migration column methods

All the methods described in columns methods still applies in migrations, to add or change columns with specific types.

This document describes common methods like default, nullable, primaryKey that have effect in both application code and migration, in methods like check, comment, collate that only have effect in migrations.

default

Set a default value for a column on a database level. Value can be a raw SQL.

default can accept a callback when used in ORM table, but it's not applicable in migrations.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    active: t.boolean().default(false),
    date: t.date().default(t.sql`now()`),
  }));
});

If you provide a function to the default, it will be called by ORM before creating records, and it won't have any default value on a database level.

import { change } from '../dbScript';
import { uuidv7 } from 'uuidv7';

change(async (db) => {
  await db.createTable('table', (t) => ({
    // uuidv7 is a function, it is ignored in migrations,
    // column won't have a `DEFAULT` on a database level:
    id: t.uuid().primaryKey().default(uuidv7),
  }));
});

uuid().primaryKey() has a default gen_random_uuid() by default, and if you'd like to drop it use default(null):

id: t.uuid().primaryKey().default(null),

nullable

By default, NOT NULL is added to every column. Use nullable to prevent this:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    name: t.text().nullable(),
  }));
});

enum

In the migration enum takes a single argument for enum name, unlike the enum column in the ORM.

To create a new enum type, use createEnum before creating a table.

import { change } from '../dbScript';

change(async (db) => {
  await db.createEnum('mood', ['sad', 'ok', 'happy']);

  await db.createTable('table', (t) => ({
    mood: t.enum('mood'),
  }));
});

generated column

Define a generated column. generated accepts a raw SQL.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    two: t.integer().generated`1 + 1`,
  }));
});

For tsvector column type, it can also accept language (optional) and columns:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('post', (t) => ({
    id: t.id(),
    title: t.text(),
    body: t.text(),
    // join title and body into a single ts_vector
    generatedTsVector: t.tsvector().generated(['title', 'body']).searchIndex(),
    // with language:
    spanishTsVector: t
      .tsvector()
      .generated('spanish', ['title', 'body'])
      .searchIndex(),
  }));
});

primaryKey

Mark the column as a primary key. This column type becomes an argument of the .find method. So if the primary key is of integer type, .find will accept the number, or if the primary key is of uuid type, .find will expect a string.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    id: t.identity().primaryKey(),
    // optionally, specify a database-level constraint name:
    id: t.identity().primaryKey('primary_key_name'),
  }));
});

composite primary key

Specify primaryKey on multiple columns to have a composite primary key. .find works only with single primary key.

Composite key is useful when defining a join table which is designed to connect other tables.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    id: t.identity().primaryKey(),
    name: t.text().primaryKey(),
    active: t.boolean().primaryKey(),
  }));
});

Alternatively, use t.primaryKey([column1, column2, ...columns]) to specify the primary key consisting of multiple columns.

By default, Postgres will name an underlying constraint as ${table name}_pkey. You can pass the second argument for a custom name.

In createTable, pass the composite primary key into a second function. Unlike changeTable that takes only a single function for all.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.integer(),
      name: t.text(),
      active: t.boolean(),
    }),
    () => t.primaryKey(['id', 'name', 'active'], { name: 'tablePkeyName' }),
  );

  await db.changeTable(
    'otherTable',
    (t) => ({
      newColumn: t.add(t.integer()),
      ...t.change(t.primaryKey(['id'], ['id', 'newColumn']),
    }),
  );
});

foreignKey

Defines a reference between different tables to enforce data integrity.

