📦 Ecto Migrator — 技能工具

Generating Migrations

From Natural Language

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

File Naming

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

Name conventions: create_, add__to_

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

# 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

# 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])

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)

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