如何插入通过外键引用另一个 postgres 表的行，并在它不存在时创建外行？

Question

In Postgres, is there a way to atomically insert a row into a table, where one column references another table, and we look up to see if the desired row exists in the referenced table and inserts it as well if it is not?

For example, say we have a US states table and a cities table which references the states table:

CREATE TABLE states (
  state_id serial primary key,
  name text
);
CREATE TABLE cities (
  city_id serial,
  name text,
  state_id int references states(state_id)
);

When I want to add the city of Austin, Texas, I want to be able to see whether Texas exists in the states table, and if so use its state_id in the new row I'm inserting in the cities table. If Texas doesn't exist in the states table, I want to create it and then use its id in the cities table.

I tried this query, but I got an error saying

ERROR:  WITH clause containing a data-modifying statement must be at the top level
LINE 2:   WITH inserted AS (
               ^

WITH state_id AS (
  WITH inserted AS (
         INSERT INTO states(name)
           VALUES ('Texas')
           ON CONFLICT DO NOTHING
           RETURNING state_id),
       already_there AS (
         SELECT state_id FROM states
           WHERE name='Texas')
  SELECT * FROM inserted
  UNION
  SELECT * FROM already_there)
INSERT INTO cities(name, state_id)
  VALUES
    ('Austin', (SELECT state_id FROM state_id));

Am I overlooking a simple solution?

Answer 1

You can force the insert statement to return a value:

WITH inserted AS (
      INSERT INTO states (name)
          VALUES ('Texas')
          ON CONFLICT (name) DO UPDATE SET name = EXCLUDED.NAME
          RETURNING state_id
     )
. . .

The DO UPDATE SET forces the INSERT to return something.

I notice that you don't have a unique constraint, so you also need that:

ALTER TABLE states ADD CONSTRAINT unq_state_name
    UNIQUE (name);

Otherwise the ON CONFLICT doesn't have anything to work with.

Answer 2

Here is one option:

with inserted as (
    insert into states(name) values ('Texas')
    on conflict do nothing
    returning state_id
)
insert into cities(name, state_id)
values (
    'Dallas', 
    coalesce(
        (select state_id from inserted),
        (select state_id from states where name = 'Texas')
    )
);

The idea is to attempt to insert in a CTE, and then, in the main insert, check if a value was inserted, else select it.

For this to work properly, you need a unique constraint on states(name) :

create table states (
  state_id serial primary key,
  name text unique
);

Demo on DB Fiddlde