Re: exposing pg_controldata and pg_config as functions

On Wed, Dec 30, 2015 at 9:08 AM, Joe Conway <mail(at)joeconway(dot)com> wrote:
> 1) Change NextXID output format from "%u/%u" to "%u:%u"
> (see recent hackers thread)

! printf(_("Latest checkpoint's NextXID: %u/%u\n"),
ControlFile.checkPointCopy.nextXidEpoch,
ControlFile.checkPointCopy.nextXid);
printf(_("Latest checkpoint's NextOID: %u\n"),
--- 646,652 ----
ControlFile.checkPointCopy.ThisTimeLineID);
printf(_("Latest checkpoint's full_page_writes: %s\n"),
ControlFile.checkPointCopy.fullPageWrites ? _("on") : _("off"));
! printf(_("Latest checkpoint's NextXID: %u:%u\n"),
This should be definitely a separate patch.

> 2) Refactor bin/pg_controldata (there should be no visible change to
> pg_controldata output)
> 3) Adds new functions, more or less in line with previous discussions:
> * pg_checkpoint_state()
> * pg_controldata_state()
> * pg_recovery_state()
> * pg_init_state()

Taking the opposite direction of Josh upthread, why is this split
actually necessary? Isn't the idea to provide a SQL interface of what
pg_controldata shows? If this split proves to be useful, shouldn't we
do it as well for pg_controldata?

> ===============
> Missing (TODO once agreement on the above is reached):
> ---------------
> a) documentation

This would be good to have.

> b) catversion bump

That's committer work.

> c) regression tests

Hm, what would be the value of those tests? I think we could live
without for simple functions like that honestly.

I think that those functions should be superuser-only. They provide
information about the system globally.
--
Michael