You were absolutely right Tom.
Rising work_mem did the trick!
Many thanks :-)
What's the best value for work_mem ?
Gaëtan
Le 7 févr. 2010 à 07:38, Tom Lane a écrit :
> =?iso-8859-1?Q?Ga=EBtan_Allart?= <gallart@free.fr> writes:
>> I'm experiencing an interesting issue with PostgreSQL 8.4.2-r2.
>> I was running a pretty big database on another server (8.3) pretty fine.
>> When I moved it to this new server, behaviour changed.
>
> Did you ANALYZE the database after loading it? Are you sure you
> have all the parameters set similarly to what you used in the 8.3
> installation?
>
> (The described behavior sounds suspiciously like work_mem is not
> large enough to allow a hashagg to be chosen.)
>
> regards, tom lane