Design pattern for global async reactive

I'm looking for a design pattern that could be used to have a http request called every N seconds and use the response to set a global reactive.Value. It's a potentially expensive request/calculation, but the result is independent of the user requesting it. So i'd ideally like to perform the request once, (async to avoid blocking the reactivity) and then "distribute" the new value(s) to all users??

I've tried it with async keyword reactives as in this gist but the reactivity of inputs (not shown in the gist) and outputs e.g. the timer_out() output are "frozen" while waiting on the long-running api request.

I've had a look at multiprocessing to use the pattern described here, which i've used in shiny-R before, but that still seems to block from within the global async reactive.

I'm a bit rusty with python so I might well be missing something important.