Database schema versioning workflow

SQLAlchemy migrate provides the migrate.versioning API that is also available as the migrate command.

Purpose of this package is frontend for migrations. It provides commands to manage migrate repository and database selection as well as script versioning.

Project setup

Create a change repository

To begin, we’ll need to create a repository for our project.

All work with repositories is done using the migrate command. Let’s create our project’s repository:

$ migrate create my_repository "Example project"

This creates an initially empty repository relative to current directory at my_repository/ named Example project.

The repository directory contains a sub directory versions that will store the schema versions, a configuration file migrate.cfg that contains repository configuration and a script manage.py that has the same functionality as the migrate command but is preconfigured with repository specific parameters.

Note

Repositories are associated with a single database schema, and store collections of change scripts to manage that schema. The scripts in a repository may be applied to any number of databases. Each repository has an unique name. This name is used to identify the repository we’re working with.

Version control a database

Next we need to declare database to be under version control. Information on a database’s version is stored in the database itself; declaring a database to be under version control creates a table named migrate_version and associates it with your repository.

The database is specified as a SQLAlchemy database url.

The version_control command assigns a specified database with a repository:

$ python my_repository/manage.py version_control sqlite:///project.db my_repository

We can have any number of databases under this repository’s version control.

Each schema has a version that SQLAlchemy Migrate manages. Each change script applied to the database increments this version number. You can retrieve a database’s current version:

$ python my_repository/manage.py db_version sqlite:///project.db my_repository
0

A freshly versioned database begins at version 0 by default. This assumes the database is empty or does only contain schema elements (tables, views, constraints, indices, …) that will not be affected by the changes in the repository. (If this is a bad assumption, you can specify the version at the time the database is put under version control, with the version_control command.) We’ll see that creating and applying change scripts changes the database’s version number.

Similarly, we can also see the latest version available in a repository with the command:

$ python my_repository/manage.py version my_repository
0

We’ve entered no changes so far, so our repository cannot upgrade a database past version 0.

Project management script

Many commands need to know our project’s database url and repository path - typing them each time is tedious. We can create a script for our project that remembers the database and repository we’re using, and use it to perform commands:

$ migrate manage manage.py --repository=my_repository --url=sqlite:///project.db
$ python manage.py db_version
0

The script manage.py was created. All commands we perform with it are the same as those performed with the migrate tool, using the repository and database connection entered above. The difference between the script manage.py in the current directory and the script inside the repository is, that the one in the current directory has the database URL preconfigured.

Note

Parameters specified in manage.py should be the same as in versioning api. Preconfigured parameter should just be omitted from migrate command.

Making schema changes

All changes to a database schema under version control should be done via change scripts - you should avoid schema modifications (creating tables, etc.) outside of change scripts. This allows you to determine what the schema looks like based on the version number alone, and helps ensure multiple databases you’re working with are consistent.

Create a change script

Our first change script will create a simple table

account = Table(
    'account', meta,
    Column('id', Integer, primary_key=True),
    Column('login', String(40)),
    Column('passwd', String(40)),
)

This table should be created in a change script. Let’s create one:

$ python manage.py script "Add account table"

This creates an empty change script at my_repository/versions/001_Add_account_table.py. Next, we’ll edit this script to create our table.

Edit the change script

Our change script predefines two functions, currently empty: upgrade() and downgrade(). We’ll fill those in:

from sqlalchemy import Table, Column, Integer, String, MetaData

meta = MetaData()

account = Table(
    'account', meta,
    Column('id', Integer, primary_key=True),
    Column('login', String(40)),
    Column('passwd', String(40)),
)


def upgrade(migrate_engine):
    meta.bind = migrate_engine
    account.create()


def downgrade(migrate_engine):
    meta.bind = migrate_engine
    account.drop()

Note

The generated script contains * imports from sqlalchemy and migrate. You should tailor the imports to fit your actual demand.

As you might have guessed, upgrade() upgrades the database to the next version. This function should contain the schema changes we want to perform (in our example we’re creating a table).

downgrade() should reverse changes made by upgrade(). You’ll need to write both functions for every change script. (Well, you don’t have to write downgrade, but you won’t be able to revert to an older version of the database or test your scripts without it.) If you really don’t want to support downgrades it is a good idea to raise a NotImplementedError or some equivalent custom exception. If you let downgrade() pass silently you might observe undesired behaviour for subsequent downgrade operations if downgrading multiple versions.

Note

As you can see, migrate_engine is passed to both functions. You should use this in your change scripts, rather than creating your own engine.

Warning

You should be very careful about importing files from the rest of your application, as your change scripts might break when your application changes. Read more about writing scripts with consistent behavior.

