How to use Background Tasks?

[Risiamu]

Sorry, I’m just an AI engineer who has recently started trying to set up a server, so maybe I’ve asked a silly question. Please bear with me.

I have an API that initiates a PyTorch training process based on the data sent from the frontend. Later, I realized that this locks up the backend, making it impossible for other APIs to be accessed.

I looked into the official documentation, and it seems that using BackgroundTasks might be the correct solution? Like this?

Is this the correct way to tackle long cpu-lock tasks?

@post("/upload")
async def upload_file(
        data: UploadFile = Body(media_type=RequestEncodingType.MULTI_PART, default=None)
) -> Response[dict[str, str]]:
 
        return Response({
            "status": "success",
        }, background=BackgroundTask(train, max_epochs=10000, data_path=cleaned_file_path))

I’d also like to ask for another suggestion. On my server, I first preprocess the data and then run the training. From this perspective, the preprocessing part should also be implemented as a Background Task.

I can split it into two APIs, where the frontend controls when to preprocess and when to train—I understand that concept.

Alternatively, I could use a single API with two background tasks, passing them as a list. (If my understanding of Background is correct) That seems feasible to me as well.

My question is: Can I directly call one API from another API? Like this:

@post("/a")
async def a()

@post("/b")
async def b()
    # call a here

Is above code a silly choice?

[provinzkraut]

Is this the correct way to tackle long cpu-lock tasks?

No, it is not. I recommend reading this article from our docs: https://docs.litestar.dev/latest/topics/sync-vs-async

The background tasks still run in the same process as your application, so for CPU bound tasks, you're not gaining anything.

For long running / compute heavy tasks, you want to offload the process to a task queue, like celery.

[Risiamu]

Is this the correct way to tackle long cpu-lock tasks?

No, it is not. I recommend reading this article from our docs: https://docs.litestar.dev/latest/topics/sync-vs-async

The background tasks still run in the same process as your application, so for CPU bound tasks, you're not gaining anything.

For long running / compute heavy tasks, you want to offload the process to a task queue, like celery.

Thank you for the guidance. I will checkout them ASAP.

[Risiamu]

Is this the correct way to tackle long cpu-lock tasks?

No, it is not. I recommend reading this article from our docs: https://docs.litestar.dev/latest/topics/sync-vs-async

The background tasks still run in the same process as your application, so for CPU bound tasks, you're not gaining anything.

For long running / compute heavy tasks, you want to offload the process to a task queue, like celery.

I notice in https://github.com/litestar-org/litestar-fullstack It uses something called SAQ, https://saq-py.readthedocs.io/en/latest/
Will this also be a viable option?

[provinzkraut]

While I use saq in a few projects, I wouldn't recommend it for this case. It's async based, which doesn't really help you, for the same reasons background tasks won't. Celery is fundamentally synchronous, and scaled up by using multiple workers. Of course you could do the same with saq, adding multiple workers, but you won't benefit from that. One task per process is the best model if one task can saturate a CPU core. And for that, celery is the best option.

[Risiamu]

Thank you for such detailed answer. I really appreciate it.