On Tue, 2005-08-02 at 04:24, Havasvölgyi Ottó wrote:
> Tom,
>
> Thanks for the suggestion. I have just applied both switch , -f (I have
> applied this in the previous case too) and -n, but it becomes slow again. At
> the beginning it reads about 300 KB a second, and when it has read 1.5 MB,
> it reads only about 10 KB a second, it slows down gradually. Maybe others
> should also try this scenario. Can I help anything?
I be you've got an issue where a seq scan on an fk field or something
works fine for the first few thousand rows. At some point, pgsql should
switch to an index scan, but it just doesn't know it.
Try wrapping every 10,000 or so inserts with
begin;
<insert 10,000 rows>
commit;
analyze;
begin;
rinse, wash repeat.
You probably won't need an analyze after the first one though.