I have a table column that uses an enum type. I wish to update that enum type to have an additional possible value. I don't want to delete any existing values, just add the new value. What is the simplest way to do this?
This question is related to
database
postgresql
enums
For those looking for an in-transaction solution, the following seems to work.
Instead of an ENUM, a DOMAIN shall be used on type TEXT with a constraint checking that the value is within the specified list of allowed values (as suggested by some comments). The only problem is that no constraint can be added (and thus neither modified) to a domain if it is used by any composite type (the docs merely says this "should eventually be improved"). Such a restriction may be worked around, however, using a constraint calling a function, as follows.
START TRANSACTION;
CREATE FUNCTION test_is_allowed_label(lbl TEXT) RETURNS BOOL AS $function$
SELECT lbl IN ('one', 'two', 'three');
$function$ LANGUAGE SQL IMMUTABLE;
CREATE DOMAIN test_domain AS TEXT CONSTRAINT val_check CHECK (test_is_allowed_label(value));
CREATE TYPE test_composite AS (num INT, word test_domain);
CREATE TABLE test_table (val test_composite);
INSERT INTO test_table (val) VALUES ((1, 'one')::test_composite), ((3, 'three')::test_composite);
-- INSERT INTO test_table (val) VALUES ((4, 'four')::test_composite); -- restricted by the CHECK constraint
CREATE VIEW test_view AS SELECT * FROM test_table; -- just to show that the views using the type work as expected
CREATE OR REPLACE FUNCTION test_is_allowed_label(lbl TEXT) RETURNS BOOL AS $function$
SELECT lbl IN ('one', 'two', 'three', 'four');
$function$ LANGUAGE SQL IMMUTABLE;
INSERT INTO test_table (val) VALUES ((4, 'four')::test_composite); -- allowed by the new effective definition of the constraint
SELECT * FROM test_view;
CREATE OR REPLACE FUNCTION test_is_allowed_label(lbl TEXT) RETURNS BOOL AS $function$
SELECT lbl IN ('one', 'two', 'three');
$function$ LANGUAGE SQL IMMUTABLE;
-- INSERT INTO test_table (val) VALUES ((4, 'four')::test_composite); -- restricted by the CHECK constraint, again
SELECT * FROM test_view; -- note the view lists the restricted value 'four' as no checks are made on existing data
DROP VIEW test_view;
DROP TABLE test_table;
DROP TYPE test_composite;
DROP DOMAIN test_domain;
DROP FUNCTION test_is_allowed_label(TEXT);
COMMIT;
Previously, I used a solution similar to the accepted answer, but it is far from being good once views or functions or composite types (and especially views using other views using the modified ENUMs...) are considered. The solution proposed in this answer seems to work under any conditions.
The only disadvantage is that no checks are performed on existing data when some allowed values are removed (which might be acceptable, especially for this question). (A call to ALTER DOMAIN test_domain VALIDATE CONSTRAINT val_check ends up with the same error as adding a new constraint to the domain used by a composite type, unfortunately.)
Note that a slight modification such as (it works, actually - it was my error)CHECK (value = ANY(get_allowed_values())), where get_allowed_values() function returned the list of allowed values, would not work - which is quite strange, so I hope the solution proposed above works reliably (it does for me, so far...).
As discussed above, ALTER command cannot be written inside a transaction. The suggested way is to insert into the pg_enum table directly, by retrieving the typelem from pg_type table and calculating the next enumsortorder number;
Following is the code that I use. (Checks if duplicate value exists before inserting (constraint between enumtypid and enumlabel name)
INSERT INTO pg_enum (enumtypid, enumlabel, enumsortorder)
SELECT typelem,
'NEW_ENUM_VALUE',
(SELECT MAX(enumsortorder) + 1
FROM pg_enum e
JOIN pg_type p
ON p.typelem = e.enumtypid
WHERE p.typname = '_mytypename'
)
FROM pg_type p
WHERE p.typname = '_mytypename'
AND NOT EXISTS (
SELECT * FROM
pg_enum e
JOIN pg_type p
ON p.typelem = e.enumtypid
WHERE e.enumlabel = 'NEW_ENUM_VALUE'
AND p.typname = '_mytypename'
)
Note that your type name is prepended with an underscore in the pg_type table. Also, the typname needs to be all lowercase in the where clause.
