From: "Ahmad Fajar" <gendowo@konphalindo.or.id>
To: "'Oleg Bartunov'" <oleg@sai.msu.su>
Cc: <pgsql-performance@postgresql.org>
Subject: Re: tsearch2 seem very slow
Date: Mon, 26 Sep 2005 02:14:38 +0700
Keywords: Postgresql
MIME-Version: 1.0
Content-Type: text/plain;
	charset="US-ASCII"
Content-Transfer-Encoding: 7bit
In-Reply-To: <Pine.GSO.4.63.0509252212010.27150@ra.sai.msu.su>
Thread-Index: AcXB/4OudUZ5HDloQ+eQJFZPZjge7QAAnprA
Message-ID: <SVONE8rBKXGA7zyWR4A0000023b@ki-communication.com>

Hi Oleg, 

> what king of garbage ? Probably you index not needed token types, for
> example, email address, file names....

> do you need proximity ? If no, use strip(tsvector) function to remove
> coordinate information from tsvector.

I need proximity. Some time I have to rank my article and make a chart for
that.

> don't index default configuration and index only needed tokens, for 
> example, to index only 3 type of tokens, first create 'qq' configuration
> and specify tokens to index.

> insert into pg_ts_cfg values('qq','default','en_US');
-- tokens to index
> insert into pg_ts_cfgmap values('qq','lhword','{en_ispell,en_stem}');
> insert into pg_ts_cfgmap values('qq','lword','{en_ispell,en_stem}');
> insert into pg_ts_cfgmap values('qq','lpart_hword','{en_ispell,en_stem}');

I still don't understand about tsearch2 configuration, so until now I just
use default configuration. I will try your suggestion. But how can I get the
en_ispell? Does my system will know if I use: ....,'{en_ispell,en_stem}';
From default configuration I only see: ..., '{en_stem}';

> Beside that, I still have problem, if I do a simple query like:
> Select ids, keywords from dict where keywords='blabla' ('blabla' is a
single
> word); The table have 200 million rows, I have index the keywords field.
On
> the first time my query seem to slow to get the result, about 15-60 sec to
> get the result. I use latest pgAdmin3 to test all queries. But if I repeat
> the query I will get fast result. My question is why on the first time the
> query seem to slow.

> because index pages should be readed from disk into shared buffers, so 
> next query will benefit from that. You need enough shared memory to get 
> real benefit. You may get postgresql stats and look on cache hit ration.

> btw, how does your query ( keywords='blabla') relates to tsearch2 ?

(Keywords='blabla') isn't related to tsearch2, I just got an idea from
tsearch2 and try different approach. But I stuck on the query result speed.
Very slow to get result on the first query. 
And how to see postgresql stats and look on cache hit ratio? I still don't
know how to get it.

> I try to cluster the table base on keyword index, but after 15 hours 
> waiting and it doesn't finish I stop clustering. 

> don't use cluster for big tables ! simple
>  select *  into clustered_foo from foo order by indexed_field
> would be faster and does the same job.

What the use of clustered_foo table? And how to use it?
I think it will not distinct duplicate rows. And the clustered_foo table
still not have an index, so if query to this table, I think the query will
be very slow to get a result.

Regards,
ahmad fajar