In snakeCase mode, columns of both tables are translated to a snake_case.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    otherId: t.integer().foreignKey('otherTableName', 'columnName'),
  }));
});

In the migration it's different from OrchidORM table code where a callback with a table is expected:

export class SomeTable extends BaseTable {
  readonly table = 'someTable';
  columns = this.setColumns((t) => ({
    otherTableId: t.integer().foreignKey(() => OtherTable, 'id'),
  }));
}

Optionally you can pass the third argument to foreignKey with options:

type ForeignKeyOptions = {
  // name of the constraint
  name?: string;
  // see database docs for MATCH in FOREIGN KEY
  match?: 'FULL' | 'PARTIAL' | 'SIMPLE';

  onUpdate?: 'NO ACTION' | 'RESTRICT' | 'CASCADE' | 'SET NULL' | 'SET DEFAULT';
  onDelete?: 'NO ACTION' | 'RESTRICT' | 'CASCADE' | 'SET NULL' | 'SET DEFAULT';
};

composite foreign key

Set foreign key from multiple columns in the current table to corresponding columns in the other table.

The first argument is an array of columns in the current table, the second argument is another table name, the third argument is an array of columns in another table, and the fourth argument is for options.

Options are the same as in a single-column foreign key.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.integer(),
      name: t.string(), // string is varchar(255)
    }),
    (t) =>
      t.foreignKey(['id', 'name'], 'otherTable', ['foreignId', 'foreignName'], {
        name: 'constraintName',
        match: 'FULL',
        onUpdate: 'RESTRICT',
        onDelete: 'CASCADE',
      }),
  );
});

index

Add an index to the column.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    // add an index to the name column with default settings:
    name: t.text().index(),
    // options are described below:
    name: t.text().index({ ...options }),
    // with a database-level name:
    name: t.text().index('custom_index_name'),
    // with name and options:
    name: t.text().index('custom_index_name', { ...options }),
  }));
});

Possible options are:

type IndexOptions = {
  // NULLS NOT DISTINCT: availabe in Postgres 15+, makes sense only for unique index
  nullsNotDistinct?: true;
  // index algorithm to use such as GIST, GIN
  using?: string;
  // specify collation:
  collate?: string;
  // see `opclass` in the Postgres document for creating the index
  opclass?: string;
  // specify index order such as ASC NULLS FIRST, DESC NULLS LAST
  order?: string;
  // include columns to an index to optimize specific queries
  include?: MaybeArray<string>;
  // see "storage parameters" in the Postgres document for creating an index, for example, 'fillfactor = 70'
  with?: string;
  // The tablespace in which to create the index. If not specified, default_tablespace is consulted, or temp_tablespaces for indexes on temporary tables.
  tablespace?: string;
  // WHERE clause to filter records for the index
  where?: string;
  // mode is for dropping the index
  mode?: 'CASCADE' | 'RESTRICT';
};

composite index

To defines an index for multiple columns.

The first argument is an array of columns, where the column can be a simple string or an object with such options:

type IndexColumnOptions = {
  // column name OR expression is required
  column: string;
  // SQL expression, like 'lower(name)'
  expression: string;

  collate?: string;
  opclass?: string; // for example, varchar_ops
  order?: string; // ASC, DESC, ASC NULLS FIRST, DESC NULLS LAST
};

The second argument is an optional object with index options:

type IndexOptions = {
  // see the comments above for these options
  using?: string;
  include?: MaybeArray<string>;
  nullsNotDistinct?: true;
  with?: string;
  tablespace?: string;
  where?: string;
  mode?: 'CASCADE' | 'RESTRICT';
};

Example:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.identity().primaryKey(),
      name: t.text(),
    }),
    (t) => [
      t.index(['id', { column: 'name', order: 'ASC' }], { ...options }),
      // index name can be set before the options:
      t.index(['id', { column: 'name', order: 'ASC' }], 'index_name', {
        ...options,
      }),
    ],
  );
});

unique

Accepts the same parameters as index.

Columns marked with unique becomes available for filtering with findBy, and in onConflict(['column']).

composite unique index

For unique indexes on multiple columns, accepts the same parameters as composite index.

As well as unique defined for a single column, it is recognized by findBy and onConflict(['column']).

