Ecto SQL v3.2.0 Ecto.Migration behaviour View Source

Migrations are used to modify your database schema over time.

This module provides many helpers for migrating the database, allowing developers to use Elixir to alter their storage in a way that is database independent.

Migrations typically provide two operations: up and down, allowing us to migrate the database forward or roll it back in case of errors.

In order to manage migrations, Ecto creates a table called schema_migrations in the database, which stores all migrations that have already been executed. You can configure the name of this table with the :migration_source configuration option.

Ecto also locks the schema_migrations table when running migrations, guaranteeing two different servers cannot run the same migration at the same time.

Creating your first migration

Migrations are defined inside the "priv/REPO/migrations" where REPO is the last part of the repository name in underscore. For example, migrations for MyApp.Repo would be found in "priv/repo/migrations". For MyApp.CustomRepo, it would be found in "priv/custom_repo/migrations".

Each file in the migratitons directory have the following structure:

NUMBER_NAME.exs

The NUMBER is a unique number that identifies the migration. It is usually the timestamp of when the migration was created. The NAME must also be unique and it quickly identifies what the migration does. For example, if you need to track the "weather" in your system, you can start a new file at "priv/repo/migraitons/2019041714000_add_weather_table.exs" that will have the following contents:

defmodule MyRepo.Migrations.AddWeatherTable do
  use Ecto.Migration

  def up do
    create table("weather") do
      add :city,    :string, size: 40
      add :temp_lo, :integer
      add :temp_hi, :integer
      add :prcp,    :float

      timestamps()
    end
  end

  def down do
    drop table("weather")
  end
end

The up/0 function is responsible to migrate your database forward. the down/0 function is executed whenever you want to rollback. The down/0 function must always do the opposite of up/0. Inside those functions, we invoke the API defined in this module, you will find conveniences for managing tables, indexes, columns, references, as well as running custom SQL commands.

To run a migration, we generally use Mix tasks. For example, you can run the migration above by going to the root of your project and typing:

$ mix ecto.migrate

You can also it rollback by calling

$ mix ecto.rollback --step 1

Note rollback requires us to say how much we want to rollback. On the other hand, mix ecto.migrate will always run all pending migrations.

In practice, we don't create migration files by hand either, we typically use mix ecto.gen.migration to generate the file with the proper timestamp and then we just fill in its contents:

$ mix ecto.gen.migration add_weather_table

Mix tasks

As seen above, Ecto provides many Mix tasks to help developers work with migrations. We summarize them below:

Run mix help COMMAND for more information on a particular command. For a lower level API for running migrations, see Ecto.Migrator.

Change

Having to write both up/0 and down/0 functions for every migration is tedious and error prone. For this reason, Ecto allows you to defined a change/0 callback with all of the code you want to execute when migrating and Ecto will automatically figure out the down/0 for you. For example, the migration above can be written as:

defmodule MyRepo.Migrations.AddWeatherTable do
  use Ecto.Migration

  def change do
    create table("weather") do
      add :city,    :string, size: 40
      add :temp_lo, :integer
      add :temp_hi, :integer
      add :prcp,    :float

      timestamps()
    end
  end
end

However, note that not all commands are reversible. Trying to rollback a non-reversible command will raise an Ecto.MigrationError.

A notable command in this regard is execute/2, which is reversible in change/0 by accepting a pair of plain SQL strings. The first is run on forward migrations (up/0) and the second when rolling back (down/0).

If up/0 and down/0 are implemented in a migration, they take precedence, and change/0 isn't invoked.

Field Types

The Ecto primitive types are mapped to the appropriate database type by the various database adapters. For example, :string is converted to :varchar, :binary to :bits or :blob, and so on.

Similarly, you can pass any field type supported by your database as long as it maps to an Ecto type. For instance, you can use :text, :varchar, or :char in your migrations as add :field_name, :text. In your Ecto schema, they will all map to the same :string type.

Remember, atoms can contain arbitrary characters by enclosing in double quotes the characters following the colon. So, if you want to use a field type with database-specific options, you can pass atoms containing these options like :"int unsigned", :"time without time zone", etc.

Flushing

Instructions inside migrations are not executed immedidately. Instead they are performed once the relevant up, change, or down callback terminates.