Now this can be written safely into your db migrate script.
Complementing @Dariusz 1
For Rails 4.2.1, there's this doc section:
== Transactional Migrations
If the database adapter supports DDL transactions, all migrations will automatically be wrapped in a transaction. There are queries that you can't execute inside a transaction though, and for these situations you can turn the automatic transactions off.
class ChangeEnum < ActiveRecord::Migration
disable_ddl_transaction!
def up
execute "ALTER TYPE model_size ADD VALUE 'new_value'"
end
end
Simplest: get rid of enums. They are not easily modifiable, and thus should very rarely be used.
Disclaimer: I haven't tried this solution, so it might not work ;-)
You should be looking at pg_enum. If you only want to change the label of an existing ENUM, a simple UPDATE will do it.
To add a new ENUM values:
pg_enum. If the new value has to be the last, you're done.pg_enum in the opposite order.Illustration
You have the following set of labels:
ENUM ('enum1', 'enum2', 'enum3')
and you want to obtain:
ENUM ('enum1', 'enum1b', 'enum2', 'enum3')
then:
INSERT INTO pg_enum (OID, 'newenum3');
UPDATE TABLE SET enumvalue TO 'newenum3' WHERE enumvalue='enum3';
UPDATE TABLE SET enumvalue TO 'enum3' WHERE enumvalue='enum2';
then:
UPDATE TABLE pg_enum SET name='enum1b' WHERE name='enum2' AND enumtypid=OID;
And so on...
I don't know if have other option but we can drop the value using:
select oid from pg_type where typname = 'fase';'
select * from pg_enum where enumtypid = 24773;'
select * from pg_enum where enumtypid = 24773 and enumsortorder = 6;
delete from pg_enum where enumtypid = 24773 and enumsortorder = 6;
A possible solution is the following; precondition is, that there are not conflicts in the used enum values. (e.g. when removing an enum value, be sure that this value is not used anymore.)
-- rename the old enum
alter type my_enum rename to my_enum__;
-- create the new enum
create type my_enum as enum ('value1', 'value2', 'value3');
-- alter all you enum columns
alter table my_table
alter column my_column type my_enum using my_column::text::my_enum;
-- drop the old enum
drop type my_enum__;
Also in this way the column order will not be changed.
PostgreSQL 9.1 introduces ability to ALTER Enum types:
ALTER TYPE enum_type ADD VALUE 'new_value'; -- appends to list
ALTER TYPE enum_type ADD VALUE 'new_value' BEFORE 'old_value';
ALTER TYPE enum_type ADD VALUE 'new_value' AFTER 'old_value';
When using Navicat you can go to types (under view -> others -> types) - get the design view of the type - and click the "add label" button.
I can't seem to post a comment, so I'll just say that updating pg_enum works in Postgres 8.4 . For the way our enums are set up, I've added new values to existing enum types via:
INSERT INTO pg_enum (enumtypid, enumlabel)
SELECT typelem, 'NEWENUM' FROM pg_type WHERE
typname = '_ENUMNAME_WITH_LEADING_UNDERSCORE';
It's a little scary, but it makes sense given the way Postgres actually stores its data.
just in case, if you are using Rails and you have several statements you will need to execute one by one, like:
execute "ALTER TYPE XXX ADD VALUE IF NOT EXISTS 'YYY';"
execute "ALTER TYPE XXX ADD VALUE IF NOT EXISTS 'ZZZ';"
From Postgres 9.1 Documentation:
ALTER TYPE name ADD VALUE new_enum_value [ { BEFORE | AFTER } existing_enum_value ]
Example:
ALTER TYPE user_status ADD VALUE 'PROVISIONAL' AFTER 'NORMAL'
If you are using Postgres 12 you can just run ALTER TYPE ... ADD VALUE inside of transaction (documentation).
If ALTER TYPE ... ADD VALUE (the form that adds a new value to an enum type) is executed inside a transaction block, the new value cannot be used until after the transaction has been committed.
