Ecto Migrator

Generating Migrations

From Natural Language

Parse the user's description and generate a migration file. Common patterns:

User Says	Migration Action
"Create users table with email and name"	create table(:users) with columns
"Add phone to users"	alter table(:users), add :phone
"Make email unique on users"	create unique_index(:users, [:email])
"Add tenant_id to all tables"	Multiple alter table with index
"Rename status to state on orders"	rename table(:orders), :status, to: :state
"Remove the legacy_id column from users"	alter table(:users), remove :legacy_id
"Add a check constraint on orders amount > 0"	create constraint(:orders, ...)

File Naming

mix ecto.gen.migration <name>
# Generates: priv/repo/migrations/YYYYMMDDHHMMSS_<name>.exs

Name conventions: create_<table>, add_<column>_to_<table>, create_<table>_<column>_index, alter_<table>_add_<columns>.

Migration Template

defmodule MyApp.Repo.Migrations.CreateUsers do
  use Ecto.Migration

  def change do
    create table(:users, primary_key: false) do
      add :id, :binary_id, primary_key: true
      add :email, :string, null: false
      add :name, :string, null: false
      add :role, :string, null: false, default: "member"
      add :metadata, :map, default: %{}
      add :tenant_id, :binary_id, null: false

      add :team_id, references(:teams, type: :binary_id, on_delete: :delete_all)

      timestamps(type: :utc_datetime_usec)
    end

    create unique_index(:users, [:tenant_id, :email])
    create index(:users, [:tenant_id])
    create index(:users, [:team_id])
  end
end

Column Types

See references/column-types.md for complete type mapping and guidance.

Key decisions:

• IDs: Use :binary_id (UUID) — set primary_key: false on table, add :id manually.
• Money: Use :integer (cents) or :decimal — never :float.
• Timestamps: Always timestamps(type: :utc_datetime_usec).
• Enums: Use :string with app-level Ecto.Enum — avoid Postgres enums (hard to migrate).
• JSON: Use :map (maps to jsonb).
• Arrays: Use {:array, :string} etc.

Index Strategies

See references/index-patterns.md for detailed index guidance.

When to Add Indexes

Always index:

• Foreign keys (_id columns)
• tenant_id (first column in composite indexes)
• Columns used in WHERE clauses
• Columns used in ORDER BY
• Unique constraints

Index Types

# Standard B-tree
create index(:users, [:tenant_id])

# Unique
create unique_index(:users, [:tenant_id, :email])

# Partial (conditional)
create index(:orders, [:status], where: "status != 'completed'", name: :orders_active_status_idx)

# GIN for JSONB
create index(:events, [:metadata], using: :gin)

# GIN for array columns
create index(:posts, [:tags], using: :gin)

# Composite
create index(:orders, [:tenant_id, :status, :inserted_at])

# Concurrent (no table lock — use in separate migration)
@disable_ddl_transaction true
@disable_migration_lock true

def change do
  create index(:users, [:email], concurrently: true)
end

Constraints

# Check constraint
create constraint(:orders, :amount_must_be_positive, check: "amount > 0")

# Exclusion constraint (requires btree_gist extension)
execute "CREATE EXTENSION IF NOT EXISTS btree_gist", ""
create constraint(:reservations, :no_overlapping_bookings,
  exclude: ~s|gist (room_id WITH =, tstzrange(starts_at, ends_at) WITH &&)|
)

# Unique constraint (same as unique_index for most purposes)
create unique_index(:accounts, [:slug])

References (Foreign Keys)

add :user_id, references(:users, type: :binary_id, on_delete: :delete_all), null: false
add :team_id, references(:teams, type: :binary_id, on_delete: :nilify_all)
add :parent_id, references(:categories, type: :binary_id, on_delete: :nothing)

on_delete	Use When
:delete_all	Child can't exist without parent (memberships, line items)
:nilify_all	Child should survive parent deletion (optional association)
:nothing	Handle in application code (default)
:restrict	Prevent parent deletion if children exist

Multi-Tenant Patterns

Every Table Gets tenant_id

def change do
  create table(:items, primary_key: false) do
    add :id, :binary_id, primary_key: true
    add :name, :string, null: false
    add :tenant_id, :binary_id, null: false
    timestamps(type: :utc_datetime_usec)
  end

  # Always composite index with tenant_id first
  create index(:items, [:tenant_id])
  create unique_index(:items, [:tenant_id, :name])
end

Adding tenant_id to Existing Tables

def change do
  alter table(:items) do
    add :tenant_id, :binary_id
  end

  # Backfill in a separate data migration, then:
  # alter table(:items) do
  #   modify :tenant_id, :binary_id, null: false
  # end
end

Data Migrations

Rule: Never mix schema changes and data changes in the same migration.

Safe Data Migration Pattern

defmodule MyApp.Repo.Migrations.BackfillUserRoles do
  use Ecto.Migration

  # Don't use schema modules — they may change after this migration runs
  def up do
    execute """
    UPDATE users SET role = 'member' WHERE role IS NULL
    """
  end

  def down do
    # Data migrations may not be reversible
    :ok
  end
end

Batched Data Migration (large tables)

def up do
  execute """
  UPDATE users SET role = 'member'
  WHERE id IN (
    SELECT id FROM users WHERE role IS NULL LIMIT 10000
  )
  """

  # For very large tables, use a Task or Oban job instead
end

Reversible vs Irreversible

Reversible (use change)

These are auto-reversible:

• create table ↔ drop table
• add column ↔ remove column
• create index ↔ drop index
• rename ↔ rename

Irreversible (use up/down)

Must define both directions:

• modify column type — Ecto can't infer the old type
• execute raw SQL
• Data backfills
• Dropping columns with data

def up do
  alter table(:users) do
    modify :email, :citext, from: :string  # from: helps reversibility
  end
end

def down do
  alter table(:users) do
    modify :email, :string, from: :citext
  end
end

Using modify with from:

Phoenix 1.7+ supports from: for reversible modify:

def change do
  alter table(:users) do
    modify :email, :citext, null: false, from: {:string, null: true}
  end
end

PostgreSQL Extensions

def change do
  execute "CREATE EXTENSION IF NOT EXISTS citext", "DROP EXTENSION IF EXISTS citext"
  execute "CREATE EXTENSION IF NOT EXISTS pgcrypto", "DROP EXTENSION IF EXISTS pgcrypto"
  execute "CREATE EXTENSION IF NOT EXISTS pg_trgm", "DROP EXTENSION IF EXISTS pg_trgm"
end

Enum Types (PostgreSQL native — use sparingly)

Prefer Ecto.Enum with :string columns. If you must use Postgres enums:

def up do
  execute "CREATE TYPE order_status AS ENUM ('pending', 'confirmed', 'shipped', 'delivered')"

  alter table(:orders) do
    add :status, :order_status, null: false, default: "pending"
  end
end

def down do
  alter table(:orders) do
    remove :status
  end

  execute "DROP TYPE order_status"
end

Warning: Adding values to Postgres enums requires ALTER TYPE ... ADD VALUE which cannot run inside a transaction. Prefer :string + Ecto.Enum.

Checklist

• Primary key: primary_key: false + add :id, :binary_id, primary_key: true
• null: false on required columns
• timestamps(type: :utc_datetime_usec)
• Foreign keys with appropriate on_delete
• Index on every foreign key column
• tenant_id indexed (composite with lookup fields)
• Unique constraints where needed
• Concurrent indexes in separate migration with @disable_ddl_transaction true
• Data migrations in separate files from schema migrations