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.identity().primaryKey(),
      name: t.text(),
    }),
    (t) => [
      t.unique(['id', 'name']),
      // with a name
      t.unique(['id', 'name'], 'unique_index_name'),
      // with name and options
      t.unique(['id', 'name'], 'unique_index_name', { ...options }),
    ],
  );
});

searchIndex

searchIndex is designed for full text search.

It can accept the same options as a regular index, but it is USING GIN by default, and it is concatenating columns into a tsvector database type.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.identity().primaryKey(),
      title: t.text(),
      body: t.text(),
    }),
    (t) => t.searchIndex(['title', 'body']),
  );
});

Produces the following index ('english' is a default language, see full text search for changing it):

CREATE INDEX "table_title_body_idx" ON "table" USING GIN (
  to_tsvector('english', "title" || ' ' || "body")
)

You can set different search weights (A to D) on different columns inside the index:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.identity().primaryKey(),
      title: t.text(),
      body: t.text(),
    }),
    (t) => [
      ...t.searchIndex([
        { column: 'title', weight: 'A' },
        { column: 'body', weight: 'B' },
      ]),
    ],
  );
});

When the table has localized columns, you can define different indexes for different languages by setting the language parameter:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.identity().primaryKey(),
      titleEn: t.text(),
      bodyEn: t.text(),
      titleFr: t.text(),
      bodyFr: t.text(),
    }),
    (t) => [
      t.searchIndex(['titleEn', 'bodyEn'], { language: 'english' }),
      t.searchIndex(['titleFr', 'bodyFr'], { language: 'french' }),
    ],
  );
});

Alternatively, different table records may correspond to a single language, then you can define a search index that relies on a language column by using languageColumn parameter:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      id: t.identity().primaryKey(),
      lang: t.type('regconfig'),
      title: t.text(),
      body: t.text(),
    }),
    (t) => t.searchIndex(['title', 'body'], { languageColumn: 'lang' }),
  );
});

It can be more efficient to use a generated column instead of indexing text column in the way described above, and to set a searchIndex on it:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    id: t.identity().primaryKey(),
    title: t.text(),
    body: t.text(),
    generatedTsVector: t.tsvector().generated(['title', 'body']).searchIndex(),
  }));
});

Produces the following index:

CREATE INDEX "table_generatedTsVector_idx" ON "table" USING GIN ("generatedTsVector")

timestamps

Adds createdAt and updatedAt columns of type timestamp (with time zone) with default SQL now().

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    ...t.timestamps(),
  }));
});

timestampsNoTZ

The same as timestamps, but without a time zone.

check

Set a database-level validation check to a column. check accepts a raw SQL.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    // validate rank to be from 1 to 10
    rank: t.integer().check(t.sql`1 >= "rank" AND "rank" <= 10`),
    // constraint name can be passed as a second argument
    column: t.integer().check(t.sql`...`, 'check_name'),
  }));
});

multi-column check

Define a check for multiple column by using a spread syntax:

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable(
    'table',
    (t) => ({
      a: t.integer(),
      b: t.integer(),
    }),
    (t) => [
      t.check(t.sql`a < b`),
      // constraint name can be passed as a second argument
      t.integer().check(t.sql`...`, 'check_name'),
    ],
  );
});

comment

Add database comment to the column.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    name: t.text().comment('This is a column comment'),
  }));
});

compression

Set compression for the column, see Postgres docs for it.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    name: t.text().compression('value'),
  }));
});

collate

Set collation for the column.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    name: t.text().collate('es_ES'),
  }));
});

unsupported types

For user-defined types or for types that are not supported yet, use type:

When using type to define columns in application, you need to also specify as so application knows the actual type behind the domain.

In migration, as won't have effect.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    name: t.type('type_name'),
  }));
});

domain

Domain is a custom database type that allows to predefine a NOT NULL and a CHECK (see postgres tutorial).

Before adding a domain column, create the domain type itself, see create domain.

as works exactly like as when using type, it has no effect in the migration.

import { change } from '../dbScript';

change(async (db) => {
  await db.createTable('table', (t) => ({
    name: t.domain('domainName'),
  }));
});