I haven't yet explored parallelization of shutdowns. This is all presently sequential and operates on the main node application thread before network start and after network stop. Isolating the store from the network greatly simplifies things, including mock store injection.
(edited)