Possible to condense primary key/serial?

Question

I have a database where everything is linked with foreign keys, so the Postgres knows exactly how the database is layed out..

Well, Lets say I have Table1 and Table2.

Answer 1

You probably don't want to do this generally, as gapless sequences are problematic for performance.

If you want to do it as a cleanup step later you can use the rank() window function to achieve the desired effect.

CREATE TABLE table1 (id integer primary key);
INSERT INTO table1 values (1),(2),(4),(5);

CREATE TABLE table2 (
  id serial primary key,
  rid integer references table1(id) ON UPDATE CASCADE
);

insert into table2 (rid) values (1),(1),(4),(4),(4),(5);



UPDATE table1 
  SET id = gapless_id
FROM (
  SELECT *, row_number() OVER () FROM table1
) AS x(old_id, gapless_id)
WHERE id = x.old_id;

Result:

regress=# select * from table1 ;
 id 
----
  1
  2
  3
  4
(4 rows)

If your FK's aren't ON UPDATE CASCADE you can ALTER TABLE to make them so. This will be quite slow though, especially if there are no indexes on the foreign keys. A faster approach is to do the change in two passes:

• Begin a transaction
• LOCK TABLE table1;
• Add a new_id column to table1 and populate it with the new IDs using row_number() as shown above
• Drop foreign key constraints referring to table1(id)
• Update all foreign keys to refer to the values in new_id
• drop id in table1
• rename the new_id column of table1 to id
• re-create foreign key constraints
• commit