On Thu, Apr 7, 2011 at 8:56 AM, French, Martin <frenchm@cromwell.co.uk> wrote:
> Thanks for the info Tom.
>
> The table has been analyzed (somewhat repeatedly...), with the stats
> target set at various limits.
>
> At the moment default_statistics_target = 50.
>
> I've had work_mem as low as 1MB and as high as 128MB, with little to no
> avail.
>
> Setting enable_hashagg = off, does allow the query to complete, but
> mashes the next query in the batch with it having to be killed at 8
> hours. Setting it off for this query alone causes the next query to do
> exactly the same.
>
> The thing that perplexes me is; that on RHEL 5.0 with all the same
> settings and data, the query works fine... so it's confusing as to what
> can have changed so much to cause this. I know that we had to move to
> RHEL 5.4 because of PERC Raid drivers, but I wouldn't have thought that
> would've made too much of a difference.
So what's the difference between the query plans on the two machines?
Any difference in non-default settings in postgresql.conf between the
two machines?