So no hacks needed in migrations.
UPD: here is an example (thanks to Nick for it)
ALTER TYPE enum_type ADD VALUE 'new_value';
Here is a more general but a rather fast-working solution, which apart from changing the type itself updates all columns in the database using it. The method can be applied even if a new version of ENUM is different by more than one label or misses some of the original ones. The code below replaces my_schema.my_type AS ENUM ('a', 'b', 'c') with ENUM ('a', 'b', 'd', 'e'):
CREATE OR REPLACE FUNCTION tmp() RETURNS BOOLEAN AS
$BODY$
DECLARE
item RECORD;
BEGIN
-- 1. create new type in replacement to my_type
CREATE TYPE my_schema.my_type_NEW
AS ENUM ('a', 'b', 'd', 'e');
-- 2. select all columns in the db that have type my_type
FOR item IN
SELECT table_schema, table_name, column_name, udt_schema, udt_name
FROM information_schema.columns
WHERE
udt_schema = 'my_schema'
AND udt_name = 'my_type'
LOOP
-- 3. Change the type of every column using my_type to my_type_NEW
EXECUTE
' ALTER TABLE ' || item.table_schema || '.' || item.table_name
|| ' ALTER COLUMN ' || item.column_name
|| ' TYPE my_schema.my_type_NEW'
|| ' USING ' || item.column_name || '::text::my_schema.my_type_NEW;';
END LOOP;
-- 4. Delete an old version of the type
DROP TYPE my_schema.my_type;
-- 5. Remove _NEW suffix from the new type
ALTER TYPE my_schema.my_type_NEW
RENAME TO my_type;
RETURN true;
END
$BODY$
LANGUAGE 'plpgsql';
SELECT * FROM tmp();
DROP FUNCTION tmp();
The whole process will run fairly quickly, because if the order of labels persists, no actual change of data will happen. I applied the method on 5 tables using my_type and having 50,000-70,000 rows in each, and the whole process took just 10 seconds.
Of course, the function will return an exception in case if labels that are missing in the new version of the ENUM are used somewhere in the data, but in such situation something should be done beforehand anyway.
Can't add a comment to the appropriate place, but ALTER TABLE foo ALTER COLUMN bar TYPE new_enum_type USING bar::text::new_enum_type with a default on the column failed. I had to:
ALTER table ALTER COLUMN bar DROP DEFAULT;
and then it worked.
Updating pg_enum works, as does the intermediary column trick highlighted above. One can also use USING magic to change the column's type directly:
CREATE TYPE test AS enum('a', 'b');
CREATE TABLE foo (bar test);
INSERT INTO foo VALUES ('a'), ('b');
ALTER TABLE foo ALTER COLUMN bar TYPE varchar;
DROP TYPE test;
CREATE TYPE test as enum('a', 'b', 'c');
ALTER TABLE foo ALTER COLUMN bar TYPE test
USING CASE
WHEN bar = ANY (enum_range(null::test)::varchar[])
THEN bar::test
WHEN bar = ANY ('{convert, these, values}'::varchar[])
THEN 'c'::test
ELSE NULL
END;
As long as you've no functions that explicitly require or return that enum, you're good. (pgsql will complain when you drop the type if there are.)
Also, note that PG9.1 is introducing an ALTER TYPE statement, which will work on enums:
http://developer.postgresql.org/pgdocs/postgres/release-9-1-alpha.html
If you fall into situation when you should add enum values in transaction, f.e. execute it in flyway migration on ALTER TYPE statement you will be get error ERROR: ALTER TYPE ... ADD cannot run inside a transaction block (see flyway issue #350) you could add such values into pg_enum directly as workaround (type_egais_units is name of target enum):
INSERT INTO pg_enum (enumtypid, enumlabel, enumsortorder)
SELECT 'type_egais_units'::regtype::oid, 'NEW_ENUM_VALUE', ( SELECT MAX(enumsortorder) + 1 FROM pg_enum WHERE enumtypid = 'type_egais_units'::regtype )
Source: Stackoverflow.com