C# BackgroundService in ASP.NET Core: Scopes and stoppingToken

A BackgroundService is a class the generic host starts with the process and stops on shutdown. ExecuteAsync is the loop. It is not an HTTP request. _ = NotifyAsync() after return Ok() is not a worker — nobody owns the exceptions, and RequestAborted is already canceled.

New to this → stay here. Merging a PR → wrong vs right. On-call / interview → stoppingToken · scope · if an interviewer asks.

Terms used here: IHostedService = the interface; BackgroundService is the usual base class. stoppingToken = host shutdown flag passed into ExecuteAsync. IServiceScopeFactory = create a DI scope so each message gets its own DbContext. Durable queue = Service Bus / RabbitMQ — survives recycle. In-process buffer: Channel.

HTTP POST checkout
  await Writer.WriteAsync(orderId)     ← request path, then Ok()
           │
           ▼
  BackgroundService.ExecuteAsync
    stoppingToken (shutdown, not Angular)
    CreateAsyncScope() → DbContext → send

Smallest example

builder.Services.AddSingleton<OrderNotifyChannel>();
builder.Services.AddHostedService<OrderNotifyWorker>();

public sealed class OrderNotifyWorker(
    OrderNotifyChannel channel,
    IServiceScopeFactory scopes,
    ILogger<OrderNotifyWorker> logger) : BackgroundService
{
    protected override async Task ExecuteAsync(CancellationToken stoppingToken)
    {
        await foreach (var msg in channel.Reader.ReadAllAsync(stoppingToken))
        {
            try
            {
                await using var scope = scopes.CreateAsyncScope();
                var mail = scope.ServiceProvider.GetRequiredService<IOrderMailer>();
                await mail.SendAsync(msg, stoppingToken);
            }
            catch (OperationCanceledException) when (stoppingToken.IsCancellationRequested)
            {
                break;
            }
            catch (Exception ex)
            {
                logger.LogError(ex, "Notify failed for {OrderId}", msg.OrderId);
                // do not crash the loop — one poison message is not a process death
            }
        }
    }
}

Register the channel as singleton. The worker is a hosted service. Pass ids in the payload, not a request DbContext.

Wrong vs right

I would reject _ = Task.Run(() => NotifyAsync(orderId)) after checkout. That is fire-and-forget, not a hosted service.

// Wrong — dying RequestAborted, lost exceptions, stolen Kestrel workers
[HttpPost]
public async Task<ActionResult<OrderDto>> Checkout(CheckoutRequest body, CancellationToken ct)
{
    var order = await _orders.PlaceAsync(body, ct);
    _ = Task.Run(() => _mail.SendAsync(order.Id, ct)); // ct dies with the response
    return Ok(order);
}

// Right — enqueue; the worker owns send and shutdown
await _notify.Writer.WriteAsync(new OrderNotify(order.Id), ct);
return Ok(order);

AddHostedService is how the host sees the work. Task.Run is how it disappears.

stoppingToken vs RequestAborted

Comparison: which token a worker may use.

Token	Meaning	In a BackgroundService
HttpContext.RequestAborted	Angular left this request	Never — already canceled after Ok()
Action CancellationToken captured into Task.Run	Same as abort	Never
ExecuteAsync(stoppingToken)	Process is shutting down	Yes — pass it to ReadAllAsync, EF, HttpClient
CancellationToken.None	Ignore shutdown	Only if you accept work after SIGTERM

A clinic checkout that passed RequestAborted into Task.Run “so email cancels if they leave” canceled the email on every success. The tab already navigated. Use stoppingToken for host lifetime. If the business must cancel a job, give that job its own CancellationTokenSource stored by id — not the HTTP token.

One scope per message

The worker is a singleton. DbContext is scoped. Injecting AppDbContext into the worker is a captive dependency: first tenant’s context lives for the process.

// Wrong — singleton holds scoped DbContext
public sealed class NightlyReportWorker(AppDbContext db) : BackgroundService { }

// Right — new scope per tick or per message
await using var scope = scopes.CreateAsyncScope();
var db = scope.ServiceProvider.GetRequiredService<AppDbContext>();
await db.Orders.Where(o => o.Open).ExecuteUpdateAsync(..., stoppingToken);

Same rule for IOrderMailer if it takes DbContext. Resolve it inside the scope. Do not close over a request-scoped service from the controller.

Shutdown

Honor stoppingToken. Complete the channel writer so ReadAllAsync ends:

public sealed class ChannelShutdown(OrderNotifyChannel channel, IHostApplicationLifetime life)
    : IHostedService
{
    public Task StartAsync(CancellationToken cancellationToken)
    {
        life.ApplicationStopping.Register(() => channel.Writer.TryComplete());
        return Task.CompletedTask;
    }

    public Task StopAsync(CancellationToken cancellationToken) => Task.CompletedTask;
}

If the process dies after Ok and before send, the email is gone. Cards, claims, and compliance mail need an outbox or a durable bus — not a bigger in-memory buffer.

PeriodicTimer vs Delay

A nightly job should not while (!stoppingToken.IsCancellationRequested) { await Task.Delay(TimeSpan.FromHours(24), stoppingToken); } if ticks can overlap. Prefer PeriodicTimer:

using var timer = new PeriodicTimer(TimeSpan.FromHours(1));
while (await timer.WaitForNextTickAsync(stoppingToken))
{
    await using var scope = scopes.CreateAsyncScope();
    await scope.ServiceProvider.GetRequiredService<IFeeImport>().RunAsync(stoppingToken);
}

Thread.Sleep in a worker parks a pool thread for the whole interval. That is the same class of bug as Sleep vs Delay.

Common mistakes

• Task.Run after Ok() and calling it a “background job”
• Injecting DbContext or IHttpContextAccessor into the worker
• Passing RequestAborted into the loop
• Crashing ExecuteAsync on the first poison message
• Forgetting Writer.Complete() so shutdown hangs
• Treating a Channel as durable across App Service recycle

What this is not

The in-process queue: Channel. Per-request CPU offload: Task.Run vs await. Request tokens: CancellationToken. Captive DI: lifetimes. Topic map: async & threading hub.

If an interviewer asks

What a hosted service is; why RequestAborted is wrong after Ok(); how you get a DbContext in a singleton worker; Channel vs bus.

Strong answer: host-owned loop, stoppingToken, scope per message, not Task.Run.

If checkout “sends email” with Task.Run and nights drop notifies on recycle, contact me. Bring the worker class.