Re: [GENERAL] Creation of tsearch2 index is very slow

On Sat, 21 Jan 2006, Martijn van Oosterhout wrote:

> On Sat, Jan 21, 2006 at 04:29:13PM +0300, Oleg Bartunov wrote:
>> Martijn, you're right! We want not only to split page to very
>> different parts, but not to increase the number of sets bits in
>> resulted signatures, which are union (OR'ed) of all signatures
>> in part. We need not only fast index creation (thanks, Tom !),
>> but a better index. Some information is available here
>> http://www.sai.msu.su/~megera/oddmuse/index.cgi/Tsearch_V2_internals
>> There are should be more detailed document, but I don't remember where:)
>
> I see how it works, what I don't quite get is whether the "inverted
> index" you refer to is what we're working with here, or just what's in
> tsearchd?

just tsearchd. We plan to implement inverted index into PostgreSQL core
and then adapt tsearch2 to use it as option for read-only archives.

>
>>> That's harder though (this algorithm does approximate it sort of)
>>> and I havn't come up with an algorithm yet
>>
>> Don't ask how hard we thought :)
>
> Well, looking at how other people are struggling with it, it's
> definitly a Hard Problem. One thing though, I don't think the picksplit
> algorithm as is really requires you to strictly have the longest
> distance, just something reasonably long. So I think the alternate
> algorithm I posted should produce equivalent results. No idea how to
> test it though...

you may try our development module 'gevel' to see how dense is a signature.

www=# \d v_pages
Table "public.v_pages"
Column | Type | Modifiers
-----------+-------------------+-----------
tid | integer | not null
path | character varying | not null
body | character varying |
title | character varying |
di | integer |
dlm | integer |
de | integer |
md5 | character(22) |
fts_index | tsvector |
Indexes:
"v_pages_pkey" PRIMARY KEY, btree (tid)
"v_pages_path_key" UNIQUE, btree (path)
"v_gist_key" gist (fts_index)

# select * from gist_print('v_gist_key') as t(level int, valid bool, a gtsvector) where level =1;
level | valid | a
-------+-------+--------------------------------
1 | t | 1698 true bits, 318 false bits
1 | t | 1699 true bits, 317 false bits
1 | t | 1701 true bits, 315 false bits
1 | t | 1500 true bits, 516 false bits
1 | t | 1517 true bits, 499 false bits
(5 rows)

Regards,
Oleg
_____________________________________________________________
Oleg Bartunov, Research Scientist, Head of AstroNet (www.astronet.ru)
Sternberg Astronomical Institute, Moscow University, Russia
Internet: oleg(at)sai(dot)msu(dot)su, http://www.sai.msu.su/~megera/
phone: +007(495)939-16-83, +007(495)939-23-83