My company is having trouble managing how we upgrade schema changes across many versions of our software. I imagine this is a common problem and there're probably some neat solutions that we don't know about.
For the last 10 years we have been writing bash shell scripts essentially numbered in order db0001, db0002, db0003.... The number represents the schema version which is recorded in the database and updated by the shell scripts. We have a template that provides all the functionality we need, we just copy the script and fill in the blanks. The schema upgrade scripts are committed to svn along with the software changes, and we have a process when installing the software at a site that runs the scripts on the DB in order before starting up the new version of the software.
This has worked really well so far. But we've effectively only had one version of the software in development at any time. We're now in the habit of branching the software to form releases to promote stability when making large changes. The idea is that only really important changes are merged in to the releases. This introduces a bit of a problem when some change needs to be merged from one release to another. The typical problem is that we might have two versions of the software 10.0 at schema version 10057 and 11.0 at 11023 and we need to merge an important bug fix from schema 11023 in to 10.0. The issue is that 11023 might depend upon changes introduced in the schema versions before it. Or 11023 might introduce changes that cause later scripts to break (11000 - 11023) when upgrading from 10.0 to 11.0.
One potential solution is to require that schema changes are never merged in to a release, but of course sometimes business requires we do =(
I'm really interested to hear how you guys manage schema upgrades in the face of branches and upgrading from many different versions of the database.
I've been reading http://pyrseas.wordpress.com/2011/02/07/version-control-part-2-sql-databases/ but I have a feeling that this blog post won't address branches.
pgsql-general by date
|Next:||From: Rob Sargent||Date: 2011-02-10 21:58:15|
|Subject: Re: Schema version control|
|Previous:||From: Raymond O'Donnell||Date: 2011-02-10 19:43:43|
|Subject: Re: COPY statement REAL vs VARCHAR precision issue|