However, under certain situations, you may want to guarantee all of the previous steps have been executed before continuing. This is generally useful in case you need to apply some changes to the table before continuing with the migration. This can be done with flush/0:

def up do
  ...
  flush()
  ...
end

Comments

Migrations where you create or alter a table support specifying table and column comments. The same can be done when creating constraints and indexes. Not all databases support this feature.

def up do
  create index("posts", [:name], comment: "Index Comment")
  create constraint("products", "price_must_be_positive", check: "price > 0", comment: "Constraint Comment")
  create table("weather", prefix: "north_america", comment: "Table Comment") do
    add :city, :string, size: 40, comment: "Column Comment"
    timestamps()
  end
end

Repo configuration

The following migration configuration options are available for a given repository:

  • :migration_source - Version numbers of migrations will be saved in a table named schema_migrations by default. You can configure the name of the table via:

    config :app, App.Repo, migration_source: "my_migrations"
  • :migration_primary_key - By default, Ecto uses the :id column with type :bigserial, but you can configure it via:

    config :app, App.Repo, migration_primary_key: [name: :uuid, type: :binary_id]
  • :migration_timestamps - By default, Ecto uses the :naive_datetime type, but you can configure it via:

    config :app, App.Repo, migration_timestamps: [type: :utc_datetime]
  • :migration_lock - By default, Ecto will lock the migration table. This allows multiple nodes to attempt to run migrations at the same time but only one will succeed. You can disable the migration_lock by setting it to nil:

    config :app, App.Repo, migration_lock: nil
  • :migration_default_prefix - Ecto defaults to nil for the database prefix for migrations, but you can configure it via:

    config :app, App.Repo, migration_default_prefix: "my_prefix"
  • :start_apps_before_migration - A list of applications to be started before running migrations. Used by Ecto.Migrator.with_repo/3 and the migration tasks:

    config :app, App.Repo, start_apps_before_migration: [:ssl, :some_custom_logger]

Prefixes

Migrations support specifying a table prefix or index prefix which will target either a schema (if using PostgreSQL) or a different database (if using MySQL). If no prefix is provided, the default schema or database is used.

Any reference declared in the table migration refers by default to the table with the same declared prefix. The prefix is specified in the table options:

def up do
  create table("weather", prefix: "north_america") do
    add :city,    :string, size: 40
    add :temp_lo, :integer
    add :temp_hi, :integer
    add :prcp,    :float
    add :group_id, references(:groups)

    timestamps()
  end

  create index("weather", [:city], prefix: "north_america")
end

Note: if using MySQL with a prefixed table, you must use the same prefix for the references since cross-database references are not supported.

When using a prefixed table with either MySQL or PostgreSQL, you must use the same prefix for the index field to ensure that you index the prefix-qualified table.

Transaction Callbacks

If possible, each migration runs inside a transaction. This is true for Postgres, but not true for MySQL, as the later does not support DDL transactions.

In some rare cases, you may need to execute some common behavior after beginning a migration transaction, or before commiting that transaction. For instance, one might desire to set a lock_timeout for each lock in the migration transaction.

You can do so by defining after_begin/0 and before_commit/0 callbacks to your migration.

However, if you need do so for every migration module, implemment this callback for every migration can be quite repetitive. Luckily, you can handle this by providing your migration module:

defmodule MyApp.Migration do
  defmacro __using__(_) do
    quote do
      use Ecto.Migration

      def after_begin() do
        repo().query! "SET lock_timeout TO '5s'", "SET lock_timeout TO '10s'"
      end
    end
  end
end

Then in your migrations you can use MyApp.Migration to share this behavior among all your migrations.

Link to this section Summary

Functions

Adds a column when creating or altering a table.

Adds a column if it not exists yet when altering a table.

Alters a table.

Defines a constraint (either a check constraint or an exclusion constraint) to be evaluated by the database when a row is inserted or updated.

Creates one of the following

Creates a table.

Creates an index or a table with only :id field if one does not yet exist.

Creates a table if it does not exist.

Gets the migrator direction.

Drops one of the following

Drops a table or index if it exists.

Executes arbitrary SQL or a keyword command.

Executes reversible SQL commands.

Executes queue migration commands.

Generates a fragment to be used as a default value.

Returns an index struct that can be given to create/1, drop/1, etc.

