Should server shutdown after receiving "lifespan.shutdown.failed"? about uvicorn HOT 6 CLOSED

the issue and patch seems legit to me, can you submit a PR @peterschutt ?

from uvicorn.

Thanks @euri10 - I'll polish up the patch and submit a PR.

from uvicorn.

I don't think this is a bug. My interpretation from the spec is that once the "lifespan.startup.complete" is sent, your application is the only task running.

Also, this is the same behavior as the other ASGI servers.

I think what you want is something like this:

import anyio
from starlette.applications import Starlette
from starlette.requests import Request
from starlette.responses import JSONResponse
from starlette.routing import Route

import uvicorn


async def infinite_task():
    while True:
        await anyio.sleep(1)
        print("I am alive!")


async def health(request: Request):
    return JSONResponse({"status": "ok"})


app = Starlette(routes=[Route("/health", health)])


async def main():
    async with anyio.create_task_group() as tg:
        tg.start_soon(infinite_task)

        config = uvicorn.Config(app=app, host="0.0.0.0", port=8002)
        server = uvicorn.Server(config=config)
        await server.serve()


if __name__ == "__main__":
    anyio.run(main, backend_options={"use_uvloop": True})

from uvicorn.

Thanks @Kludex - that pattern will work just fine!

once the "lifespan.startup.complete" is sent, your application is the only task running.

Do you feel there is a disconnect between the spec, and how some frameworks advertise lifespan? For example, from starlette's docs (ref):

Consider using anyio.create_task_group() for managing asynchronous tasks.

And from litestar (ref):

This can be useful when dealing with long running tasks

And lastly, from a framework perspective, can we do any better than returning the "lifespan.shutdown.failed" message, in the case where an exception is caught within the lifespan?

from uvicorn.

Do you feel there is a disconnect between the spec, and how some frameworks advertise lifespan? For example, from starlette's docs (ref):

I'm not sure.

And lastly, from a framework perspective, can we do any better than returning the "lifespan.shutdown.failed" message, in the case where an exception is caught within the lifespan?

I don't know.

from uvicorn.

@euri10 can we close this please?

from uvicorn.