Tsearch2 trigger firing... - Mailing list pgsql-general

From Net Virtual Mailing Lists
Subject Tsearch2 trigger firing...
Date
Msg-id 20041015093924.24704@mail.e-na.net
Whole thread Raw
In response to Rules and locking within a transaction?...  ("Net Virtual Mailing Lists" <mailinglists@net-virtual.com>)
Responses Re: Tsearch2 trigger firing...  (Oleg Bartunov <oleg@sai.msu.su>)
List pgsql-general
Hello,

I have a table that uses tsearch2 and, of course, and index and trigger
to keep everything updated.....  Something like:

CREATE TABLE sometable (
    id           SERIAL,
    someinteger  INTEGER
    sometext     TEXT,
    sometext2    TEXT,
    sometext3    TEXT,
    sometext_fti TSVECTOR
);

UPDATE sometable SET sometext_fti=to_tsvector('default',
COALESCE(sometext, ''));
CREATE INDEX sometext_fti_idx ON sometable USING gist(sometext_fti);
CREATE TRIGGER text_update BEFORE UPDATE OR INSERT ON sometable FOR EACH
ROW EXECUTE PROCEDURE tsearch2(sometext_fti, sometable);


There are two issues:

#1. I need a way to be able to do a full-text search any combination of
sometext, sometext2, and sometext3... So far the only way I've been able
to figure out to do this is to actually create additional fields and
indexes for all combinations (a total of 7), like the following:

CREATE TABLE sometable (
    id            SERIAL,
    someinteger   INTEGER,
    sometext_fti  TSVECTOR,
    sometext      TEXT,
    sometext_fti  TSVECTOR,
    sometext2     TEXT,
    sometext2_fti TSVECTOR,
    sometext3     TEXT,
    sometext3_fti TSVECTOR

    sometext1_2_fti TSVECTOR
    sometext1_3_fti TSVECTOR
    sometext2_3_fti TSVECTOR

    sometext1_2_3_fti TSVECTOR

);

UPDATE sometable SET sometext_fti=to_tsvector('default',
COALESCE(sometext, ''));
CREATE INDEX sometext_fti_idx ON sometable USING gist(sometext_fti);
CREATE TRIGGER sometext_update BEFORE UPDATE OR INSERT ON sometable FOR
EACH ROW EXECUTE PROCEDURE tsearch2(sometext_fti, sometable);

UPDATE sometable SET sometext2_fti=to_tsvector('default',
COALESCE(sometext2, ''));
CREATE INDEX sometext2_fti_idx ON sometable USING gist(sometext2_fti);
CREATE TRIGGER sometext2_update BEFORE UPDATE OR INSERT ON sometable FOR
EACH ROW EXECUTE PROCEDURE tsearch2(sometext2_fti, sometable);

UPDATE sometable SET sometext3_fti=to_tsvector('default',
COALESCE(sometext3, ''));
CREATE INDEX sometext3_fti_idx ON sometable USING gist(sometext3_fti);
CREATE TRIGGER sometext3_update BEFORE UPDATE OR INSERT ON sometable FOR
EACH ROW EXECUTE PROCEDURE tsearch2(sometext3_fti, sometable);

UPDATE sometable SET sometext1_2_fti=to_tsvector('default',
COALESCE(sometext1, '')||' '||COALESC(sometext2, ''));
CREATE INDEX sometext1_2_fti_idx ON sometable USING gist(sometext1_2_fti);
CREATE TRIGGER sometext1_2_update BEFORE UPDATE OR INSERT ON sometable
FOR EACH ROW EXECUTE PROCEDURE tsearch2(sometext_1_2_fti, sometable);

UPDATE sometable SET sometext1_3_fti=to_tsvector('default',
COALESCE(sometext1, '')||' '||COALESC(sometext3, ''));
CREATE INDEX sometext1_3_fti_idx ON sometable USING gist(sometext1_3_fti);
CREATE TRIGGER sometext1_3_update BEFORE UPDATE OR INSERT ON sometable
FOR EACH ROW EXECUTE PROCEDURE tsearch2(sometext1_3_fti, sometable);

UPDATE sometable SET sometext2_3_fti=to_tsvector('default',
COALESCE(sometext2, '')||' '||COALESC(sometext3, ''));
CREATE INDEX sometext2_3_fti_idx ON sometable USING gist(sometext2_3_fti);
CREATE TRIGGER sometext2_3_update BEFORE UPDATE OR INSERT ON sometable
FOR EACH ROW EXECUTE PROCEDURE tsearch2(sometext2_3_fti, sometable);

UPDATE sometable SET sometext1_2_3_fti=to_tsvector('default',
COALESCE(sometext1, '')||' '||COALESCE(sometext2, '')||'
'||COALESC(sometext3, ''));
CREATE INDEX sometext1_2_3_fti_idx ON sometable USING gist(sometext1_2_3_fti);
CREATE TRIGGER sometext1_2_3_update BEFORE UPDATE OR INSERT ON sometable
FOR EACH ROW EXECUTE PROCEDURE tsearch2(sometext1_2_3_fti, sometable);

... of course, this creates an enormous load on the server for any
update/delete operation...  I tried just creating 3 indexes (one for each
field) and doing a query like:

SELECT * FROM sometable WHERE sometext1 @@ to_tsquery('default',
'postgres') OR sometext2 @@ to_tsquery('default', 'postgres') OR
sometext3 @@ to_tsquery('default', 'postgres') -- but found this to be
enormously expensive for queries....

.. with the arrangement above, I make some decision as to which field to
query, for example:

SELECT * FROM sometable WHERE sometext1_2_3 @@ to_tsquery('default',
'postgres'); -- this executes MUCH faster


.. is there some way to get the best of both worlds here or am I trying
to jam a square peg into a round hole?...

#2. When doing an update on the above table, such as: "UPDATE sometable
SET someinteger=0", it ends up firing off the triggers which of course
takes a long time to update... Is there someway to make the trigger only
fire if a the field it is tied to is updated?....    This is not a

Thanks as always!


- Greg



pgsql-general by date:

Previous
From: Alexander Pucher
Date:
Subject: Mathematical operations with NULL values
Next
From: Martijn van Oosterhout
Date:
Subject: Re: 'NOW' in UTC with no timezone