Re: Subscription tests fail under CLOBBER_CACHE_ALWAYS

I wrote:
> BTW, I wonder whether it wouldn't be a good idea for the
> postmaster to log something along the lines of "stopping
> because restart_after_crash is off". The present behavior
> can be quite mysterious otherwise (it certainly confused me).

Concretely, I suggest the attached.

While checking the other ExitPostmaster calls to see if any of
them lacked suitable log messages, I noticed that there's one
after a call to AuxiliaryProcessMain, which is marked
pg_attribute_noreturn(). So that's dead code, and if it
weren't dead it'd be wrong, because we shouldn't use
ExitPostmaster to exit a child process.

regards, tom lane