Current location

narf Source control manager Git

diff options
authorChristopher Allan Webber <>2016-03-26 11:37:52 -0700
committerChristopher Allan Webber <>2016-03-26 11:39:08 -0700
commitdcccf654f57cb884f66b4da41121b991bdbb4071 (patch)
parent20adec8f26e8b88a6fb07c0be342170893758370 (diff)
doc: Update developer migration docs for Alembic as primary migration system
* docs/source/devel/migrations.rst: Update to reflect sqlalchemy-migrate as being deprecated and Alembic as being the new way of doing migrations.
1 files changed, 22 insertions, 22 deletions
diff --git a/docs/source/devel/migrations.rst b/docs/source/devel/migrations.rst
index 16c02b04..5e0a6bee 100644
--- a/docs/source/devel/migrations.rst
+++ b/docs/source/devel/migrations.rst
@@ -28,35 +28,35 @@ every migration is run with dbupdate.
There's a few things you need to know:
-- We use `sqlalchemy-migrate
+- We use `Alembic <>`_ to run
+ migrations. We also make heavy use of the
+ `branching model <>`_
+ for our plugins. Every plugin gets its own migration branc.
+- We used to use `sqlalchemy-migrate
See `their docs <>`_.
-- `Alembic <>`_ might be a better
- choice than sqlalchemy-migrate now or in the future, but we
- originally decided not to use it because it didn't have sqlite
- support. It's not clear if that's changed.
+ sqlalchemy-migrate is now only kept around for legacy migrations;
+ don't add any more! But some users are still using older databases,
+ and we need to provide the intermediary "old" migrations for a while.
- SQLAlchemy has two parts to it, the ORM and the "core" interface.
We DO NOT use the ORM when running migrations. Think about it: the
ORM is set up with an expectation that the models already reflect a
certain pattern. But if a person is moving from their old patern
and are running tools to *get to* the current pattern, of course
their current database structure doesn't match the state of the ORM!
-- How to write migrations? Maybe there will be a tutorial here in the
- future... in the meanwhile, look at existing migrations in
- `mediagoblin/db/` and look in
- `mediagoblin/tests/` for examples.
-- Common pattern: use `inspect_table` to get the current state
- of the table before we run alterations on it.
-- Make sure you set the RegisterMigration to be the next migration in
- order.
-- What happens if you're adding a *totally new* table? In this case,
- you should copy the table in entirety as it exists into
- then create the tables based off of that... see
- add_collection_tables. This is easier than reproducing the SQL by
- hand.
-- If you're writing a feature branch, you don't need to keep adding
- migrations every time you change things around if your database
- structure is in flux. Just alter your migrations so that they're
- correct for the merge into master.
+ Anyway, Alembic has its own conventions for migrations; follow those.
+- Alembic's documentation is pretty good; you don't need to worry about
+ setting up the migration environment or the config file so you can
+ skip those parts. You can start at the
+ `Create a Migration Script <>`_
+ section.
+- *Users* should only use `./bin/gmg dbupdate`. However, *developers*
+ may wish to use the `./bin/gmg alembic` subcommand, which wraps
+ alembic's own command line interface. Alembic has some tools for
+ `autogenerating migrations <>`_,
+ and they aren't perfect, but they are helpful. (You can pass in
+ `./bin/gmg alembic --with-plugins revision --autogenerate` if you need
+ to include plugins in the generated output; see the
+ :ref:`plugin database chapter <plugin-database-chapter>` for more info.)
That's it for now! Good luck!