On Fri, 01 Aug 2008 10:33:59 -0400 Tom Lane <tgl@xxxxxxxxxxxxx> wrote: > Ivan Sergio Borgonovo <mail@xxxxxxxxxxxxxxx> writes: > > Well I reached 3Gb of work_mem and still I got: > > > "Seq Scan on catalog_categoryitem (cost=31747.84..4019284477.13 > > rows=475532 width=6)" > > " Filter: (NOT (subplan))" > > " SubPlan" > > " -> Materialize (cost=31747.84..38509.51 rows=676167 > > width=8)" " -> Seq Scan on catalog_items > > (cost=0.00..31071.67 rows=676167 width=8)" > > Huh. The only way I can see for that to happen is if the datatypes > involved aren't hashable. What's the datatypes of the two columns > being compared, anyway? I changed both columns to bigint. I added 2 indexes on the ItemID column of both tables and increased work_mem to 3Gb [sic]. The query got executed in ~1300ms... but explain gave the same output as the one above. The problem is solved... but curious mind want to know. -- Ivan Sergio Borgonovo http://www.webthatworks.it
