In order to use the PostgreSQL adapter you need to have at least version 9.3 installed. Older versions are not supported.
To get started with PostgreSQL have a look at the configuring Rails guide. It describes how to properly set up Active Record for PostgreSQL.
1. Datatypes
PostgreSQL offers a number of specific datatypes. Following is a list of types, that are supported by the PostgreSQL adapter.
1.1. Bytea
# db/migrate/20140207133952_create_documents.rbcreate_table:documentsdo|t|t.binary"payload"end# app/models/document.rbclassDocument<ApplicationRecordend# Usagedata=File.read(Rails.root+"tmp/output.pdf")Document.createpayload: data1.2. Array
# db/migrate/20140207133952_create_books.rbcreate_table:booksdo|t|t.string"title"t.string"tags",array: truet.integer"ratings",array: trueendadd_index:books,:tags,using: "gin"add_index:books,:ratings,using: "gin"# app/models/book.rbclassBook<ApplicationRecordend# UsageBook.createtitle: "Brave New World",tags: ["fantasy","fiction"],ratings: [4,5]## Books for a single tagBook.where("'fantasy' = ANY (tags)")## Books for multiple tagsBook.where("tags @> ARRAY[?]::varchar[]",["fantasy","fiction"])## Books with 3 or more ratingsBook.where("array_length(ratings, 1) >= 3")1.3. Hstore
You need to enable the hstore extension to use hstore.
# db/migrate/20131009135255_create_profiles.rbclassCreateProfiles<ActiveRecord::Migration[8.0]enable_extension"hstore"unlessextension_enabled?("hstore")create_table:profilesdo|t|t.hstore"settings"endend# app/models/profile.rbclassProfile<ApplicationRecordendirb>Profile.create(settings: {"color"=>"blue","resolution"=>"800x600"})irb>profile=Profile.firstirb>profile.settings=>{"color"=>"blue","resolution"=>"800x600"}irb>profile.settings={"color"=>"yellow","resolution"=>"1280x1024"}irb>profile.save!irb>Profile.where("settings->'color' = ?","yellow")=>#<ActiveRecord::Relation[#<Profileid: 1,settings: {"color"=>"yellow","resolution"=>"1280x1024"}>]>1.4. JSON and JSONB
# db/migrate/20131220144913_create_events.rb# ... for json datatype:create_table:eventsdo|t|t.json"payload"end# ... or for jsonb datatype:create_table:eventsdo|t|t.jsonb"payload"end# app/models/event.rbclassEvent<ApplicationRecordendirb>Event.create(payload: {kind: "user_renamed",change: ["jack","john"]})irb>event=Event.firstirb>event.payload=>{"kind"=>"user_renamed","change"=>["jack","john"]}## Query based on JSON document# The -> operator returns the original JSON type (which might be an object), whereas ->>returnstextirb>Event.where("payload->>'kind' = ?","user_renamed")1.5. Range Types
This type is mapped to Ruby Range objects.
# db/migrate/20130923065404_create_events.rbcreate_table:eventsdo|t|t.daterange"duration"end# app/models/event.rbclassEvent<ApplicationRecordendirb>Event.create(duration: Date.new(2014,2,11)..Date.new(2014,2,12))irb>event=Event.firstirb>event.duration=>Tue,11Feb2014...Thu,13Feb2014## All Events on a given dateirb>Event.where("duration @> ?::date",Date.new(2014,2,12))## Working with range boundsirb>event=Event.select("lower(duration) AS starts_at").select("upper(duration) AS ends_at").firstirb>event.starts_at=>Tue,11Feb2014irb>event.ends_at=>Thu,13Feb20141.6. Composite Types
Currently there is no special support for composite types. They are mapped to normal text columns:
CREATETYPEfull_addressAS(cityVARCHAR(90),streetVARCHAR(90));# db/migrate/20140207133952_create_contacts.rbexecute<<-SQL CREATE TYPE full_address AS ( city VARCHAR(90), street VARCHAR(90) );SQLcreate_table:contactsdo|t|t.column:address,:full_addressend# app/models/contact.rbclassContact<ApplicationRecordendirb>Contact.createaddress: "(Paris,Champs-Élysées)"irb>contact=Contact.firstirb>contact.address=>"(Paris,Champs-Élysées)"irb>contact.address="(Paris,Rue Basse)"irb>contact.save!1.7. Enumerated Types
The type can be mapped as a normal text column, or to an ActiveRecord::Enum.
