Re: Weaker shmem interlock w/o postmaster.pid

On Wed, Sep 11, 2013 at 07:57:22PM -0700, David Johnston wrote:
> Noah Misch-2 wrote
> > Making PGSharedMemoryCreate() pickier in all branches will greatly
> > diminish
> > the marginal value of preserving postmaster.pid, so I'm fine with dropping
> > the
> > postmaster.pid side of the proposal.
> 
> Its probably still worth a fresh look at the immediate shutdown process to
> see whether the current location where postmaster.pid is removed is
> acceptable.  It may not be necessary to leave it in place always but:
> 
> 1) if there is a section of shared memory that can only be reached/found if
> one knows the pid, and

Similar: one needs a sysv shared memory key to find the segment, and
postmaster.pid records that key.  The chosen key is almost always the same
from run to run, so a new postmaster typically does find the old segment even
without postmaster.pid.

> 2) postmaster.pid is removed before that area is "secured from future
> clobbering"

Clobbering shared memory is not the actual threat here.  We use the shared
memory segment as a witness to the fact that PostgreSQL processes are active
in a data directory.  If we let children of different postmasters operate in
the same directory simultaneously, they can corrupt data.

> then there may be a risk that can still be mitigated by moving its removal
> without having to go to the extreme.  

-- 
Noah Misch
EnterpriseDB                                 http://www.enterprisedb.com