Tom Lane wrote:
> Magnus Hagander <magnus@hagander.net> writes:
>> Tom Lane wrote:
>>> Another problem is that postmaster children that do
>>> PGSharedMemoryDetach will still have valid inherited handles for
>>> the shmem segment --- does that factor into the behavior? It looks
>>> to me like the CloseHandle ought to be in PGSharedMemoryDetach.
>
>> Not as long as the processes die. If they die, their handles go with
>> them, and once the reference count goes to zero, the object goes away.
>
> But the syslogger process (and maybe others) is *not* supposed to die.
Right. But are you saying we actually want to start up a new backend in
a directory where we already have a running syslogger (and maybe others)
processes, just no postmaster? I'd assume we might run into such simple
things as "sharing violations" on the logfile - if nothing inside the db
itself..
//Magnus
