2

I am trying to work out how an auto increment key is represented in Postgres, depending on whether you create it using the SERIAL type or using an IDENTITY.

When I create a table with a SERIAL primary key using this DDL:

CREATE TABLE public.test_change_column (
    id SERIAL PRIMARY KEY NOT NULL
)

I get a sequence called 'test_change_column_id_seq' and gives the column id a default value of nextval('test_change_column_id_seq'::regclass)

If I create a table without SERIAL and a primary key using this DDL:

CREATE TABLE public.test_change_column (
    id INTEGER PRIMARY KEY NOT NULL GENERATED ALWAYS AS IDENTITY,
)

The creates the sequence but the default value of id is not set to nextval('test_change_column_id_seq'::regclass). Instead the is_identity column is set to "YES".

If you create the column without an autoincrement, you cannot add it at a later stage using the SERIAL:

ALTER TABLE public.test_change_column  ALTER COLUMN id TYPE serial;

this results in the error "type "serial" does not exist".

Using the IDENTITY method, you can add the auto increment using this DDL:

ALTER TABLE public.test_change_column ALTER COLUMN id ADD GENERATED BY DEFAULT AS IDENTITY;

This SQL will show how postgres stores the metadata for the 2 different methods:

SELECT column_name, column_default, is_identity 
FROM information_schema.columns 
WHERE table_name = 'test_change_column';

When it comes removing an auto increment, the auto increment is deleted differently depending on whether you used serial or identity.

If the auto increment was created using the serial type, you have to ALTER COLUMN id DROP DEFAULT. This does not delete the associated sequence table.

If the auto increment was created using IDENTITY, you have to ALTER COLUMN id DROP IDENTITY. This also removes the sequence table.

Improve this question
1

2 Answers 2

Reset to default
1

Create table.

CREATE TABLE public.test_change_column (
    id INTEGER NOT NULL
);

Alter table in one line:

ALTER TABLE 
    public.test_change_column ALTER COLUMN id ADD GENERATED BY DEFAULT AS IDENTITY, 
    ADD CONSTRAINT test_change_column_pkey PRIMARY KEY(id);

\d test_change_column
                     Table "public.test_change_column"
 Column |  Type   | Collation | Nullable |             Default              
--------+---------+-----------+----------+----------------------------------
 id     | integer |           | not null | generated by default as identity
Indexes:
    "test_change_column_pkey" PRIMARY KEY, btree (id)

An IDENTITY column has a sequence associated with it that is used just like with a serial column. It is just not shown as the DEFAULT. To find underlying sequence:

select pg_get_serial_sequence ('public.test_change_column', 'id');
      pg_get_serial_sequence      
----------------------------------
 public.test_change_column_id_seq

select * from public.test_change_column_id_seq ;
 last_value | log_cnt | is_called 
------------+---------+-----------
          1 |       0 | f

You can manipulate the above with the IDENTITY commands here ALTER TABLE:

ALTER [ COLUMN ] column_name ADD GENERATED { ALWAYS | BY DEFAULT } AS IDENTITY [ ( sequence_options ) ]

ALTER [ COLUMN ] column_name { SET GENERATED { ALWAYS | BY DEFAULT } | SET sequence_option | RESTART [ [ WITH ] restart ] } [...]

ALTER [ COLUMN ] column_name DROP IDENTITY [ IF EXISTS ]

UPDATE

From here serial type:

The data types smallserial, serial and bigserial are not true types, but merely a notational convenience for creating unique identifier columns (similar to the AUTO_INCREMENT property supported by some other databases). In the current implementation, specifying:

CREATE TABLE tablename ( colname SERIAL );

is equivalent to specifying:

CREATE SEQUENCE tablename_colname_seq AS integer;

CREATE TABLE tablename ( colname integer NOT NULL DEFAULT nextval('tablename_colname_seq') );

ALTER SEQUENCE tablename_colname_seq OWNED BY tablename.colname;

Thus, we have created an integer column and arranged for its default values to be assigned from a sequence generator. A NOT NULL constraint is applied to ensure that a null value cannot be inserted. (In most cases you would also want to attach a UNIQUE or PRIMARY KEY constraint to prevent duplicate values from being inserted by accident, but this is not automatic.) Lastly, the sequence is marked as “owned by” the column, so that it will be dropped if the column or table is dropped.

So if you want to ALTER a table to replicate a serial 'type' you will need to do the individual steps above:

CREATE TABLE public.test_change_column (id INTEGER NOT NULL);

create sequence test_change_column_seq;


alter 
    table test_change_column alter COLUMN id 
    set default nextval('test_change_column_seq'), 
ADD CONSTRAINT test_change_column_pkey PRIMARY KEY(id);
alter sequence test_change_column_seq owned by test_change_column.id;

\d test_change_column
                           Table "public.test_change_column"
 Column |  Type   | Collation | Nullable |                   Default                   
--------+---------+-----------+----------+---------------------------------------------
 id     | integer |           | not null | nextval('test_change_column_seq'::regclass)
Indexes:
    "test_change_column_pkey" PRIMARY KEY, btree (id)
Improve this answer
Sign up to request clarification or add additional context in comments.

5 Comments

thanks. The issue I am seeing appears to be with how the column_default is being read from the information_schema (see above).
What information are you wanting to get? information_schema.columns does have columns that return the sequence values: identity_*.
I want column_default to be populated in the same as by id SERIAL PRIMARY KEY NOT NULL - this populates column_default whereas the ALTER statements do not.
Why? serial is considered deprecated in Postgres versions that have IDENTITY.
See my UPDATE on how to alter to a serial 'type'.
0

Changing the column to an identity column (and adding the PK constraint) is enough and the correct thing to do (the use of the serial pseudo type is discouraged)

You can see that e.g. psql reports this correctly as an identity column:

arthur=> CREATE TABLE public.test_change_column (id INTEGER NOT NULL);
CREATE TABLE
arthur=> \d public.test_change_column
         Table "public.test_change_column"
 Column |  Type   | Collation | Nullable | Default
--------+---------+-----------+----------+---------
 id     | integer |           | not null |


arthur=> ALTER TABLE public.test_change_column ALTER COLUMN id ADD GENERATED BY DEFAULT AS IDENTITY;
ALTER TABLE
arthur=> ALTER TABLE public.test_change_column ADD CONSTRAINT test_change_column_pkey PRIMARY KEY(id);
ALTER TABLE
arthur=> \d public.test_change_column
                     Table "public.test_change_column"
 Column |  Type   | Collation | Nullable |             Default
--------+---------+-----------+----------+----------------------------------
 id     | integer |           | not null | generated by default as identity
Indexes:
    "test_change_column_pkey" PRIMARY KEY, btree (id)

You can also verify that the default is working by inserting a row:

insert into public.test_change_column default values;

Will create a new row and will increment the id value.

Such a column will be shown in information_schema.columns with is_identity = 'YES' instead of having a default value.

I recommend to use generated always as identity so that you can't accidentally bypass the generation through the sequence (with the effect that the sequence and the values in the table are no longer "in sync").

Improve this answer

9 Comments

\d public.test_change_column does show the default but when I run "SELECT column_name, column_default FROM information_schema.columns WHERE table_name = 'test_change_column' ORDER BY ordinal_position;" column_default is null
@jmc42: but is_identity is set to YES to identify the columns as an identity column.
is_identity is set to YES but column_default is null.
@jmc42: yes, that's correct. An identity column will not show a default value. The "default" is implied by being an identity column.
@a-horse-with-no-name. Is there a way that the ALTER statement can result in the column default having a value (rather than null) in information_schema.columns?
|

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.