Tom Lane <tgl@sss.pgh.pa.us> writes:
> Palle Girgensohn <girgen@partitur.se> writes:
> > New facts: The problem "moved" when raising the -B from default (64)
> > to 1000 (-N 100 now). Now, here's what happens:
>
> Interesting.
I am also using -o -S 2048 now, but I doubt it helps...
> > query: CREATE VIEW wtabmaria AS SELECT
p.personid,p.foretag,p.fnamn,p.enamn,p.titel,p.telefon,p.mobil,p.email,p.tidpunktFROM personer p WHERE true AND low
> > er(p.enamn)~~lower('Branner%')
>
> Hm, you had not mentioned before that wtabmaria is a view. That may be
> the critical factor. However, I'm still not having any luck duplicating
> the failure.
Oh, gee. sorry! I slipped with the mouse when cut'n'pasting. The first
statement should be a create view. The view is dropped and created
"on-the-fly" with different where-clauses every time, depending on the
user's search arguments. As I might have said before, I am not the
programmer... I think I would have the view created once, and done
"select from view where..." Maybe that would help, but there is still
a bug in postgres, I guess?
> > It looks very much like an out-of-memory error,
I must take that back, since the select from view only holds two
records... My mistake, again...
> I don't think I believe that; out-of-memory problems should be reported
> as such. Moreover, I see nothing in this query that would require
> touching more than one disk buffer at a time. Now I really want to see
> the coredump backtrace...
>
> > There are about 5500 records in this table.
>
> Is that total in "personer", or total shown by the view?
in "personer".
Also, I have suddenly become a "very lousy bug reporter", sorry. I forgot the indices:
CREATE INDEX "personer_fnamn_idx" on "personer" using btree ( lower ("fnamn") "text_ops" );
CREATE UNIQUE INDEX "personer_personid_index" on "personer" using btree ( "personid" "int4_ops" );
CREATE INDEX "personer_personid_fname_idx" on "personer" using btree ( "personid" "int4_ops", "fnamn" "text_ops" );
Cheers,
Palle
