Sub application startup event never triggered

[jonathanunderwood]

Describe the bug

Startup event handlers for sub-applications never trigger.

To Reproduce

from fastapi import FastAPI


sub_app = FastAPI(openapi_prefix="/subapp")


@sub_app.on_event("startup")
async def sub_app_startup():
    print("++++ Sub-app startup event")  # Never fires


app = FastAPI()


@app.on_event("startup")
async def app_startup():
    print("++++ App startup event")


app.mount("/subapp", sub_app)


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, port=8000, host="0.0.0.0")  # nosec

Running the above shows:

$ python subapp_events.py
INFO:     Started server process [38462]
INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
INFO:     Waiting for application startup.
++++ App startup event
INFO:     Application startup complete.

Notice that sub_app_startup() is never run.

Expected behavior

I would expect the sub-application startup event handler to also run (although I am not sure what ordering I would expect).

Environment

• OS: macOS
• FastAPI Version: 0.45.0
• Python version: 3.7.5

Additional context

N/A

[jonathanunderwood]

Digging in to this a bit more, I can see that FastAPI merely exposes the Starlette lifecycle events and doesn't do anything special about them.

In addition, this issue has been raised previously against Starlette.

Since it's deemed out-of-scope by Starlette developers, I think the only action here would be to document this in the FastAPI docs i.e. call out explicitly that sub-application event handlers are ignored.

[dmontagu]

I also think it would be good to document this.

PR welcome!

[tiangolo]

Thanks for the investigation and for reporting back @jonathanunderwood , yep, it makes sense to have it documented.

[ClericPy]

I found this issue this days, so what should I do to trigger subapp's start_up callbacks?
It seems that fastapi.APIRouter's on_startup param has a similar feature.

[tiangolo]

@ClericPy it would have to be implemented in Starlette: encode/starlette#649

[tiangolo]

This was documented in #1554 by @amacfie 🚀

[github-actions]

Assuming the original issue was solved, it will be automatically closed now. But feel free to add more comments or create new issues.

[jonathanunderwood]

As an update, looks like there's activity upstream in Starlette to bring this capability in: encode/starlette#1988

[devamin]

some updates about this feature, sometimes it is required specially when dealing with non-fastapi ASGI apps like socketio one https://github.com/miguelgrinberg/python-socketio/blob/main/src/socketio/asgi.py

[OmegAshEnr01n]

Yes, But there is an easy work around for this for now.
Make a function that is responsible for the start up event for your sub app and call it in the main app startup function.

# Main app
@app.on_event("startup")
def spawn_children():
    model_sub_app.startup_function()

# Sub app
def startup_function():
    # Load model
    pass

I find this super useful for certain cases. For example if I am making an app which has one AI model loaded on each router I wouldnt be able to add features to them like socketio. So for now it appears that having a fastapi instance for such use cases and mounting it is ideal. Paired with the above mentioned startup function hack and its doable...

[dhilst]

What I did was to create a function to mount my application and register my event handlers in the parent app

def mount_app(parent_app: FastAPI, path: str) -> None:
    parent_app.on_event("startup")(app_startup)
    parent_app.on_event("shutdown")(app_shutdown)
    parent_app.mount(path, app)

[unights]

I did something similar. Provide a reference for others.

from contextlib import AsyncExitStack, asynccontextmanager

from fastapi import FastAPI
from starlette.routing import Mount


@asynccontextmanager
async def lifespan(app: FastAPI):
    async with AsyncExitStack() as stack:
        for route in app.routes:
            if isinstance(route, Mount) and isinstance(route.app, FastAPI):
                await stack.enter_async_context(
                    route.app.router.lifespan_context(route.app),  # noqa
                )
        yield