Re: reserving space in a rec for future update - Mailing list pgsql-general

From Andrew Sullivan
Subject Re: reserving space in a rec for future update
Date
Msg-id 20071114161257.GR16041@crankycanuck.ca
Whole thread Raw
In response to reserving space in a rec for future update  ("Gauthier, Dave" <dave.gauthier@intel.com>)
Responses Re: reserving space in a rec for future update  ("Gauthier, Dave" <dave.gauthier@intel.com>)
Re: reserving space in a rec for future update  (Mike Charnoky <noky@nextbus.com>)
List pgsql-general
On Wed, Nov 14, 2007 at 10:28:30AM -0500, Gauthier, Dave wrote:
> null.   My concern has to do with record fragmentation at the time of
> update because there's no room to "expand" them to accept the non-null
> data.  (BTW, the columns are floating point).

You have a mistaken idea about how this works.

> Is there a way to initially insert nulls, but reserve space for the
> future update (and avoid record fragmentation)?

No.

> Is my record fragmentation concern unfounded?

Sort of.

The way this will work in Postgres is that, when you UPDATE the row, the old
row will be marked dead, and a _new_ row will be written out with the new
data.  You will need to perform VACUUM in order to keep the table from
bloating.  You'll want to read the manual carefully about this topic, in
order to keep your table from getting so bloated that your free space map
becomes useless.  One of the weakest areas for PostgreSQL is its behaviour
under this sort of "most rows updated" scenario, and it is wise to plan
carefully how you will accomplish these sorts of activities without causing
yourself extreme pain.

A

--
Andrew Sullivan
Old sigs will return after re-constitution of blue smoke

pgsql-general by date:

Previous
From: "Scott Marlowe"
Date:
Subject: Re: reserving space in a rec for future update
Next
From: "Jan de Visser"
Date:
Subject: Re: Using generate_series to create a unique ID in a query?