Andrew Snow wrote:
> > So we were advised to use "fulltextindex".- chunk
> > this fields on single words and make new table with words and oids in
> > it. After we made this table its size was 2940360 records. And I tried
> > to measure the time:
> >
> > ....
> >
> > select f1.id from app_fti f1, app_fti f2 where f1.string~*'visual' and
> > f2.string~*'basic' and f1.id=f2.id;
>
> What is causing it to be so slow is the use of regular expression pattern
> matching. For reasons I don't know, regular expression evaluation in a
> query is fairly slow in PostgreSQL. But since you have (correctly, I think)
> made a full text index, why do you still need to use regexp's ?
>
> Replace the ~* operator in the query above with 'LIKE', and you should find
> it several times faster.
>
> Please let us know the results of your testing.
>
> - Andrew
Thank you gentelmen! It seems that regexp was the reason! I enabled
likeplanning and it i=didn't improve the performance until I replaced '~*' with
'LIKE'. Thank you again!
V.