Modifies the type of a column when altering a table.

Gets the migrator prefix.

Defines a foreign key.

Removes a column when altering a table.

Removes a column in a reversible way when altering a table.

Removes a column only if the column exists when altering the constraint if the reference type is passed once it only has the constraint name on reference structure.

Renames a table.

Renames a column outside of the alter statement.

Gets the migrator repo.

Returns a table struct that can be given to create/2, alter/2, drop/1, etc.

Adds :inserted_at and :updated_at timestamp columns.

Shortcut for creating a unique index.

Callbacks

Migration code to run immediately after the transaction is opened.

Migration code to run immediately before the transaction is closed.

Link to this section Functions

Link to this function

add(column, type, opts \\ []) View Source

Adds a column when creating or altering a table.

This function also accepts Ecto primitive types as column types that are normalized by the database adapter. For example, :string is converted to :varchar, :binary to :bits or :blob, and so on.

However, the column type is not always the same as the type used in your schema. For example, a schema that has a :string field can be supported by columns of type :char, :varchar, :text, and others. For this reason, this function also accepts :text and other type annotations that are native to the database. These are passed to the database as-is.

To sum up, the column type may be either an Ecto primitive type, which is normalized in cases where the database does not understand it, such as :string or :binary, or a database type which is passed as-is. Custom Ecto types like Ecto.UUID are not supported because they are application-level concerns and may not always map to the database.

Examples

create table("posts") do
  add :title, :string, default: "Untitled"
end

alter table("posts") do
  add :summary, :text # Database type
  add :object,  :map  # Elixir type which is handled by the database
end

Options

  • :primary_key - when true, marks this field as the primary key. If multiple fields are marked, a composite primary key will be created.
  • :default - the column's default value. It can be a string, number, empty list, list of strings, list of numbers, or a fragment generated by fragment/1.
  • :null - when false, the column does not allow null values.
  • :size - the size of the type (for example, the number of characters). The default is no size, except for :string, which defaults to 255.
  • :precision - the precision for a numeric type. Required when :scale is specified.
  • :scale - the scale of a numeric type. Defaults to 0.
Link to this function

add_if_not_exists(column, type, opts \\ []) View Source

Adds a column if it not exists yet when altering a table.

If the type value is a %Reference{}, it is used to add a constraint.

type and opts are exactly the same as in add/3.

This command is not reversible as Ecto does not know about column existence before the creation attempt.

Examples

alter table("posts") do
  add_if_not_exists :title, :string, default: ""
end
Link to this macro

alter(object, list) View Source (macro)

Alters a table.

Examples

alter table("posts") do
  add :summary, :text
  modify :title, :text
  remove :views
end
Link to this function

constraint(table, name, opts \\ []) View Source

Defines a constraint (either a check constraint or an exclusion constraint) to be evaluated by the database when a row is inserted or updated.

Examples

create constraint("users", :price_must_be_positive, check: "price > 0")
create constraint("size_ranges", :no_overlap, exclude: ~s|gist (int4range("from", "to", '[]') WITH &&)|)
drop   constraint("products", "price_must_be_positive")

Options

  • :check - A check constraint expression. Required when creating a check constraint.
  • :exclude - An exclusion constraint expression. Required when creating an exclusion constraint.
  • :prefix - The prefix for the table.

Creates one of the following:

  • an index
  • a table with only an :id field
  • a constraint

When reversing (in a change/0 running backwards), indexes are only dropped if they exist, and no errors are raised. To enforce dropping an index, use drop/1.

Examples

create index("posts", [:name])
create table("version")
create constraint("products", "price_must_be_positive", check: "price > 0")
Link to this macro

create(object, list) View Source (macro)

Creates a table.

By default, the table will also include an :id primary key field that has a type of :bigserial. Check the table/2 docs for more information.

Examples

create table(:posts) do
  add :title, :string, default: "Untitled"
  add :body,  :text

  timestamps()
end
Link to this function

create_if_not_exists(index) View Source

Creates an index or a table with only :id field if one does not yet exist.

Examples

create_if_not_exists index("posts", [:name])

create_if_not_exists table("version")
Link to this macro

create_if_not_exists(object, list) View Source (macro)

Creates a table if it does not exist.

Works just like create/2 but does not raise an error when the table already exists.

