Обсуждение: table design strategy

-----Original Message-----
From: Johnson, Shaunn
Sent: Monday, March 25, 2002 4:15 PM
To: pgsql-general@postgresql.org
Subject: Re: [GENERAL] file size issue?

--I think you've answered at least 1/2 of my question,

Andrew.

 

--I'd like to figure out if Postgres reaches a point where

it will no longer index or vacuum a table based on its size (your answer

tells me 'No' -  it will continue until it is done, splitting each

table on 1Gig increments).

 

--And if THAT is true, then why am I getting failures when

I'm vacuuming or indexing a table just after reaching 2 Gig?

 

--And if it's an OS (or any other) problem, how can I factor

out Postgres?

 

--Thanks!

 

-X

 

 

[snip]

 

 

> Has anyone seen if it is a problem with the OS or with the way

> Postgres handles large files (or, if I should compile it again

> with some new options).

 

 

What do you mean "postgres handles large files"? The filesize

problem isn't related to the size of your table, because postgres

splits files at 1 Gig.

If it's an output problem, you could see something, but you said you

were vacuuming.

 

 

A

[snip]

-- Thanks for responding:

-- Regarding the trigger suggestion, I'm not sure
-- I follow  (translation: I wouldn't know a trigger from the gun).

-- I'm looking at my PostgreSQL book and looking that
-- feature up ...

-- While I do, what about this:

-- * Append all of the 2000 and 2001 tables into one table
-- * Make index of columns in the one table
-- * For updates and new imports, create a
-- method to only append new data / updates
-- (I imagine this is where the trigger function comes
-- in, right?) when I have to pull data (the source tables
-- will be much smaller ... hopefully ...

-- I have quite a lot to research ... any suggestions
-- will be appreciated.

-- Thanks again!

-X

-----Original Message-----
From: Andrew Sullivan [mailto:andrew@libertyrms.info]

> Howdy:
> 
> Running Postgres 7.1.3 on RedHat 7.2, kernel 2.4.7  rel. 10.
> 
> My question is about table design strategies ...
> 
> I have a imported a table, broken up in such a fashion that
> there are now 8 tables on Postgres.  For example: table_north,
> table_south, table_east and table_west originally comes from
> another source on another database called 'table_direction'.
> >From that, the tables span years, so, I have tables called
> 'table_north_2000' and 'table_north_2001', 'table_south_2000'
> and table_south_2001' and so on ...
> 
> Now ... I was thinking that now that I have all 8 parts, I'd like
> to:
> 
> * create indices on the similar names in each table
> * create a view that joins all 8 tables into ONE table again

That view is going to be _painful_ in performance.  What if you wrote
a trigger that inserts/deletes/updates in table_all when any of the
other tables are touched?  It's a lot of disk, but disk is cheap.

> PS: Has anyone had a chance to test a Data Model / database
> structure modeling tool (for creating pretty pictures and relational
> info / documentation)?

DbVisualiser turns out to be much better than I expected.  You can
also get a program called postgres_autodoc.pl if you just want to
create pretty UML diagrams.  The latter requires dia, the former Java
2.  (I don't have references handy, but a google search should get
you there.)

A

On Thu, Mar 28, 2002 at 12:28:40PM -0500, Johnson, Shaunn wrote:
> -- While I do, what about this:
>
> -- * Append all of the 2000 and 2001 tables into one table
> -- * Make index of columns in the one table
> -- * For updates and new imports, create a
> -- method to only append new data / updates
> -- (I imagine this is where the trigger function comes
> -- in, right?) when I have to pull data (the source tables
> -- will be much smaller ... hopefully ...

Right, the trigger will only fire on insert/update/delete, so you'd
have to populate the big table manually.  But after that, with a
trigger on all the source tables, any change you make there will also
be reflected in table_all.

Note that doing it this way is a very un-SQL-ish thing to do, in that
you'll have redundant data all over the place.  Still, given what you
already have, doing it this way might be a quick and dirty answer for
looking at the big table.  Of course, if you hardly ever reference
the big table, maybe it isn't worth the storage cost.

A
--
----
Andrew Sullivan                               87 Mowat Avenue
Liberty RMS                           Toronto, Ontario Canada
<andrew@libertyrms.info>                              M6K 3E3
                                         +1 416 646 3304 x110