fix: support persistent lifespans in asgi

[dom96]

Ensures that mutable lifespans are persisted across requests. Adds fastapi and asgi-specific tests.

[ask-bonk]

A cancelled first request propagates cancellation into this shared task, leaving lifespan permanently cancelled and causing every later request in the isolate to be cancelled. Shield the shared startup task from an individual request cancellation.

Suggested change

	async def lifespan_state():
	nonlocal lifespan
	if lifespan is None:
	# Assign before awaiting so concurrent first requests share startup.
	lifespan = create_task(start_application(app))
	_, state = await lifespan
	return state
	async def lifespan_state():
	from asyncio import shield
	nonlocal lifespan
	if lifespan is None:
	# Assign before awaiting so concurrent first requests share startup.
	lifespan = create_task(start_application(app))
	_, state = await shield(lifespan)
	return state

[ask-bonk]

I'm Bonk, and I've done a quick review of your PR.

Makes ASGI entrypoint lifespan state persist across requests.

1. High: Posted 1 actionable inline suggestion.

github run

[hoodmane]

Generally looks reasonable, though I'd set _start_future on the AsgiWorkerEntrypoint instance rather than the class and initialize _start_future to None in AsgiWorkerEntrypoint.__init__(). Could adjust it in a followup if you like though.

[dom96]

So we do actually want it set on the class because the instance is re-created on every worker invocation. The lifespan is supposed to stay alive for the duration of the lifetime of the isolate. I added a comment to explain this.

[hoodmane]

the instance is re-created on every worker invocation

That is really weird, can we migrate away from that?

[hoodmane]

In JavaScript is the WorkerEntrypoint instantiated on every request?