It can be OS version specific (if a later OS wants to handle an older version issue that’s fine). For a specific OS, there should be an appropriate strategy to hibernate or close any stateful structures that are not temporally/network context portable.

Options of interest:

a) Includes all child processes (default)
b) Includes all other non-system processes with IPC transitivity of connections
c) wait until some idle like state
d) checkpoint on this schedule…
e) allow virtualized OS hibernation where it seems needed (i.e. system judgment about what can be safely isolated)