Re: Proposal for Allow postgresql.conf values to be changed via SQL

On Wed, Nov 7, 2012 at 6:19 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Magnus Hagander <magnus(at)hagander(dot)net> writes:
>> On Wed, Nov 7, 2012 at 5:53 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> I'm not sure that the above approach works anyway --- for instance, the
>>> "current setting" might be a SET LOCAL result, in which case you still
>>> don't know anything about what the appropriate thing to put into the
>>> file is. I think there are probably also race conditions with cases
>>> where somebody else just changed some other setting but your session
>>> hasn't absorbed it yet.
>
>> Well, you don't have to look at pg_settings specifically - since this
>> is inside the backend. You can look at the underlying structures. We
>> stack them up so we can RESET them, right? So we could just peek up in
>> that stack and find the data there.
>
> You could dig it out of the stack if it's there, but that doesn't fix
> the race-condition aspect. Now a race is inevitable if two sessions try
> to set the *same* variable, but I think people will be unhappy if a SET
> on one variable makes a recent SET on some other variable disappear.

I think if we require an exclusive lock on a single global lock for
"set permanent", people are quite ok with that, really. Changing
permanent settings concurrently doesn't seem like a veyr likely
scenario.

> The one-value-per-file solution neatly bypasses all these problems,
> which is why this topic got put on the back burner originally until
> we had the include-directory functionality. I don't see why we are
> revisiting the bugs in an approach that was already rejected.

Yeah, agreed - that certainly takes most of it away. And there is
nothing preventing somebody from having both that and another
directory-include somewhere if they'd like to...

--
Magnus Hagander
Me: http://www.hagander.net/
Work: http://www.redpill-linpro.com/