Link to this function

direction() View Source
direction() :: :up | :down

Gets the migrator direction.

Link to this function

drop(index_or_table_or_constraint) View Source

Drops one of the following:

  • an index
  • a table
  • a constraint

Examples

drop index("posts", [:name])
drop table("posts")
drop constraint("products", "price_must_be_positive")
Link to this function

drop_if_exists(index_or_table) View Source

Drops a table or index if it exists.

Does not raise an error if the specified table or index does not exist.

Examples

drop_if_exists index("posts", [:name])
drop_if_exists table("posts")

Executes arbitrary SQL or a keyword command.

Reversible commands can be defined by calling execute/2.

Examples

execute "CREATE EXTENSION postgres_fdw"

execute create: "posts", capped: true, size: 1024

Executes reversible SQL commands.

This is useful for database-specific functionality that does not warrant special support in Ecto, for example, creating and dropping a PostgreSQL extension. The execute/2 form avoids having to define separate up/0 and down/0 blocks that each contain an execute/1 expression.

Examples

execute "CREATE EXTENSION postgres_fdw", "DROP EXTENSION postgres_fdw"

Executes queue migration commands.

Reverses the order in which commands are executed when doing a rollback on a change/0 function and resets the commands queue.

Generates a fragment to be used as a default value.

Examples

create table("posts") do
  add :inserted_at, :naive_datetime, default: fragment("now()")
end
Link to this function

index(table, columns, opts \\ []) View Source

Returns an index struct that can be given to create/1, drop/1, etc.

Expects the table name as the first argument and the index field(s) as the second. The fields can be atoms, representing columns, or strings, representing expressions that are sent as-is to the database.

Options

  • :name - the name of the index. Defaults to "#{table}_#{column}_index".
  • :unique - indicates whether the index should be unique. Defaults to false.
  • :concurrently - indicates whether the index should be created/dropped concurrently.
  • :using - configures the index type.
  • :prefix - specify an optional prefix for the index.
  • :where - specify conditions for a partial index.

Adding/dropping indexes concurrently

PostgreSQL supports adding/dropping indexes concurrently (see the docs). However, this feature does not work well with the transactions used by Ecto to guarantee integrity during migrations.

Therefore, to migrate indexes concurrently, you need to set both @disable_ddl_transaction and @disable_migration_lock to true:

defmodule MyRepo.Migrations.CreateIndexes do
  use Ecto.Migration
  @disable_ddl_transaction true
  @disable_migration_lock true

  def change do
    create index("posts", [:slug], concurrently: true)
  end
end

Disabling DDL transactions removes the guarantee that all of the changes in the migration will happen at once. Disabling the migration lock removes the guarantee only a single node will run a given migration if multiple nodes are attempting to migrate at the same time.

Since running migrations outside a transaction and without locks can be dangerous, consider performing very few operations in migrations that add concurrent indexes. We recommend to run migrations with concurrent indexes in isolation and disable those features only temporarily.

Index types

When creating an index, the index type can be specified with the :using option. The :using option can be an atom or a string, and its value is passed to the generated USING clause as-is.

For example, PostgreSQL supports several index types like B-tree (the default), Hash, GIN, and GiST. More information on index types can be found in the PostgreSQL docs.

Partial indexes

Databases like PostgreSQL and MSSQL support partial indexes.

A partial index is an index built over a subset of a table. The subset is defined by a conditional expression using the :where option. The :where option can be an atom or a string; its value is passed to the generated WHERE clause as-is.

More information on partial indexes can be found in the PostgreSQL docs.

Examples

# With no name provided, the name of the below index defaults to
# products_category_id_sku_index
create index("products", [:category_id, :sku], unique: true)

# The name can also be set explicitly
drop index("products", [:category_id, :sku], name: :my_special_name)

# Indexes can be added concurrently
create index("products", [:category_id, :sku], concurrently: true)

# The index type can be specified
create index("products", [:name], using: :hash)

# Partial indexes are created by specifying a :where option
create index("products", [:user_id], where: "price = 0", name: :free_products_index)

Indexes also support custom expressions. Some databases may require the index expression to be written between parentheses:

# Create an index on a custom expression
create index("products", ["(lower(name))"], name: :products_lower_name_index)

# Create a tsvector GIN index on PostgreSQL
create index("products", ["(to_tsvector('english', name))"],
             name: :products_name_vector, using: "GIN")
Link to this function

modify(column, type, opts \\ []) View Source

Modifies the type of a column when altering a table.

