Thank you Marti,
I will go with the ``reduced number of matched rows'' and naturally be waiting for postgres 9.1 expectantly.
Kind regards,
Kim
On 2011-02-15 22:13, Marti Raudsepp wrote:
> On Tue, Feb 15, 2011 at 21:33, Kim A. Brandt<kimabrandt@gmx.de> wrote:
>> removing the ORDER BY worked. But I am afraid to ask this. How can I order
>> by partition? It seams that the planner has picked a random(!?) order of
>> partition to select from. The returned records, from the selected partition,
>> are correctly sorted bythe index though.
>
> If a single query accesses more than one partition, PostgreSQL
> currently cannot read the values in index-sorted order. Hence with
> ORDER BY and LIMIT, PostgreSQL cannot return *any* results before it
> has read all matching rows and then sorted them. Adding a LIMIT
> doesn't help much. Your only bet is to reduce the number of matched
> rows, or make sure that you only access a single partition.
>
> Increasing work_mem may speed up the sort step if you're hitting the
> disk (EXPLAIN ANALYZE VERBOSE will tell you whether that's the case).
>
> This will change in PostgreSQL 9.1 which has a new Merge Append plan node.
>
> Regards,
> Marti