# db/migrate/20131220144913_create_articles.rbdefchangecreate_enum:article_status,["draft","published","archived"]create_table:articlesdo|t|t.enum:status,enum_type: :article_status,default: "draft",null: falseendendYou can also create an enum type and add an enum column to an existing table:
# db/migrate/20230113024409_add_status_to_articles.rbdefchangecreate_enum:article_status,["draft","published","archived"]add_column:articles,:status,:enum,enum_type: :article_status,default: "draft",null: falseendThe above migrations are both reversible, but you can define separate #up and #down methods if required. Make sure you remove any columns or tables that depend on the enum type before dropping it:
defdowndrop_table:articles# OR: remove_column :articles, :statusdrop_enum:article_statusendDeclaring an enum attribute in the model adds helper methods and prevents invalid values from being assigned to instances of the class:
# app/models/article.rbclassArticle<ApplicationRecordenum:status,{draft: "draft",published: "published",archived: "archived"},prefix: trueendirb>article=Article.createirb>article.status=>"draft"# default status from PostgreSQL, as defined in migration aboveirb>article.status_published!irb>article.status=>"published"irb>article.status_archived?=>falseirb>article.status="deleted"ArgumentError: 'deleted' is not a valid statusTo rename the enum you can use rename_enum along with updating any model usage:
# db/migrate/20150718144917_rename_article_status.rbdefchangerename_enum:article_status,:article_stateendTo add a new value you can use add_enum_value:
# db/migrate/20150720144913_add_new_state_to_articles.rbdefupadd_enum_value:article_state,"archived"# will be at the end after publishedadd_enum_value:article_state,"in review",before: "published"add_enum_value:article_state,"approved",after: "in review"add_enum_value:article_state,"rejected",if_not_exists: true# won't raise an error if the value already existsendEnum values can't be dropped, which also means add_enum_value is irreversible. You can read why here.
To rename a value you can use rename_enum_value:
# db/migrate/20150722144915_rename_article_state.rbdefchangerename_enum_value:article_state,from: "archived",to: "deleted"endHint: to show all the values of the all enums you have, you can call this query in bin/rails db or psql console:
SELECTn.nspnameASenum_schema,t.typnameASenum_name,e.enumlabelASenum_valueFROMpg_typetJOINpg_enumeONt.oid=e.enumtypidJOINpg_catalog.pg_namespacenONn.oid=t.typnamespace1.8. UUID
If you're using PostgreSQL earlier than version 13.0 you may need to enable special extensions to use UUIDs. Enable the pgcrypto extension (PostgreSQL >= 9.4) or uuid-ossp extension (for even earlier releases).
# db/migrate/20131220144913_create_revisions.rbcreate_table:revisionsdo|t|t.uuid:identifierend# app/models/revision.rbclassRevision<ApplicationRecordendirb>Revision.createidentifier: "A0EEBC99-9C0B-4EF8-BB6D-6BB9BD380A11"irb>revision=Revision.firstirb>revision.identifier=>"a0eebc99-9c0b-4ef8-bb6d-6bb9bd380a11"You can use uuid type to define references in migrations:
# db/migrate/20150418012400_create_blog.rbenable_extension"pgcrypto"unlessextension_enabled?("pgcrypto")create_table:posts,id: :uuidcreate_table:comments,id: :uuiddo|t|# t.belongs_to :post, type: :uuidt.references:post,type: :uuidend# app/models/post.rbclassPost<ApplicationRecordhas_many:commentsend# app/models/comment.rbclassComment<ApplicationRecordbelongs_to:postendSee this section for more details on using UUIDs as primary key.
1.9. Bit String Types
# db/migrate/20131220144913_create_users.rbcreate_table:users,force: truedo|t|t.column:settings,"bit(8)"end# app/models/user.rbclassUser<ApplicationRecordendirb>User.createsettings: "01010011"irb>user=User.firstirb>user.settings=>"01010011"irb>user.settings="0xAF"irb>user.settings=>"10101111"irb>user.save!1.10. Network Address Types
The types inet and cidr are mapped to Ruby IPAddr objects. The macaddr type is mapped to normal text.
