Migraine

Migraine helps with painful data migrations.

It provides a framework for running cross-model and SQL-to-model data migrations for Django. Migraine's Migrator classes provide a declarative approach to importing data from external databases and Django models into other Django models, with a syntax somewhat similar to Django's ModelForms. Migraine will also run you your migrations in order derived from inter-migration dependencies.

Building a migration project

To use Migraine, you will need to create a project containing two basic elements: a migrators package containg one module per Django app and a bootstrap script you will use to set up django settings and start a migration.

Migraine projects are recommended to be placed outside of your main application's source code, so make sure the target app is available on the PYTHONPATH. You can append its path in an environment variable or in the migrate.py script.

Assuming we want to migrate to a single app called polls, here is how our project structure will look like:

polls_migration/
    __init__.py
    migrate.py
    migrators/
        __init__.py
        polls.py

We created a migrate.py module that will contain our configuration code, and a polls.py module where we will define our migrator classes.

Writing a bootstrap script

Your migrate.py script needs to do two things:

1. Import your migrators package.
2. Call run_from_command_line.

You can also use it for additional configuration, like loading Django settings.

Here is a basic example:

#!/usr/bin/env python
import sys
from migraine import run_from_command_line
import migrators

if __name__ == "__main__":
    run_from_command_line(migrators, sys.argv)

Defining Migrators

A Migrator class defines how we want to process the data we're going to migrate. It can be any class providing a run_migration method. Inside each of the migrators package submodules define a list called migrators, containing names of classes from that submodule that you wish to be detected by migraine's migration-running mechanism.

Model to Model migrations

Migraine provides a base ModelToModelMigrator that will create a single record in the target model per each record from the source model. We will use it to migrate data from an old model called OldPoll to a fresh model called NewPoll.

# our app's models:

from django.db import models

class OldPoll(models.Model):
    old_poll_name = models.CharField(max_length=30)

class NewPoll(models.Model):
    new_poll_name = models.CharField(max_length=36)


# migrators/polls.py:

from migraine.migrators import ModelToModelMigrator
from polls.models import OldPoll, NewPoll

migrators = ['PollsMigrator']

class PollsMigrator(ModelToModelMigrator):
    source_model = OldPoll
    target_model = NewPoll
    fields = [
        ('old_poll_name', 'new_poll_name')
    ]

We've just created a Migrator that will copy over OldPolls to NewPolls.

You can also define more complex rules for processing fields. Let's assume we want the old polls' names to end with "(old)". For each such field we can define a method that will return a processed value. Migraine uses a convention of prepending such methods' names with import_:

class AppendingPollsMigrator(ModelToModelMigrator):
    source_model = OldPoll
    target_model = NewPoll

    def import_new_poll_name(self, source):
        return source.old_poll_name + ' (old)'

Effect of running such a migration will be identical to running

new_poll.new_poll_name = source.old_poll_name + ' (old)'

for each newly created NewPoll object.

Instead of a source_model, you can also define a query_set field if you need more control over source data.

SQL table to model migrations

Migraine can handle importing data from a raw SQL database. For this, there is an SQLToModelMigrator.

from blog.models import Author, BlogPost
migrators = ['BlogPostMigrator']

class BlogPostMigrator(SQLToModelMigrator):
    source_db = 'oldblog'
    source_table = 'blog_posts'
    target_model = BlogPost
    skip_on_match = ['name']
    fields = [
        ('title', 'title'),
        ('content', 'content'),
    ]

    def import_author(self, source):
        return Author.objects.get_or_create(name=source['author_name'])

This simple example will populate the BlogPost model with data from blog_post table's rows. The import_ methods' source argument contains a dict mapping column names to values for each of source table's rows.

The source_db field declares the database to be used. The database needs to be decared in the DATABASES dict in django settings. It is optional and defaults to default.

Intead of source_table, you can define an sql field. This will cause the Migrator to use query's result rows as the source feed.

Running migrations

To launch all migrations, run your bootstrap script:

python migrate.py

You can also specify individual migrations to run. To see a list of available migrations run migrate.py --list.

Migrator dependencies

Migraine can sort your migrations using topological sorting based on inter-migration dependencies. To use this feature, declare a depends_on field on your Migrators that will contain a list of migrator names:

# migrators/foo.py
migrators = ['MigratorA', MigratorB']

class MigratorA:
    depends_on = ['foo.MigratorB']
    # ...

class MigratorB:
    # ...

In this example, MigratorB will always be run before MigratorA.

Running tests

cd testapp
pip install -r requirements.txt  # you probably want to make a virtualenv
                                 # for this
DJANGO_SETTINGS_MODULE=settings PYTHONPATH=`pwd` py.test