Test the change script

Change scripts should be tested before they are committed. Testing a script will run its upgrade() and downgrade() functions on a specified database; you can ensure the script runs without error. You should be testing on a test database - if something goes wrong here, you’ll need to correct it by hand. If the test is successful, the database should appear unchanged after upgrade() and downgrade() run.

To test the script:

$ python manage.py test
Upgrading... done
Downgrading... done
Success

Our script runs on our database (sqlite:///project.db, as specified in manage.py) without any errors.

Our repository’s version is:

$ python manage.py version
1

Note

Due to #41 the database must be exactly one version behind the repository version.

Warning

The test command executes actual scripts, be sure you are NOT doing this on production database.

If you need to test production changes you should:

  1. get a dump of your production database

  2. import the dump into an empty database

  3. run test or upgrade on that copy

Upgrade the database

Now, we can apply this change script to our database:

$ python manage.py upgrade
0 -> 1...
done

This upgrades the database (sqlite:///project.db, as specified when we created manage.py above) to the latest available version. (We could also specify a version number if we wished, using the --version option.) We can see the database’s version number has changed, and our table has been created:

$ python manage.py db_version
1
$ sqlite3 project.db
sqlite> .tables
account migrate_version
sqlite> .schema account
CREATE TABLE account (
   id INTEGER NOT NULL,
   login VARCHAR(40),
   passwd VARCHAR(40),
   PRIMARY KEY (id)
);

Our account table was created - success!

Modifying existing tables

After we have initialized the database schema we now want to add another Column to the account table that we already have in our schema.

First start a new changeset by the commands learned above:

$ python manage.py script "Add email column"

This creates a new changeset template. Edit the resulting script my_repository/versions/002_Add_email_column.py:

from sqlalchemy import Table, MetaData, String, Column


def upgrade(migrate_engine):
    meta = MetaData(bind=migrate_engine)
    account = Table('account', meta, autoload=True)
    emailc = Column('email', String(128))
    emailc.create(account)


def downgrade(migrate_engine):
    meta = MetaData(bind=migrate_engine)
    account = Table('account', meta, autoload=True)
    account.c.email.drop()

As we can see in this example we can (and should) use SQLAlchemy’s schema reflection (autoload) mechanism to reference existing schema objects. We could have defined the table objects as they are expected before upgrade or downgrade as well but this would have been more work and is not as convenient.

We can now apply the changeset to sqlite:///project.db:

$ python manage.py upgrade
1 -> 2...
done

and get the following expected result:

$ sqlite3 project.db
sqlite> .schema account
CREATE TABLE account (
   id INTEGER NOT NULL,
   login VARCHAR(40),
   passwd VARCHAR(40), email VARCHAR(128),
   PRIMARY KEY (id)
);

Writing change scripts

As our application evolves, we can create more change scripts using a similar process.

By default, change scripts may do anything any other SQLAlchemy program can do.

SQLAlchemy Migrate extends SQLAlchemy with several operations used to change existing schemas - ie. ALTER TABLE stuff. See changeset documentation for details.

Writing scripts with consistent behavior

Normally, it’s important to write change scripts in a way that’s independent of your application - the same SQL should be generated every time, despite any changes to your app’s source code. You don’t want your change scripts’ behavior changing when your source code does.

Warning

Consider the following example of what NOT to do

Let’s say your application defines a table in the model.py file:

from sqlalchemy import *

meta = MetaData()
table = Table('mytable', meta,
    Column('id', Integer, primary_key=True),
)

… and uses this file to create a table in a change script:

from sqlalchemy import *
from migrate import *
import model

def upgrade(migrate_engine):
    model.meta.bind = migrate_engine

def downgrade(migrate_engine):
    model.meta.bind = migrate_engine
    model.table.drop()

This runs successfully the first time. But what happens if we change the table definition in model.py?

from sqlalchemy import *

meta = MetaData()
table = Table('mytable', meta,
    Column('id', Integer, primary_key=True),
    Column('data', String(42)),
)

We’ll create a new column with a matching change script

from sqlalchemy import *
from migrate import *
import model

def upgrade(migrate_engine):
    model.meta.bind = migrate_engine
    model.table.create()

def downgrade(migrate_engine):
    model.meta.bind = migrate_engine
    model.table.drop()

This appears to run fine when upgrading an existing database - but the first script’s behavior changed! Running all our change scripts on a new database will result in an error - the first script creates the table based on the new definition, with both columns; the second cannot add the column because it already exists.

To avoid the above problem, you should use SQLAlchemy schema reflection as shown above or copy-paste your table definition into each change script rather than importing parts of your application.

Note

Sometimes it is enough to just reflect tables with SQLAlchemy instead of copy-pasting - but remember, explicit is better than implicit!

