The logic for serving prometheus metrics was somewhat broken: It would attempt to check that the cancellation context didn't indicate that we should exit, but it would only do that before starting to serve.
Instead, first serve, and if serve returns a fault, check whether we're meant to exit. Oops.
The logic for serving prometheus metrics was somewhat broken: It would attempt to check that the cancellation context didn't indicate that we should exit, but it would only do that before starting to serve.
Instead, first serve, and if serve returns a fault, check whether we're meant to exit. Oops.