Migrations

Migrations are a convenient way for you to alter your database in a structured and organised manner. You could edit fragments of SQL by hand but you would then be responsible for telling other developers that they need to go and run it. You'd also have to keep track of which changes need to be run against the production machines next time you deploy. Active Record tracks which migrations have already been run so all you have to do is update your source and run rake db:migrate. Active Record will work out which migrations should be run.

Migrations also allow you to describe these transformations using Ruby. The great thing about this is that (like most of Active Record's functionality) it is database independent: you don't need to worry about the precise syntax of CREATE TABLE any more that you worry about variations on SELECT * (you can drop down to raw SQL for database specific features). For example you could use SQLite3 in development, but MySQL in production.

You'll learn all about migrations including:

  • The generators you can use to create them

  • The methods Active Record provides to manipulate your database

  • The Rake tasks that manipulate them

  • How they relate to schema.rb

1. Anatomy Of A Migration

Before I dive into the details of a migration, here are a few examples of the sorts of things you can do:

class CreateProducts < ActiveRecord::Migration
  def self.up
    create_table :products do |t|
      t.string :name
      t.text :description

      t.timestamps
    end
  end

  def self.down
    drop_table :products
  end
end

This migration adds a table called products with a string column called name and a text column called description. A primary key column called id will also be added, however since this is the default we do not need to ask for this. The timestamp columns created_at and updated_at which Active Record populates automatically will also be added. Reversing this migration is as simple as dropping the table.

Migrations are not limited to changing the schema. You can also use them to fix bad data in the database or populate new fields:

class AddReceiveNewsletterToUsers < ActiveRecord::Migration
  def self.up
    change_table :users do |t|
      t.boolean :receive_newsletter, :default => false
    end
    User.update_all ["receive_newsletter = ?", true]
  end

  def self.down
    remove_column :users, :receive_newsletter
  end
end

This migration adds an receive_newsletter column to the users table. We want it to default to false for new users, but existing users are considered to have already opted in, so we use the User model to set the flag to true for existing users.

Note Some caveats apply to using models in your migrations.

1.1. Migrations are classes

A migration is a subclass of ActiveRecord::Migration that implements two class methods: up (perform the required transformations) and down (revert them).

Active Record provides methods that perform common data definition tasks in a database independent way (you'll read about them in detail later):

  • create_table

  • change_table

  • drop_table

  • add_column

  • remove_column

  • change_column

  • rename_column

  • add_index

  • remove_index

If you need to perform tasks specific to your database (for example create a foreign key constraint) then the execute function allows you to execute arbitrary SQL. A migration is just a regular Ruby class so you're not limited to these functions. For example after adding a column you could write code to set the value of that column for existing records (if necessary using your models).

On databases that support transactions with statements that change the schema (such as PostgreSQL), migrations are wrapped in a transaction. If the database does not support this (for example MySQL and SQLite) then when a migration fails the parts of it that succeeded will not be rolled back. You will have to unpick the changes that were made by hand.

1.2. What's in a name

Migrations are stored in files in db/migrate, one for each migration class. The name of the file is of the form YYYYMMDDHHMMSS_create_products.rb, that is to say a UTC timestamp identifying the migration followed by an underscore followed by the name of the migration. The migration class' name must match (the camelcased version of) the latter part of the file name. For example 20080906120000_create_products.rb should define CreateProducts and 20080906120001_add_details_to_products.rb should define AddDetailsToProducts. If you do feel the need to change the file name then you MUST update the name of the class inside or Rails will complain about a missing class.

Internally Rails only uses the migration's number (the timestamp) to identify them. Prior to Rails 2.1 the migration number started at 1 and was incremented each time a migration was generated. With multiple developers it was easy for these to clash requiring you to rollback migrations and renumber them. With Rails 2.1 this is largely avoided by using the creation time of the migration to identify them. You can revert to the old numbering scheme by setting config.active_record.timestamped_migrations to false in environment.rb.

The combination of timestamps and recording which migrations have been run allows Rails to handle common situations that occur with multiple developers.

