django-mass-migration

Django Mass Migration

A Django app which provides utilities for performing data operations on (optionally) large datasets.

Similar to Django's built-in migration system, it allows you to define data migrations to be performed on the database, and to apply those migration operations to your database, and it tracks which migrations have been applied and which haven't.

Unlike Django's built-in migration system, it is designed for performing long-running data modification tasks where there could be a significant amount of time between when the migration is started and when it's finished. See Concepts. It is particularly useful for schemaless databases where Django's normal concept of schema changes doesn't apply.

Due to the expectation that migrations maybe be long-running, and just for awesomeness, it provides a web based (rather than terminal based) interface for managing the migrations.

The actual running of the operations can be done on a "backend" of your choosing, e.g. a task queue. A backend for running migration operations on Djangae with Glcoudc is bundled with the app.

Installation

1. Install the package: pip install django-mass-migration
2. Add massmigration to settings.INSTALLED_APPS.
3. Add path("migrations/", include("massmigration.urls")) to your root urlconf.
4. If you're using the bundled backend, set settings.MASSMIGRATION_TASK_QUEUE to a Google Cloud Tasks queue name.

Creating A Migration

To create a new migration, run the makemassmigration management command:

python manage.py makemassmigration myapp my_migration_name

This command has two required, positional arguments:

1. app_label - this should be the name of an installed app in your project.
2. migration_name - this should be the name that you want to give you migration.

The command also takes the following optional arguments:

• --template - this should be the name of one of the supplied templates in massmigration/templates/migration_templates, without the extension, e.g. --template=mapper. See Migration Types. The default is mapper.

A file will be created inside a folder called 'massmigrations' in your app, using the supplied migration name. E.g. myapp/massmigrations/001_my_migration.py. This is a blank slate for you to write your migration code in.

Migration Types

There are three broad types of migration for you to choose from.

simple

This is for a migration which is just a single Python function. It's intended for small operations which you know your backend can perform in one step.

mapper

This is for mapping a function over the instances of a Django queryset. You define a queryset and an operation function, and the migration will call that function on each object in the queryset.

The backend must be able to handle iterating over the queryset. The bundled DjangaeBackend can handle almost infinite sized querysets.

custom

If you want to take matters into your own hands you can write an entirely custom migration. These can still be tracked the same as the other operations, but the implementation of what your migration does and how your the backend handles it are up to you.

Running with Multiple DBs

If your project connects to multiple databases it is possible to run your migrations against each database individually. The migrations are not forced on a specific DB but rather the db_alias is passed to both to get_queryset(db_alias) (for mappers migrations) and operation (for simple migrations) giving control back to the developer to customise what is retrieved for the migration.

By default a migration can be run on any available database. That can be customised but setting allowed_db_aliases on the migration class.

Applying Migrations

There are two ways to apply a migration to the database:

Via the Web UI

1. Optionally: ensure that the Django admin is set up in your project.
2. Deploy your project, including the new migration file.
3. Either:
   • Go to the Django Admin site and under Mass Migrations -> Migration Record, click the "Manage Migrations" button; or
   • Go directly to the URL of reverse("massmigration_manage") whatever path you've configured that to be on.
4. Next to your new migration, click "Run...".
5. Click "Run migration".
6. Wait for the migration to be listed as applied in the Migration Record list view, or check the logging from your backend.

Programmatically

If you want to create your own system for applying migrations, you can use the API functions. Or you will be able to, once I've written them.

Protecting Code Which Requires Migrations

Sometimes in the development cycle of your application, you will want to ensure that a particular migration has been run, usually when you've added new code which expects the migration to have been applied before the code is run.

To handle this, Mass Migration provides the following utilities:

massmigration.enforcement.requires_migration

This is a function decorator which ensures that the function cannot be executed until the specified migration has been applied. For example:

@requires_migration(("myapp", "0001_my_migration"))
def my_function():
	...

