine(user.Value); 
}

Note: As can be seen from the example above AsynLocal.Value has copy on read semantic, but if you flow some reference type and change its properties you will affect other tasks. Hence, best practice with AsyncLocal is to use value types or immutable types.

# Parallel.ForEach in VB.NET

For Each row As DataRow In FooDataTable.Rows
    Me.RowsToProcess.Add(row)
Next

Dim myOptions As ParallelOptions = New ParallelOptions()
myOptions.MaxDegreeOfParallelism = environment.processorcount

Parallel.ForEach(RowsToProcess, myOptions, Sub(currentRow, state)
                                               ProcessRowParallel(currentRow, state)
                                           End Sub)

# Remarks

# Purpose And Use Cases

The purpose of the Task Parallel Library is to simplify the process of writing and maintaining multithreaded and parallel code.

Some Use Cases*:

  • Keeping a UI responsive by running background work on separate task
  • Distributing workload
  • Allowing a client application to send and receive requests at the same time (rest, TCP/UDP, ect)
  • Reading and/or writing multiple files at once

*Code should be considered on a case by case basis for multithreading. For example, if a loop only has a few iterations or only does a small amount of the work, the overhead for parallelism may outweigh the benefits.

TPL with .Net 3.5

The TPL is also available for .Net 3.5 included in a NuGet package, it is called Task Parallel Library.

← Task Parallel Library (TPL) API Overviews →