For example Alice adds migrations 20080906120000 and 20080906123000 and Bob adds 20080906124500 and runs it. Alice finishes her changes and checks in her migrations and Bob pulls down the latest changes. Rails knows that it has not run Alice's two migrations so rake db:migrate would run them (even though Bob's migration with a later timestamp has been run), and similarly migrating down would not run their down methods.

Of course this is no substitution for communication within the team, for example if Alice's migration removed a table that Bob's migration assumed the existence of then trouble will still occur.

1.3. Changing migrations

Occasionally you will make a mistake while writing a migration. If you have already run the migration then you cannot just edit the migration and run the migration again: Rails thinks it has already run the migration and so will do nothing when you run rake db:migrate. You must rollback the migration (for example with rake db:rollback), edit your migration and then run rake db:migrate to run the corrected version.

In general editing existing migrations is not a good idea: you will be creating extra work for yourself and your co-workers and cause major headaches if the existing version of the migration has already been run on production machines. Instead you should write a new migration that performs the changes you require. Editing a freshly generated migration that has not yet been committed to source control (or more generally which has not been propagated beyond your development machine) is relatively harmless. Just use some common sense.

2. Creating A Migration

2.1. Creating a model

The model and scaffold generators will create migrations appropriate for adding a new model. This migration will already contain instructions for creating the relevant table. If you tell Rails what columns you want then statements for adding those will also be created. For example, running

ruby script/generate model Product name:string description:text will create a migration that looks like this

class CreateProducts < ActiveRecord::Migration
  def self.up
    create_table :products do |t|
      t.string :name
      t.text :description

      t.timestamps
    end
  end

  def self.down
    drop_table :products
  end
end

You can append as many column name/type pairs as you want. By default t.timestamps (which creates the updated_at and created_at columns that are automatically populated by Active Record) will be added for you.

2.2. Creating a standalone migration

If you are creating migrations for other purposes (for example to add a column to an existing table) then you can use the migration generator:

ruby script/generate migration AddPartNumberToProducts

This will create an empty but appropriately named migration:

class AddPartNumberToProducts < ActiveRecord::Migration
  def self.up
  end

  def self.down
  end
end

If the migration name is of the form AddXXXToYYY or RemoveXXXFromY and is followed by a list of column names and types then a migration containing the appropriate add and remove column statements will be created.

ruby script/generate migration AddPartNumberToProducts part_number:string

will generate

class AddPartNumberToProducts < ActiveRecord::Migration
  def self.up
    add_column :products, :part_number, :string
  end

  def self.down
    remove_column :products, :part_number
  end
end

Similarly,

ruby script/generate migration RemovePartNumberFromProducts part_number:string

generates

class RemovePartNumberFromProducts < ActiveRecord::Migration
  def self.up
    remove_column :products, :part_number
  end

  def self.down
    add_column :products, :part_number, :string
  end
end

You are not limited to one magically generated column, for example

ruby script/generate migration AddDetailsToProducts part_number:string price:decimal

generates

class AddDetailsToProducts < ActiveRecord::Migration
  def self.up
    add_column :products, :part_number, :string
    add_column :products, :price, :decimal
  end

  def self.down
    remove_column :products, :price
    remove_column :products, :part_number
  end
end

As always, what has been generated for you is just a starting point. You can add or remove from it as you see fit.

3. Writing a Migration

Once you have created your migration using one of the generators it's time to get to work!

3.1. Creating a table

create_table will be one of your workhorses. A typical use would be

create_table :products do |t|
  t.string :name
end

which creates a products table with a column called name (and as discussed below, an implicit id column).

The object yielded to the block allows you create columns on the table. There are two ways of doing this. The first looks like

create_table :products do |t|
  t.column :name, :string, :null => false
end

the second form, the so called "sexy" migrations, drops the somewhat redundant column method. Instead, the string, integer etc. methods create a column of that type. Subsequent parameters are identical.

create_table :products do |t|
  t.string :name, :null => false
end

By default create_table will create a primary key called id. You can change the name of the primary key with the :primary_key option (don't forget to update the corresponding model) or if you don't want a primary key at all (for example for a HABTM join table) you can pass :id ⇒ false. If you need to pass database specific options you can place an sql fragment in the :options option. For example

create_table :products, :options => "ENGINE=BLACKHOLE" do |t|
  t.string :name, :null => false
end

Will append ENGINE=BLACKHOLE to the sql used to create the table (when using MySQL the default is "ENGINE=InnoDB").

