PostgreSQL 9.3 - How to insert using json_populate_recordset but still get auto ids by sequence - sql

Question

This is a followup question to "Populate multiple tables from a single JSON object with json_populate_recordset".
Say my tables (table_a, table_b) have sequences for a_id and b_id,
so they look like this:
CREATE TABLE table_a (a_id serial, b_id integer, c_id integer);
CREATE TABLE table_b (b_id serial, name text, z_id integer);
...
CREATE TABLE table_z (z_id serial, some_text_entry text);
A regular insert looking like this
INSERT INTO table_a (b_id, c_id) values (1, 2);
INSERT INTO table_b (name, z_id) values ("whatever", 3);
...
INSERT INTO table_z (some_text_entry) values ("Some text");
lets PostgreSQL increment the ids for a_id (table_a), b_id(table_b), etc., so the new dataset can be created and stored.
But how could I insert data using the more complex approach following and still get automatically created ids?
WITH input AS (
SELECT '{
"tablename_a":[{"a_id":1,"b_id":2,"c_id":3},
{"a_id":2,"b_id":51,"c_id":3}],
"tablename_b":[{"b_id":2,"name":"John Doe", "z_id":123},
{"b_id":51,"name":"Mary Ann", "z_id":412}],
"tablename_z":[{"z_id":123, "some_text_entry":"Something"},
{"z_id":123, "some_text_entry":"Something else"}]
}'::json AS j
)
, a AS (
INSERT INTO tablename_a
SELECT t.*
FROM input i
, json_populate_recordset(NULL::tablename_a, i.j->'tablename_a') t
)
, b AS (
INSERT INTO table_b
SELECT t.*
FROM input i
, json_populate_recordset(NULL::tablename_b, i.j->'tablename_b') t
)
-- ... more ...
INSERT INTO tablename_z
SELECT t.*
FROM input i
, json_populate_recordset(NULL::tablename_z, i.j->'tablename_z') t
;
If I just set the id to null
"tablename_a":[{"a_id": null,"b_id":2,"c_id":3},
{"a_id": null,"b_id":51,"c_id":3}]
or leave out the respective ids like this:
"tablename_a":[{"b_id":2,"c_id":3},
{"b_id":51,"c_id":3}]
I get the error 'null value in column "a_id" violates not-null constraint'.
In addition:
Would be really great if there was a way to deal with this, without specifically listing all column names (except the serial-ones).
In MySQL if you take my table_a, I know you could do something like
INSERT INTO table_a VALUES (1,2)
meaning that MySQL would fill the a_id field with an autoincrement id and the rest with the data given by the insert command. Is there no way to accomplish that with json, too?

Answer 1

Usually, when you want to insert into a table that has a column of type serial or bigserial, you omit that column from the list of columns. I tried to change your original syntax as little as possible to show how this works.
-- I made no changes to the CTE, except to cut some lines.
WITH input AS (
SELECT '{
"tablename_a":[{"a_id":1,"b_id":2,"c_id":3}, {"a_id":2,"b_id":51,"c_id":3}]
}'::json AS j
)
-- Ignore a_id. It's set automatically by PostgreSQL.
-- Think about cutting it from input.
INSERT INTO table_a (b_id, c_id)
SELECT b_id, c_id
FROM input i
, json_populate_recordset(NULL::table_a, i.j->'tablename_a') t