I have a web application using sqlalchemy (within Pylons). I need to effiently change the schema to be able to change the production version at least on a daily basis, maybe more, without losing the data.
I have played a little bit with sqlalchemy-migrate over the week-end and I would say that it gave me a bad impression. First I think it cannot help with migration between two databases engines; that's something that could probably be done with sqlalchemy alone. Second the docs do not seem up to date. I had to change some command-line options, like giving the repository path at each command, this could be a bug of migrate.
But the worst thing it the "manage.py test" command. Not only it actually modifies the database (this point is clearly indicated in the documentation so I can't blame migrate) but my first migration script just made plain stupid schema migration, leaving the upgraded-downgraded db with a different schema than the original. But the "manage.py test" just answered something like
success !
That is, it did not even check if the schema was left in a coherent state. So is it worth using migrate? Is there any advantage compared to the Do It Yourself method associated with good practices as proposed by S.Lott ? Are there alternatives to sqlalchemy-migrate actually simplifying the migration process or am I just trying to use migrate with a bad a priori (then please show me why is't clearly superior to creating CSV columns as proposed in the link above)?
Many Thanks!
Use Alembic instead:
http://pypi.python.org/pypi/alembic
Thanks for comments, edited to add some reasoning --
It's developed by the author of SQLAlchemy, and it's brand new and well supported. I don't know enough about sqlalchemy-migrate to give a good comparison. But I took a quick read through the clear and concise Alembic docs, then got my own autogenerated migration working in a very short time.
Autogeneration: Not its only mode of operation, but if you choose, Alembic will read your application's sqlalchemy configuration (for instance, your declarative model classes that set up all your tables, constraints, and mappings) and compare to the actual current state of your database, and output a Python script that represents the delta between the two. You then pass that script to Alembic's upgrade command and there you go, the differences are resolved. A small amount of editing the migration script by hand is usually needed, and that's (a) just the nature of migrations, and (b) something you want to do anyway to make sure you were fully aware of the exact steps that the migration is going to perform before you run it.
Alembic brings a DVCS-like ability to the way your migrations are tracked, too. It makes it really easy to return to any past state of your db schema.