Re: Providing catalog view to pg_hba.conf file - Patch submission

2015-02-28 3:12 GMT+01:00 Stephen Frost <sfrost(at)snowman(dot)net>:

> * Josh Berkus (josh(at)agliodbs(dot)com) wrote:
> > On 02/27/2015 04:41 PM, Stephen Frost wrote:
> > >> we can do copy of pg_hba.conf somewhere when postmaster starts or
> when it
> > >> is reloaded.
> > >
> > > Please see my reply to Tom. There's no trivial way to reach into the
> > > postmaster from a backend- but we do get a copy of whatever the
> > > postmaster had when we forked, and the postmaster only reloads
> > > pg_hba.conf on a sighup and that sighup is passed down to the children,
> > > so we simply need to also reload the pg_hba.conf in the children when
> > > they get a sighup.
> > >
> > > That's how postgresql.conf is handled, which is what pg_settings is
> > > based off of, and I believe is the behavior folks are really looking
> > > for.
> >
> > I thought the patch in question just implemented reading the file from
> > disk, and nothing else?
> >
> > Speaking for my uses, I would rather have just that for 9.5 than wait
> > for something more sophisticated in 9.6.
>
> From my perspective, at least, the differences we're talking about are
> not enough to raise this to a 9.5-vs-9.6 issue. I can see the use cases
> for both (which is exactly why I suggested providing both). Having one
> would be better than nothing, but I foretell lots of subsequent
> complaints along the lines of "everything looks right according to
> pg_hba_config, but I'm getting this error!!" Now, perhaps that's the
> right approach to go for 9.5 since it'd more-or-less force our hand to
> deal with it in 9.6 properly, but, personally, I'd be happier if we
> moved forward with providing both because everyone agrees that it makes
> sense rather than waiting to see if user complaints force our hand.
>

+1

Probably we can implement simple load pg_hba.conf and tab transformation
early. There is a agreement and not any problem.

But if we start to implement some view, then it should be fully functional
without potential issues.

Regards

Pavel

>
> Thanks!
>
> Stephen
>