107
votes

I have a small table (~30 rows) in my Postgres 9.0 database with an integer ID field (the primary key) which currently contains unique sequential integers starting at 1, but which was not created using the 'serial' keyword.

How can I alter this table such that from now on inserts to this table will cause this field to behave as if it had been created with 'serial' as a type?

3
FYI, the SERIAL pseudo-type is now legacy, supplanted by the new GENERATED … AS IDENTITY feature defined in SQL:2003, in Postgres 10 and later. See explanation.Basil Bourque
For modern Postgres version (>= 10) see this question: stackoverflow.com/questions/2944499a_horse_with_no_name

3 Answers

147
votes

Look at the following commands (especially the commented block).

DROP TABLE foo;
DROP TABLE bar;

CREATE TABLE foo (a int, b text);
CREATE TABLE bar (a serial, b text);

INSERT INTO foo (a, b) SELECT i, 'foo ' || i::text FROM generate_series(1, 5) i;
INSERT INTO bar (b) SELECT 'bar ' || i::text FROM generate_series(1, 5) i;

-- blocks of commands to turn foo into bar
CREATE SEQUENCE foo_a_seq;
ALTER TABLE foo ALTER COLUMN a SET DEFAULT nextval('foo_a_seq');
ALTER TABLE foo ALTER COLUMN a SET NOT NULL;
ALTER SEQUENCE foo_a_seq OWNED BY foo.a;    -- 8.2 or later

SELECT MAX(a) FROM foo;
SELECT setval('foo_a_seq', 5);  -- replace 5 by SELECT MAX result

INSERT INTO foo (b) VALUES('teste');
INSERT INTO bar (b) VALUES('teste');

SELECT * FROM foo;
SELECT * FROM bar;
55
votes

You can also use START WITH to start a sequence from a particular point, although setval accomplishes the same thing, as in Euler's answer, eg,

SELECT MAX(a) + 1 FROM foo;
CREATE SEQUENCE foo_a_seq START WITH 12345; -- replace 12345 with max above
ALTER TABLE foo ALTER COLUMN a SET DEFAULT nextval('foo_a_seq');
42
votes

TL;DR

Here's a version where you don't need a human to read a value and type it out themselves.

CREATE SEQUENCE foo_a_seq OWNED BY foo.a;
SELECT setval('foo_a_seq', coalesce(max(a), 0) + 1, false) FROM foo;
ALTER TABLE foo ALTER COLUMN a SET DEFAULT nextval('foo_a_seq'); 

Another option would be to employ the reusable Function shared at the end of this answer.


A non-interactive solution

Just adding to the other two answers, for those of us who need to have these Sequences created by a non-interactive script, while patching a live-ish DB for instance.

That is, when you don't wanna SELECT the value manually and type it yourself into a subsequent CREATE statement.

In short, you can not do:

CREATE SEQUENCE foo_a_seq
    START WITH ( SELECT max(a) + 1 FROM foo );

... since the START [WITH] clause in CREATE SEQUENCE expects a value, not a subquery.

Note: As a rule of thumb, that applies to all non-CRUD (i.e.: anything other than INSERT, SELECT, UPDATE, DELETE) statements in pgSQL AFAIK.

However, setval() does! Thus, the following is absolutely fine:

SELECT setval('foo_a_seq', max(a)) FROM foo;

If there's no data and you don't (want to) know about it, use coalesce() to set the default value:

SELECT setval('foo_a_seq', coalesce(max(a), 0)) FROM foo;
--                         ^      ^         ^
--                       defaults to:       0

However, having the current sequence value set to 0 is clumsy, if not illegal.
Using the three-parameter form of setval would be more appropriate:

--                                             vvv
SELECT setval('foo_a_seq', coalesce(max(a), 0) + 1, false) FROM foo;
--                                                  ^   ^
--                                                is_called

Setting the optional third parameter of setval to false will prevent the next nextval from advancing the sequence before returning a value, and thus:

the next nextval will return exactly the specified value, and sequence advancement commences with the following nextval.

— from this entry in the documentation

On an unrelated note, you also can specify the column owning the Sequence directly with CREATE, you don't have to alter it later:

CREATE SEQUENCE foo_a_seq OWNED BY foo.a;

In summary:

CREATE SEQUENCE foo_a_seq OWNED BY foo.a;
SELECT setval('foo_a_seq', coalesce(max(a), 0) + 1, false) FROM foo;
ALTER TABLE foo ALTER COLUMN a SET DEFAULT nextval('foo_a_seq'); 

Using a Function

Alternatively, if you're planning on doing this for multiple columns, you could opt for using an actual Function.

CREATE OR REPLACE FUNCTION make_into_serial(table_name TEXT, column_name TEXT) RETURNS INTEGER AS $$
DECLARE
    start_with INTEGER;
    sequence_name TEXT;
BEGIN
    sequence_name := table_name || '_' || column_name || '_seq';
    EXECUTE 'SELECT coalesce(max(' || column_name || '), 0) + 1 FROM ' || table_name
            INTO start_with;
    EXECUTE 'CREATE SEQUENCE ' || sequence_name ||
            ' START WITH ' || start_with ||
            ' OWNED BY ' || table_name || '.' || column_name;
    EXECUTE 'ALTER TABLE ' || table_name || ' ALTER COLUMN ' || column_name ||
            ' SET DEFAULT nextVal(''' || sequence_name || ''')';
    RETURN start_with;
END;
$$ LANGUAGE plpgsql VOLATILE;

Use it like so:

INSERT INTO foo (data) VALUES ('asdf');
-- ERROR: null value in column "a" violates not-null constraint

SELECT make_into_serial('foo', 'a');
INSERT INTO foo (data) VALUES ('asdf');
-- OK: 1 row(s) affected