On Sep 4, 2008, at 6:29 AM, Andrew Sullivan wrote:
> On Thu, Sep 04, 2008 at 01:26:44AM +0300, Hannu Krosing wrote:
>
>> So Andrews opinion was that Mb (meaning Mbit) is different from MB
>> (for
>> megabyte) and that if someone thinks that we define shared buffers in
>> megabits can get confused and order wrong kind of network card ?
>
> I know it's fun to point and laugh instead of giving an argument, but
> the above is not what I said. What I said is that there is a
> technical difference between at least some of these units, and one
> that is relevant in some contexts where we have good reason to believe
> Postgres is used. So it seems to me that there is at least a _prima
> facie_ reason in favour of making case-based decisions. Your argument
> against that appears to be, "Well, people can be sloppy."
Settings in postgresql.conf are currently case-insensitive. Except
for the units.
> Alvaro's suggestion seems to me to be a better one. It is customary,
> in servers with large complicated configuration systems, for the
> server to come with a tool that validates the configuration file
> before you try to load it. Postfix does this; apache does it; so does
> BIND. Heck, even NSD (which is way less configurable than BIND) does
> this. Offering such a tool provides considerable more benefit than
> the questionable one of allowing people to type whatever they want
> into the configuration file and suppose that the server will by magic
> know what they meant.
How would such a tool cope with, for example, shared_buffers
being set to one eighth the size the DBA intended, due to their
use of Mb rather than MB? Both of which are perfectly valid
units to use to set shared buffers, even though we only support
one right now. If the answer to that is something along the lines
of we don't support megaabits for shared_buffers, and never will because
nobody in their right mind would ever intend to use megabits
to set their shared buffer size... that's a useful datapoint when
it comes to designing for usability.
Cheers, Steve