How long does the `httpx.AsyncClient` can "live" without closing the event loop?

Question

I have an endpoint in Django that orchestrates a bunch of API calls to other underlying services. I intend to use a long-lived instance of httpx.AsyncClient - which will be ultimately responsible for the API calls.

from functools import partial


class Api:  # for historical reasons, this class has both sync and async methods
   ...


AsyncApi = partial(Api, async_client=httpx.AsyncClient(timeout=30))


class MyViewSet(viewsets.GenericViewSet):
   @action(...)
   def merge(self, request: Request) -> Response:
      ...
      result = async_to_sync(merge_task)(..., api=AsyncApi(token=request.auth))
      ...

I'm doing some tests in a dummy server and I can see that sometimes merge_task captures RuntimeError('Event loop is closed') when calling api (or, ultimately, the long-lived instance of httpx.AsyncClient) so I'd bet that the loop for httpx.AsyncClient is closed after some time and there's no issue with asgiref's async_to_sync.

Am I correct? If so, the only solution is to instantiate httpx.AsyncClient on every request?

I tried do call merge_task using the endpoint but it captured RuntimeError('Event loop is closed'). The expectation is to not raise any error related to async programming