The types Active Record supports are :primary_key, :string, :text, :integer, :float, :decimal, :datetime, :timestamp, :time, :date, :binary, :boolean.

These will be mapped onto an appropriate underlying database type, for example with MySQL :string is mapped to VARCHAR(255). You can create columns of types not supported by Active Record when using the non sexy syntax, for example

create_table :products do |t|
  t.column :name, 'polygon', :null => false
end

This may however hinder portability to other databases.

3.2. Changing tables

create_table's close cousin is change_table. Used for changing existing tables, it is used in a similar fashion to create_table but the object yielded to the block knows more tricks. For example

change_table :products do |t|
  t.remove :description, :name
  t.string :part_number
  t.index :part_number
  t.rename :upccode, :upc_code
end

removes the description column, creates a part_number column and adds an index on it. Finally it renames the upccode column. This is the same as doing

remove_column :products, :description
remove_column :products, :name
add_column :products, :part_number, :string
add_index :products, :part_number
rename_column :products, :upccode, :upc_code

You don't have to keep repeating the table name and it groups all the statements related to modifying one particular table. The individual transformation names are also shorter, for example remove_column becomes just remove and add_index becomes just index.

3.3. Special helpers

Active Record provides some shortcuts for common functionality. It is for example very common to add both the created_at and updated_at columns and so there is a method that does exactly that:

create_table :products do |t|
  t.timestamps
end

will create a new products table with those two columns whereas

change_table :products do |t|
  t.timestamps
end

adds those columns to an existing table.

The other helper is called references (also available as belongs_to). In its simplest form it just adds some readability

create_table :products do |t|
  t.references :category
end

will create a category_id column of the appropriate type. Note that you pass the model name, not the column name. Active Record adds the _id for you. If you have polymorphic belongs_to associations then references will add both of the columns required:

create_table :products do |t|
  t.references :attachment, :polymorphic => {:default => 'Photo'}
end

will add an attachment_id column and a string attachment_type column with a default value of Photo.

Note The references helper does not actually create foreign key constraints for you. You will need to use execute for that or a plugin that adds foreign key support.

If the helpers provided by Active Record aren't enough you can use the execute function to execute arbitrary SQL.

For more details and examples of individual methods check the API documentation, in particular the documentation for ActiveRecord::ConnectionAdapters::SchemaStatements (which provides the methods available in the up and down methods), ActiveRecord::ConnectionAdapters::TableDefinition (which provides the methods available on the object yielded by create_table) and ActiveRecord::ConnectionAdapters::Table (which provides the methods available on the object yielded by change_table).

3.4. Writing your down method

The down method of your migration should revert the transformations done by the up method. In other words the database should be unchanged if you do an up followed by a down. For example if you create a table in the up you should drop it in the down method. It is wise to do things in precisely the reverse order to in the up method. For example

class ExampleMigration < ActiveRecord::Migration

  def self.up
    create_table :products do |t|
      t.references :category
    end
    #add a foreign key
    execute "ALTER TABLE products ADD CONSTRAINT fk_products_categories FOREIGN KEY (category_id) REFERENCES categories(id)"

    add_column :users, :home_page_url, :string

    rename_column :users, :email, :email_address
  end

  def self.down
    rename_column :users, :email_address, :email
    remove_column :users, :home_page_url
    execute "ALTER TABLE products DROP FOREIGN KEY fk_products_categories"
    drop_table :products
  end
end

Sometimes your migration will do something which is just plain irreversible, for example it might destroy some data. In cases like those when you can't reverse the migration you can raise IrreversibleMigration from your down method. If someone tries to revert your migration an error message will be displayed saying that it can't be done.

