On Wed, Jan 22, 2014 at 09:32:09AM -0500, Robert Haas wrote:
> On Tue, Jan 21, 2014 at 2:58 PM, Noah Misch <noah@leadboat.com> wrote:
> >> What do people prefer?
> >
> > I recommend performing cleanup on the control segment named in PGShmemHeader
> > just before shmdt() in PGSharedMemoryCreate(). No new ERROR or WARNING sites
> > are necessary. Have dsm_postmaster_startup() continue to perform a cleanup on
> > the control segment named in the state file.
>
> I think I'm on board with the first two sentences of this, but after
> Fujii Masao's email yesterday, I can't help thinking that what you
> propose the third sentence is a bad idea. He cloned a master to
> create a standby server on the same machine, and the standby startup
> ate the master's dynamic shared memory. We could teach pg_basebackup
> not to copy the state file, but that wouldn't help people who take
> base backups using the file system copy method, which is a lot of
> people.
I agree we should not rely on folks learning to omit the state file from base
backups. Abandoning the state file is one way to resolve that, and the
reasons I outlined for preferring to keep it were not overriding concerns. We
could instead store a postmaster PID in dsm_control_header and only clean if
that PID is dead. We could make DSM startup aware of whether we're using a
backup label, but that would be awkward thanks to StartupXLOG() happening a
good bit later. Yeah, abandoning the state file is looking attractive.
--
Noah Misch
EnterpriseDB http://www.enterprisedb.com