# db/migrate/20140508144913_create_devices.rbcreate_table(:devices,force: true)do|t|t.inet"ip"t.cidr"network"t.macaddr"address"end# app/models/device.rbclassDevice<ApplicationRecordendirb>macbook=Device.create(ip: "192.168.1.12",network: "192.168.2.0/24",address: "32:01:16:6d:05:ef")irb>macbook.ip=>#<IPAddr:IPv4:192.168.1.12/255.255.255.255>irb>macbook.network=>#<IPAddr:IPv4:192.168.2.0/255.255.255.0>irb>macbook.address=>"32:01:16:6d:05:ef"1.11. Geometric Types
All geometric types, with the exception of points are mapped to normal text. A point is cast to an array containing x and y coordinates.
1.12. Interval
This type is mapped to ActiveSupport::Duration objects.
# db/migrate/20200120000000_create_events.rbcreate_table:eventsdo|t|t.interval"duration"end# app/models/event.rbclassEvent<ApplicationRecordendirb>Event.create(duration: 2.days)irb>event=Event.firstirb>event.duration=>2days2. UUID Primary Keys
You need to enable the pgcrypto (only PostgreSQL >= 9.4) or uuid-ossp extension to generate random UUIDs.
# db/migrate/20131220144913_create_devices.rbenable_extension"pgcrypto"unlessextension_enabled?("pgcrypto")create_table:devices,id: :uuiddo|t|t.string:kindend# app/models/device.rbclassDevice<ApplicationRecordendirb>device=Device.createirb>device.id=>"814865cd-5a1d-4771-9306-4268f188fe9e"gen_random_uuid() (from pgcrypto) is assumed if no :default option was passed to create_table.
To use the Rails model generator for a table using UUID as the primary key, pass --primary-key-type=uuid to the model generator.
For example:
$rails generate model Device --primary-key-type=uuid kind:string When building a model with a foreign key that will reference this UUID, treat uuid as the native field type, for example:
$rails generate model Case device_id:uuid 3. Indexing
PostgreSQL includes a variety of index options. The following options are supported by the PostgreSQL adapter in addition to the common index options
3.1. Include
When creating a new index, non-key columns can be included with the :include option. These keys are not used in index scans for searching, but can be read during an index only scan without having to visit the associated table.
# db/migrate/20131220144913_add_index_users_on_email_include_id.rbadd_index:users,:email,include: :idMultiple columns are supported:
# db/migrate/20131220144913_add_index_users_on_email_include_id_and_created_at.rbadd_index:users,:email,include: [:id,:created_at]4. Generated Columns
Generated columns are supported since version 12.0 of PostgreSQL.
# db/migrate/20131220144913_create_users.rbcreate_table:usersdo|t|t.string:namet.virtual:name_upcased,type: :string,as: "upper(name)",stored: trueend# app/models/user.rbclassUser<ApplicationRecordend# Usageuser=User.create(name: "John")User.last.name_upcased# => "JOHN"5. Deferrable Foreign Keys
By default, table constraints in PostgreSQL are checked immediately after each statement. It intentionally does not allow creating records where the referenced record is not yet in the referenced table. It is possible to run this integrity check later on when the transaction is committed by adding DEFERRABLE to the foreign key definition though. To defer all checks by default it can be set to DEFERRABLE INITIALLY DEFERRED. Rails exposes this PostgreSQL feature by adding the :deferrable key to the foreign_key options in the add_reference and add_foreign_key methods.
One example of this is creating circular dependencies in a transaction even if you have created foreign keys:
add_reference:person,:alias,foreign_key: {deferrable: :deferred}add_reference:alias,:person,foreign_key: {deferrable: :deferred}If the reference was created with the foreign_key: true option, the following transaction would fail when executing the first INSERT statement. It does not fail when the deferrable: :deferred option is set though.
ActiveRecord::Base.lease_connection.transactiondoperson=Person.create(id: SecureRandom.uuid,alias_id: SecureRandom.uuid,name: "John Doe")Alias.create(id: person.alias_id,person_id: person.id,name: "jaydee")endWhen the :deferrable option is set to :immediate, let the foreign keys keep the default behavior of checking the constraint immediately, but allow manually deferring the checks using set_constraints within a transaction. This will cause the foreign keys to be checked when the transaction is committed:
ActiveRecord::Base.lease_connection.transactiondoActiveRecord::Base.lease_connection.set_constraints(:deferred)person=Person.create(alias_id: SecureRandom.uuid,name: "John Doe")Alias.create(id: person.alias_id,person_id: person.id,name: "jaydee")endBy default :deferrable is false and the constraint is always checked immediately.
