Changefiles
The simplest format for tracking database changes - the format you may already be using - is to keep
each changeset in a .sql file. These files group together changes, and allow you to track them in a
version control system.
- You should store each script in a separate file. The contents of the file will be
issued as a single sql command.
- By default, scripts will be applied in alphabetical order. Name your script file with the
date first. A script named 2004-12-13-update-customer-types.sql will always be applied in an intuitive and
predictable order.
- Change files are immutable. Once a changefile has been committed to your VCS, it
should never be edited. If you need to alter a script that you have already committed, write a new script instead of
editing the existing script.
If someone were to apply your change script, and then the script was
subsequently modified, what should happen? Should the change be reapplied? No. Therefore, we track
the checksum for each script, and use the checksum to determine whether a candidate change script has
already been applied to the selected database.
- At the risk of being obvious, all database change scripts must be stored in a version control system.
When a new release is built from the VCS, all database changes should be included automatically. Also,
edits to a script that has already been created and distributed are very very dangerous, and could put the
database schema into an inconistent state (from the point of view of the application).
Why not use an xml defined format? Short answer: maybe. We just haven't had a need for it yet.