Re: ALTER TABLE ADD COLUMN fast default

Andres Freund <andres@anarazel.de>

From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>, David Rowley <david.rowley@2ndquadrant.com>, Thomas Munro <thomas.munro@enterprisedb.com>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2018-03-29T21:31:29Z
Lists: pgsql-hackers
Hi,

On 2018-03-29 17:27:47 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > There's plenty databases with pg_attribute being many gigabytes large,
> > and this is going to make that even worse.
> 
> Only if you imagine that a sizable fraction of the columns have fast
> default values, which seems somewhat unlikely.

Why is that unlikely?  In the field it's definitely not uncommon to
define default values for just about every column. And in a lot of cases
that'll mean we'll end up with pg_attribute containing default values
for most columns but the ones defined at table creation. A lot of
frameworks make it a habit to add columns near exclusively in
incremental steps.  You'd only get rid of them if you force an operation
that does a full table rewrite, which often enough is impractical.

Greetings,

Andres Freund


Commits

  1. Clean up treatment of missing default and CHECK-constraint records.

  2. Fast ALTER TABLE ADD COLUMN with a non-NULL default

  3. Fix application of identity values in some cases