This will raise massmigration.exceptions.RequiredMigrationNotApplied if the function is called before the migration is applied.

Multiple DBs Support

When running on multiple databases, by default, the utilities assume that the migration must have been applied on all the databases specified by the Migration's allowed_db_aliases attribute. To customise that behaviour the decorators accept an optional db_aliases list to specify a subset of the aliases allowed for the migration and require only those to have been applied.

@requires_migration(("myapp", "0001_my_migration"), db_aliases=["my_db_alias"])
def my_function():
	...

Notes:

• If the migration is applied, then this will cache that fact, so it will only query the database the first time the function is called.
• The migration identifier can either be a tuple of (app_label, migration_name) or can be a string of "app_label:migration_name".
• There is an optional second argument skip_in_tests, which defaults to True.

massmigration.enforcement.view_requires_migration

This is the same as requires_migration but for decorating a view function. If the specified migration is not applied then the view will return a 503 response.

Concepts

Why Mass Migration?

Reason 1

Django's migration system is built for relational databases. These have a table schema and allow you to make a change to that schema as a single "operation", often inside a transaction. This means that a change to a schema is either "applied" or "not applied".

On schemaless databases (such as Google's Cloud Datastore/Firestore), there is no schema, and therefore making a change to a "table", such as adding or removing a column, requires mapping over each row of the table to apply that change. This means that a change to a schema can be "applied", "not applied" or "being applied".

Django's migration system doesn't allow for this in-between state. Its internal Migration model (which is nested inside the MigrationRecorder and cannot be edited) which tracks migrations cannot store the fact that a migration is currently being applied; a migration is either applied or it isn't.

This makes Django's migration recorder unsuitable for tracking changes to a schemaless database.

Reason 2

If you're using a schemaless database with Django, e.g. you're using Google Cloud Datastore with a connector such as gcloudc then making a change to your model doesn't necessarily require you to create any kind of migration at all.

For example, if you add a new field to one of your models and you specify a default value, you can just start using that new field with no need for a migration. Existing rows which don't yet have that column will simply have the column added the next time they're resaved via your model.

But in other cases, the more limited querying capabilities of a schemaless database mean that there are times when something which could be done with an UPDATE table statement on SQL can't be done so easily. For these cases, django-mass-migration provides tools to let you iterate over your table rows and perform data manipulation operations.

General approach

TODO: long-running migrations, errors, etc. Why long-running migrations are not run as part of the deployment process (because they take too long and you don't want to take your site down while they run).

Model State

TODO: Stuff about why there's no model history (apps/schema_editor).

Workflow & Code Protection

TODO: Deployment workflow and use of enforcement utilities

Settings

The following settings can be used:

MASSMIGRATION_BACKEND

This should be a dotted path string to the backend class that you want to use. The default is "massmigration.backends.djangae.DjangaeBackend", which works with SQL, Google Cloud Datastore and Firestore databases.

MASSMIGRATION_TASK_QUEUE

Used by the DjangaeBackend, this sets the Google Cloud Tasks queue name to be used for running migration tasks.

MASSMIGRATION_RECORD_CACHE_TIMEOUT

You're unlikely to need this. It sets the time for caching MigrationRecords for the purpose of checking a migration's status during mapper operations.

Backends

Currently massmigration includes a DjangaeBackend for running migrations on Google App Engine applications which are using djangae.

Due to the nature of different hosting platforms having different types of task queue systems, each platform will require a slightly different backend, but these can easily be written and plugged in (merge requests welcome!).

DjangaeBackend

This runs simple migrations using djangae.tasks.deferred and runs mapper migrations using djangae.tasks.defer_iteration_with_finalize.

It can be configured via the backend_params attribute on your Migration classes, using the following two items:

• defer_kwargs: a dict of kwargs which will get passed to the defer call for simple migrations.
• defer_iteration_with_finalize_kwargs - a dict of kwargs which will get passed through to defer_iteration_with_finalize for mapper migrations.