Re: Schema versioning in postgres

talk2abhinav(at)gmail(dot)com (abhinav mehrotra) writes:
> My database schema is susceptible to change between various release of
> the product. There may be new fields added or old fields removed from
> a table.
> Introduction of new fields seems to simple, but removal/truncated or
> data type does not seem to very direct.
>
> Does postgres has any support for such schema versioning?

Postgres generally conforms to the SQL standard, which (fairly properly)
does not offer any particular functionality surrounding schema
versioning.

A technique that seems commonly used is to record a version number in
the database, whether within a table, or as the name of a table:

insert into version_info.schema_version (system, version, installed)
select 'my_system', '2.2.3', now();

On one of my applications, I encode relevant stuff as a series of tables:

cbbapp(at)localhost-> \dt
List of relations
Schema | Name | Type | Owner
--------------+--------------------------------------+-------+---------------------------
_cbbaversion | branch_is_1.0.9-SNAPSHOT | table | don't_need_to_know_who
_cbbaversion | generated_on_host_cbbrowne | table | don't_need_to_know_who
_cbbaversion | schema_generated_at_2011-01-28 15:28 | table | don't_need_to_know_who
_cbbaversion | svn_version_7321 | table | don't_need_to_know_who
(4 rows)

We've "saved the day" a few times by having applications set up to
refuse to start up (e.g. - indicate a FATAL error) if the schema version
found did not match the version that an application expects.

This requires no special functionality from Postgres beyond allowing the
developer to:

- INSERT new tuples into a version table
- DROP/CREATE/RENAME tables
--
let name="cbbrowne" and tld="gmail.com" in name ^ "@" ^ tld;;
http://linuxfinances.info/info/finances.html
The world's full of apathy, but I don't care.