4. Running Migrations

Rails provides a set of rake tasks to work with migrations which boils down to running certain sets of migrations. The very first migration related rake task you use will probably be db:migrate. In its most basic form it just runs the up method for all the migrations that have not yet been run. If there are no such migrations it exits.

If you specify a target version, Active Record will run the required migrations (up or down) until it has reached the specified version. The version is the numerical prefix on the migration's filename. For example to migrate to version 20080906120000 run

rake db:migrate VERSION=20080906120000

If this is greater than the current version (i.e. it is migrating upwards) this will run the up method on all migrations up to and including 20080906120000, if migrating downwards this will run the down method on all the migrations down to, but not including, 20080906120000.

4.1. Rolling back

A common task is to rollback the last migration, for example if you made a mistake in it and wish to correct it. Rather than tracking down the version number associated with the previous migration you can run

rake db:rollback

This will run the down method from the latest migration. If you need to undo several migrations you can provide a STEP parameter:

rake db:rollback STEP=3

will run the down method fron the last 3 migrations.

The db:migrate:redo task is a shortcut for doing a rollback and then migrating back up again. As with the db:rollback task you can use the STEP parameter if you need to go more than one version back, for example

rake db:migrate:redo STEP=3

Neither of these Rake tasks do anything you could not do with db:migrate, they are simply more convenient since you do not need to explicitly specify the version to migrate to.

Lastly, the db:reset task will drop the database, recreate it and load the current schema into it.

Note This is not the same as running all the migrations - see the section on schema.rb.

4.2. Being Specific

If you need to run a specific migration up or down the db:migrate:up and db:migrate:down tasks will do that. Just specify the appropriate version and the corresponding migration will have its up or down method invoked, for example

rake db:migrate:up VERSION=20080906120000

will run the up method from the 20080906120000 migration. These tasks check whether the migration has already run, so for example db:migrate:up VERSION=20080906120000 will do nothing if Active Record believes that 20080906120000 has already been run.

4.3. Being talkative

By default migrations tell you exactly what they're doing and how long it took. A migration creating a table and adding an index might produce output like this

== 20080906170109 CreateProducts: migrating ===================================
-- create_table(:products)
   -> 0.0021s
-- add_index(:products, :name)
   -> 0.0026s
== 20080906170109 CreateProducts: migrated (0.0059s) ==========================

Several methods are provided that allow you to control all this:

  • suppress_messages suppresses any output generated by its block

  • say outputs text (the second argument controls whether it is indented or not)

  • say_with_time outputs text along with how long it took to run its block. If the block returns an integer it assumes it is the number of rows affected.

For example, this migration

class CreateProducts < ActiveRecord::Migration
  def self.up
    suppress_messages do
      create_table :products do |t|
        t.string :name
        t.text :description
        t.timestamps
      end
    end
    say "Created a table"
    suppress_messages {add_index :products, :name}
    say "and an index!", true
    say_with_time 'Waiting for a while' do
      sleep 10
      250
    end
  end

  def self.down
    drop_table :products
  end
end

generates the following output

== 20080906170109 CreateProducts: migrating ===================================
-- Created a table
   -> and an index!
-- Waiting for a while
   -> 10.0001s
   -> 250 rows
== 20080906170109 CreateProducts: migrated (10.0097s) =========================

If you just want Active Record to shut up then running rake db:migrate VERBOSE=false will suppress any output.

5. Using Models In Your Migrations

When creating or updating data in a migration it is often tempting to use one of your models. After all they exist to provide easy access to the underlying data. This can be done but some caution should be observed.

Consider for example a migration that uses the Product model to update a row in the corresponding table. Alice later updates the Product model, adding a new column and a validation on it. Bob comes back from holiday, updates the source and runs outstanding migrations with rake db:migrate, including the one that used the Product model. When the migration runs the source is up to date and so the Product model has the validation added by Alice. The database however is still old and so does not have that column and an error ensues because that validation is on a column that does not yet exist.