Writing for a specific database

Sometimes you need to write code for a specific database. Migrate scripts can run under any database, however - the engine you’re given might belong to any database. Use engine.name to get the name of the database you’re working with

>>> from sqlalchemy import *
>>> from migrate import *
>>>
>>> engine = create_engine('sqlite:///:memory:')
>>> engine.name
'sqlite'

Writings .sql scripts

You might prefer to write your change scripts in SQL, as .sql files, rather than as Python scripts. SQLAlchemy-migrate can work with that:

$ python manage.py version
1
$ python manage.py script_sql postgresql

This creates two scripts my_repository/versions/002_postgresql_upgrade.sql and my_repository/versions/002_postgresql_downgrade.sql, one for each operation, or function defined in a Python change script - upgrade and downgrade. Both are specified to run with PostgreSQL databases - we can add more for different databases if we like. Any database defined by SQLAlchemy may be used here - ex. sqlite, postgresql, oracle, mysql…

Command line usage

migrate command is used for API interface. For list of commands and help use:

$ migrate --help

migrate command executes main() function. For ease of usage, generate your own project management script, which calls main function with keywords arguments. You may want to specify url and repository arguments which almost all API functions require.

If api command looks like:

$ migrate downgrade URL REPOSITORY VERSION [--preview_sql|--preview_py]

and you have a project management script that looks like

from migrate.versioning.shell import main

main(url='sqlite://', repository='./project/migrations/')

you have first two slots filed, and command line usage would look like:

# preview Python script
$ migrate downgrade 2 --preview_py

# downgrade to version 2
$ migrate downgrade 2

Changed in version 0.5.4: Command line parsing refactored: positional parameters usage

Whole command line parsing was rewriten from scratch with use of OptionParser. Options passed as kwargs to main() are now parsed correctly. Options are passed to commands in the following priority (starting from highest):

Python API

All commands available from the command line are also available for your Python scripts by importing migrate.versioning.api. See the migrate.versioning.api documentation for a list of functions; function names match equivalent shell commands. You can use this to help integrate SQLAlchemy Migrate with your existing update process.

For example, the following commands are similar:

From the command line:

$ migrate help help
/usr/bin/migrate help COMMAND

    Displays help on a given command.

From Python

import migrate.versioning.api
migrate.versioning.api.help('help')
# Output:
# %prog help COMMAND
#
#     Displays help on a given command.

Experimental commands

Some interesting new features to create SQLAlchemy db models from existing databases and vice versa were developed by Christian Simms during the development of SQLAlchemy-migrate 0.4.5. These features are roughly documented in a thread in migrate-users.

Here are the commands’ descriptions as given by migrate help <command>:

  • compare_model_to_db: Compare the current model (assumed to be a module level variable of type sqlalchemy.MetaData) against the current database.

  • create_model: Dump the current database as a Python model to stdout.

  • make_update_script_for_model: Create a script changing the old Python model to the new (current) Python model, sending to stdout.

As this sections headline says: These features are EXPERIMENTAL. Take the necessary arguments to the commands from the output of migrate help <command>.

Repository configuration

SQLAlchemy-migrate repositories can be configured in their migrate.cfg files. The initial configuration is performed by the migrate create call explained in Create a change repository. The following options are available currently:

  • repository_id Used to identify which repository this database is versioned under. You can use the name of your project.

  • version_table The name of the database table used to track the schema version. This name shouldn’t already be used by your project. If this is changed once a database is under version control, you’ll need to change the table name in each database too.

  • required_dbs When committing a change script, SQLAlchemy-migrate will attempt to generate the sql for all supported databases; normally, if one of them fails - probably because you don’t have that database installed - it is ignored and the commit continues, perhaps ending successfully. Databases in this list MUST compile successfully during a commit, or the entire commit will fail. List the databases your application will actually be using to ensure your updates to that database work properly. This must be a list; example: [‘postgres’, ‘sqlite’]

  • use_timestamp_numbering When creating new change scripts, Migrate will stamp the new script with a version number. By default this is latest_version + 1. You can set this to ‘true’ to tell Migrate to use the UTC timestamp instead.

    New in version 0.7.2.

Customize templates

Users can pass templates_path to API functions to provide customized templates path. Path should be a collection of templates, like migrate.versioning.templates package directory.

One may also want to specify custom themes. API functions accept templates_theme for this purpose (which defaults to default)

Example:

/home/user/templates/manage $ ls
default.py_tmpl
pylons.py_tmpl

/home/user/templates/manage $ migrate manage manage.py --templates_path=/home/user/templates --templates_theme=pylons

New in version 0.6.0.