On Sun, May 10, 2009 at 2:03 AM, David Fetter <david@fetter.org> wrote:
> On Sat, May 09, 2009 at 01:28:03PM -0400, Tom Lane wrote:
>> Merlin Moncure <mmoncure@gmail.com> writes:
>> > On Fri, May 8, 2009 at 5:40 PM, Alvaro Herrera
>> > <alvherre@commandprompt.com> wrote:
>> >> Yeah, we went over this on the spanish list, turned out that I
>> >> couldn't remember about syncscan :-)
>>
>> > I like the new behavior. It really encourages proper use of order
>> > by, because the natural ordering results are effectively
>> > randomized. A class of subtle bugs has been made obvious. :)
>>
>> Not really, because the syncscan behavior only kicks in when your
>> table gets large ... you'll never see it during devel testing on toy
>> tables ...
>
> Good point. It's important not to test only on toy-sized tables for
> lots and lots of good reasons, scale-dependence of sync scans being a
> small one.
Last job I was at I was the lone pgsql guy who worked with three
Oracle DBAs, and quite a few of them were caught off guard by this
type of behaviour (it was with hash_agg and reporting queries with
group by).