Frequently I just want to update rows in the database without writing out the SQL by hand: I'm not using anything specific to the model. One pattern for this is to define a copy of the model inside the migration itself, for example:

class AddPartNumberToProducts < ActiveRecord::Migration
  class Product < ActiveRecord::Base
  end

  def self.up
    ...
  end

  def self.down
    ...
  end
end

The migration has its own minimal copy of the Product model and no longer cares about the Product model defined in the application.

5.1. Dealing with changing models

For performance reasons information about the columns a model has is cached. For example if you add a column to a table and then try and use the corresponding model to insert a new row it may try and use the old column information. You can force Active Record to re-read the column information with the reset_column_information method, for example

class AddPartNumberToProducts < ActiveRecord::Migration
  class Product < ActiveRecord::Base
  end

  def self.up
    add_column :product, :part_number, :string
    Product.reset_column_information
    ...
  end

  def self.down
    ...
  end
end

6. Schema dumping and you

6.1. What are schema files for?

Migrations, mighty as they may be, are not the authoritative source for your database schema. That role falls to either schema.rb or an SQL file which Active Record generates by examining the database. They are not designed to be edited, they just represent the current state of the database.

There is no need (and it is error prone) to deploy a new instance of an app by replaying the entire migration history. It is much simpler and faster to just load into the database a description of the current schema.

For example, this is how the test database is created: the current development database is dumped (either to schema.rb or development.sql) and then loaded into the test database.

Schema files are also useful if want a quick look at what attributes an Active Record object has. This information is not in the model's code and is frequently spread across several migrations but is all summed up in the schema file. The annotate_models plugin, which automatically adds (and updates) comments at the top of each model summarising the schema, may also be of interest.

6.2. Types of schema dumps

There are two ways to dump the schema. This is set in config/environment.rb by the config.active_record.schema_format setting, which may be either :sql or :ruby.

If :ruby is selected then the schema is stored in db/schema.rb. If you look at this file you'll find that it looks an awful lot like one very big migration:

ActiveRecord::Schema.define(:version => 20080906171750) do
  create_table "authors", :force => true do |t|
    t.string   "name"
    t.datetime "created_at"
    t.datetime "updated_at"
  end

  create_table "products", :force => true do |t|
    t.string   "name"
    t.text     "description"
    t.datetime "created_at"
    t.datetime "updated_at"
    t.string   "part_number"
  end
end

In many ways this is exactly what it is. This file is created by inspecting the database and expressing its structure using create_table, add_index and so on. Because this is database independent it could be loaded into any database that Active Record supports. This could be very useful if you were to distribute an application that is able to run against multiple databases.

There is however a trade-off: schema.rb cannot express database specific items such as foreign key constraints, triggers or stored procedures. While in a migration you can execute custom SQL statements, the schema dumper cannot reconstitute those statements from the database. If you are using features like this then you should set the schema format to :sql.

Instead of using Active Record's schema dumper the database's structure will be dumped using a tool specific to that database (via the db:structure:dump Rake task) into db/#{RAILS_ENV}_structure.sql. For example for PostgreSQL the pg_dump utility is used and for MySQL this file will contain the output of SHOW CREATE TABLE for the various tables. Loading this schema is simply a question of executing the SQL statements contained inside.

By definition this will be a perfect copy of the database's structure but this will usually prevent loading the schema into a database other than the one used to create it.

6.3. Schema dumps and source control

Because they are the authoritative source for your database schema, it is strongly recommended that you check them into source control.

7. Active Record and Referential Integrity

The Active Record way is that intelligence belongs in your models, not in the database. As such features such as triggers or foreign key constraints, which push some of that intelligence back into the database are not heavily used.

Validations such as validates_uniqueness_of are one way in which models can enforce data integrity. The :dependent option on associations allows models to automatically destroy child objects when the parent is destroyed. Like anything which operates at the application level these cannot guarantee referential integrity and so some people augment them with foreign key constraints.

Although Active Record does not provide any tools for working directly with such features, the execute method can be used to execute arbitrary SQL. There are also a number of plugins such as redhillonrails which add foreign key support to Active Record (including support for dumping foreign keys in schema.rb).

8. Changelog

Lighthouse ticket