ASP .Net Core Queued background tasks parallel processing

Question

I have an ASP .NET core Web API which uses Queued background tasks like described here.

I've used the code sample provided and added the IBackgroundTaskQueue, BackgroundTaskQueue and QueuedHostedService exactly as described in the article.

In my Startup.cs, I'm registering only one QueuedHostedService instance as follows: services.AddHostedService<QueuedHostedService>();

Tasks coming from the WebApi's controller are enqueued and then dequeued and executed one by one by the QueuedHostedService.

I'll would like to allow more than one background processing thread that will dequeue and execute the incoming Tasks. The most straight forward solution i can come up with is to register more than one instance of the QueuedHostedService in my Startup.cs. i.e, something like this:

 int maxNumOfParallelOperations;
 var isValid = int.TryParse(Configuration["App:MaxNumOfParallelOperations"], out maxNumOfParallelOperations);

 maxNumOfParallelOperations = isValid && maxNumOfParallelOperations > 0 ? maxNumOfParallelOperations : 2;

 for (int index = 0; index < maxNumOfParallelOperations; index++) 
 {
    services.AddHostedService<QueuedHostedService>();
 }

I've also noticed that thanks to the singal Semaphore in BackgroundTaskQueue, the QueuedHostedService instances are not really working all the time, but only awaken when a new Task is available in the queue.

This solution seems to works just fine in my tests.

But, In this particular use case - is it really a valid, recommended solution for parallel processing?

Using the QueuedHostedService is the best solution. Unfortunally you don't have total control over thread pool. Consider that the degree of parallelism is not given by the number of hosted services that you use but by the number of thread of the thread pool. If your using 10 threads you can perform 10 operations in parallel. This threads are shared with the webhost and I think that asp.net core gives priority to the web requestes over the background tasks. — ddfra
– ddfra, Commented Jul 12, 2020 at 22:52
OK, Thanks, I got it. But is registering multiple instances of the same background task, like described here, for parallel processing, is a good practice? My direction is to register up to 5 of them, in order to allow basic level of parallelism in my service. — LioH
– LioH, Commented Jul 12, 2020 at 23:19

Denys Vega · Accepted Answer · 2020-07-13 00:23:04Z

You can use an IHostedService with a number of threads to consume the IBackgroundTaskQueue.

Here is a basic implementation. I assume you're using the same IBackgroundTaskQueue and BackgroundTaskQueue described here.

public class QueuedHostedService : IHostedService
{
    private readonly ILogger _logger;

    private readonly Task[] _executors;
    private readonly int _executorsCount = 2; //--default value: 2
    private CancellationTokenSource _tokenSource;
    public IBackgroundTaskQueue TaskQueue { get; }

    public QueuedHostedService(IBackgroundTaskQueue taskQueue,
        ILoggerFactory loggerFactory,
        IConfiguration configuration)
    {
        TaskQueue = taskQueue;
        _logger = loggerFactory.CreateLogger<QueuedHostedService>();

        if (ushort.TryParse(configuration["App:MaxNumOfParallelOperations"], out var ct))
        {
            _executorsCount = ct;
        }
        _executors = new Task[_executorsCount];
    }

    public Task StartAsync(CancellationToken cancellationToken)
    {
        _logger.LogInformation("Queued Hosted Service is starting.");

        _tokenSource = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken);

        for (var i = 0; i < _executorsCount; i++)
        {
            var executorTask = new Task(
                async () =>
                {
                    while (!cancellationToken.IsCancellationRequested)
                    {
#if DEBUG
                    _logger.LogInformation("Waiting background task...");
#endif
                    var workItem = await TaskQueue.DequeueAsync(cancellationToken);

                        try
                        {
#if DEBUG
                        _logger.LogInformation("Got background task, executing...");
#endif
                        await workItem(cancellationToken);
                        }
                        catch (Exception ex)
                        {
                            _logger.LogError(ex,
                                "Error occurred executing {WorkItem}.", nameof(workItem)
                            );
                        }
                    }
                }, _tokenSource.Token);

            _executors[i] = executorTask;
            executorTask.Start();
        }

        return Task.CompletedTask;
    }

    public Task StopAsync(CancellationToken cancellationToken)
    {
        _logger.LogInformation("Queued Hosted Service is stopping.");
        _tokenSource.Cancel(); // send the cancellation signal

        if (_executors != null)
        {
            // wait for _executors completion
            Task.WaitAll(_executors, cancellationToken);
        }

        return Task.CompletedTask;
    }
}

You need to register the services in ConfigureServices on Startup class.

...
services.AddSingleton<IBackgroundTaskQueue, BackgroundTaskQueue>();
services.AddHostedService<QueuedHostedService>();
...

Aditionally, you can set the number of threads in configuration (appsettings.json)

...
"App": {
    "MaxNumOfParallelOperations": 4
}
...

Thanks for the solution . Its working great but the CPU usage is always high even there are no jobs in the queues . I am running default _executorsCount as 5 and CPU usage for .net counter is showing around 60% .
When you do await workItem(cancellationToken); you should do await workItem(_tokenSource.Token); so when the service is stopped on StopAsync, the token that you passed on the OnStart triggers the cancel Anyway.. great solution! thanks!

Collectives™ on Stack Overflow

ASP .Net Core Queued background tasks parallel processing

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related