6. Unique Constraint
# db/migrate/20230422225213_create_items.rbcreate_table:itemsdo|t|t.integer:position,null: falset.unique_constraint[:position],deferrable: :immediateendIf you want to change an existing unique index to deferrable, you can use :using_index to create deferrable unique constraints.
add_unique_constraint:items,deferrable: :deferred,using_index: "index_items_on_position"Like foreign keys, unique constraints can be deferred by setting :deferrable to either :immediate or :deferred. By default, :deferrable is false and the constraint is always checked immediately.
7. Exclusion Constraints
# db/migrate/20131220144913_create_products.rbcreate_table:productsdo|t|t.integer:price,null: falset.daterange:availability_range,null: falset.exclusion_constraint"price WITH =, availability_range WITH &&",using: :gist,name: "price_check"endLike foreign keys, exclusion constraints can be deferred by setting :deferrable to either :immediate or :deferred. By default, :deferrable is false and the constraint is always checked immediately.
8. Full Text Search
# db/migrate/20131220144913_create_documents.rbcreate_table:documentsdo|t|t.string:titlet.string:bodyendadd_index:documents,"to_tsvector('english', title || ' ' || body)",using: :gin,name: "documents_idx"# app/models/document.rbclassDocument<ApplicationRecordend# UsageDocument.create(title: "Cats and Dogs",body: "are nice!")## all documents matching 'cat & dog'Document.where("to_tsvector('english', title || ' ' || body) @@ to_tsquery(?)","cat & dog")Optionally, you can store the vector as automatically generated column (from PostgreSQL 12.0):
# db/migrate/20131220144913_create_documents.rbcreate_table:documentsdo|t|t.string:titlet.string:bodyt.virtual:textsearchable_index_col,type: :tsvector,as: "to_tsvector('english', title || ' ' || body)",stored: trueendadd_index:documents,:textsearchable_index_col,using: :gin,name: "documents_idx"# UsageDocument.create(title: "Cats and Dogs",body: "are nice!")## all documents matching 'cat & dog'Document.where("textsearchable_index_col @@ to_tsquery(?)","cat & dog")9. Database Views
Imagine you need to work with a legacy database containing the following table:
rails_pg_guide=# \d "TBL_ART" Table "public.TBL_ART" Column | Type | Modifiers ------------+-----------------------------+------------------------------------------------------------ INT_ID | integer | not null default nextval('"TBL_ART_INT_ID_seq"'::regclass) STR_TITLE | character varying | STR_STAT | character varying | default 'draft'::character varying DT_PUBL_AT | timestamp without time zone | BL_ARCH | boolean | default false Indexes: "TBL_ART_pkey" PRIMARY KEY, btree ("INT_ID") This table does not follow the Rails conventions at all. Because simple PostgreSQL views are updateable by default, we can wrap it as follows:
# db/migrate/20131220144913_create_articles_view.rbexecute<<-SQLCREATE VIEW articles AS SELECT "INT_ID" AS id, "STR_TITLE" AS title, "STR_STAT" AS status, "DT_PUBL_AT" AS published_at, "BL_ARCH" AS archived FROM "TBL_ART" WHERE "BL_ARCH" = 'f'SQL# app/models/article.rbclassArticle<ApplicationRecordself.primary_key="id"defarchive!update_attribute:archived,trueendendirb>first=Article.create!title: "Winter is coming",status: "published",published_at: 1.year.agoirb>second=Article.create!title: "Brace yourself",status: "draft",published_at: 1.month.agoirb>Article.count=>2irb>first.archive!irb>Article.count=>1This application only cares about non-archived Articles. A view also allows for conditions so we can exclude the archived Articles directly.
10. Structure Dumps
If your config.active_record.schema_format is :sql, Rails will call pg_dump to generate a structure dump.
You can use ActiveRecord::Tasks::DatabaseTasks.structure_dump_flags to configure pg_dump. For example, to exclude comments from your structure dump, add this to an initializer:
ActiveRecord::Tasks::DatabaseTasks.structure_dump_flags=["--no-comments"]