This command is not reversible unless the :from option is provided. If the :from value is a %Reference{}, the adapter will try to drop the corresponding foreign key constraints before modifying the type.

See add/3 for more information on supported types.

Examples

alter table("posts") do
  modify :title, :text
end

Options

  • :null - determines whether the column accepts null values.
  • :default - changes the default value of the column.
  • :from - specifies the current type of the column.
  • :size - specifies the size of the type (for example, the number of characters). The default is no size.
  • :precision - the precision for a numeric type. Required when :scale is specified.
  • :scale - the scale of a numeric type. Defaults to 0.

Gets the migrator prefix.

Link to this function

references(table, opts \\ []) View Source

Defines a foreign key.

Examples

create table("products") do
  add :group_id, references("groups")
end

Options

  • :name - The name of the underlying reference, which defaults to "#{table}_#{column}_fkey".
  • :column - The foreign key column name, which defaults to :id.
  • :prefix - The prefix for the reference. Defaults to the reference of the table if present, or nil.
  • :type - The foreign key type, which defaults to :bigserial.
  • :on_delete - What to do if the referenced entry is deleted. May be :nothing (default), :delete_all, :nilify_all, or :restrict.
  • :on_update - What to do if the referenced entry is updated. May be :nothing (default), :update_all, :nilify_all, or :restrict.

Removes a column when altering a table.

This command is not reversible as Ecto does not know what type it should add the column back as. See remove/3 as a reversible alternative.

Examples

alter table("posts") do
  remove :title
end
Link to this function

remove(column, type, opts \\ []) View Source

Removes a column in a reversible way when altering a table.

type and opts are exactly the same as in add/3, and they are used when the command is reversed.

If the type value is a %Reference{}, it is used to remove the constraint.

Examples

alter table("posts") do
  remove :title, :string, default: ""
end
Link to this function

remove_if_exists(column, type) View Source

Removes a column only if the column exists when altering the constraint if the reference type is passed once it only has the constraint name on reference structure.

This command is not reversible as Ecto does not know about column existence before the removal attempt.

Examples

alter table("posts") do
  remove_if_exists :title, :string
end
Link to this function

rename(table_current, list) View Source

Renames a table.

Examples

rename table("posts"), to: table("new_posts")
Link to this function

rename(table, current_column, list) View Source

Renames a column outside of the alter statement.

Examples

rename table("posts"), :title, to: :summary

Gets the migrator repo.

Returns a table struct that can be given to create/2, alter/2, drop/1, etc.

Examples

create table("products") do
  add :name, :string
  add :price, :decimal
end

drop table("products")

create table("products", primary_key: false) do
  add :name, :string
  add :price, :decimal
end

Options

  • :primary_key - when false, a primary key field is not generated on table creation.
  • :engine - customizes the table storage for supported databases. For MySQL, the default is InnoDB.
  • :prefix - the prefix for the table.
  • :options - provide custom options that will be appended after the generated statement. For example, "WITH", "INHERITS", or "ON COMMIT" clauses.

Adds :inserted_at and :updated_at timestamp columns.

Those columns are of :naive_datetime type and by default cannot be null. A list of opts can be given to customize the generated fields.

Options

  • :inserted_at - the name of the column for storing insertion times. Setting it to false disables the column.
  • :updated_at - the name of the column for storing last-updated-at times. Setting it to false disables the column.
  • :type - the type of the :inserted_at and :updated_at columns. Defaults to :naive_datetime.
Link to this function

unique_index(table, columns, opts \\ []) View Source

Shortcut for creating a unique index.

See index/3 for more information.

Link to this section Callbacks

Link to this callback

after_begin() View Source (optional)
after_begin() :: term()

Migration code to run immediately after the transaction is opened.

Keep in mind that it is treated like any normal migration code, and should consider both the up and down cases of the migration.

Link to this callback

before_commit() View Source (optional)
before_commit() :: term()

Migration code to run immediately before the transaction is closed.

Keep in mind that it is treated like any normal migration code, and should consider both the up and down cases of the migration.