Return multiple columns of the same row as JSON array of objects

Question

I have the following table MyTable:

 id │ value_two │ value_three │ value_four 
────┼───────────┼─────────────┼────────────
  1 │ a         │ A           │ AA
  2 │ a         │ A2          │ AA2
  3 │ b         │ A3          │ AA3
  4 │ a         │ A4          │ AA4
  5 │ b         │ A5          │ AA5

I want to query an array of objects { value_three, value_four } grouped by value_two. value_two should be present on its own in the result. The result should look like this:

 value_two │                                                                                    value_four                                                                                 
───────────┼───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
 a         │ [{"value_three":"A","value_four":"AA"}, {"value_three":"A2","value_four":"AA2"}, {"value_three":"A4","value_four":"AA4"}]
 b         │ [{"value_three":"A3","value_four":"AA3"}, {"value_three":"A5","value_four":"AA5"}]

It does not matter whether it uses json_agg() or array_agg().

However the best I can do is:

with MyCTE as ( select value_two, value_three, value_four from MyTable ) 
select value_two, json_agg(row_to_json(MyCTE)) value_four 
from MyCTE 
group by value_two;

Which returns:

 value_two │                                                                                    value_four                                                                                 
───────────┼───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
 a         │ [{"value_two":"a","value_three":"A","value_four":"AA"}, {"value_two":"a","value_three":"A2","value_four":"AA2"}, {"value_two":"a","value_three":"A4","value_four":"AA4"}]
 b         │ [{"value_two":"b","value_three":"A3","value_four":"AA3"}, {"value_two":"b","value_three":"A5","value_four":"AA5"}]

With an extra value_two key in the objects, which I would like to get rid of. Which SQL (Postgres) query should I use?

Waterer

Answer 1

Postgres 10+

Convert the whole row to a jsonb object and eliminate a single key (pg 9.5+) or an array of keys (pg 10+) with the - operator before aggregating:

SELECT val2, jsonb_agg(to_jsonb(t.*) - '{id, val2}'::text[]) AS js_34
FROM   tbl t
GROUP  BY val2;

The explicit cast in '{id, val2}'::text[] is necessary to disambiguate from the overloaded function taking a single key as text.

See:

• Select columns inside jsonb_agg

Postgres 9.4+

jsonb_build_object() or json_build_object().

SELECT val2, jsonb_agg(jsonb_build_object('val3', val3, 'val4', val4)) AS js_34
FROM   tbl 
GROUP  BY val2;

The manual:

Builds a JSON object out of a variadic argument list. By convention, the argument list consists of alternating keys and values.

Postgres 9.3+

to_jsonb() (or to_json) with a ROW expression does the trick. (Or row_to_json() with optional line feeds):

SELECT val2, jsonb_agg(to_jsonb((val3, val4))) AS js_34
FROM   tbl
GROUP  BY val2;

But you lose original column names. A cast to a registered row type avoids that. (The row type of a temporary table serves for ad hoc queries, too.)

CREATE TYPE foo AS (val3 text, val4 text);  -- once in the same session

SELECT val2, jsonb_agg((val3, val4)::foo) AS js_34
FROM   tbl
GROUP  BY val2;

Or use a subselect instead of the ROW expression. More verbose, but without type cast:

SELECT val2, jsonb_agg(to_jsonb((SELECT t FROM (SELECT val3, val4) t))) AS js_34
FROM   tbl
GROUP  BY val2;

to_jsonb() is an optional addition to add (insignificant) line breaks in the JSON document.

More in Craig's related answer:

• Using row_to_json() with nested joins

fiddle
_{Old sqlfiddle}

Astragalus

Thanks, but this returns the object with automatically generated keys: {"f1":"a","f2":"AA"}. How to rename f1 to value_three? – Waterer

@ehmicky: Right, if you want the column names, too, you need to cast the row to a well-known composite type. I'll add some more. – Astragalus

A sub-select, or even OP's cte could be a more convenient way for aliasing the row's column names. #13227642 – Simmer

@pozs: Good point (and link), I added a code variant accordingly. – Astragalus

As a style note, I think it's easier for newer folks when we put the updates at the top of the answer and the original notes below it with a note on what version they're applicable for. So long as you maintain the answer it's good forever, and the advice for older PostgreSQL's becomes less useful as time moves on. – Gad

@Evan: I agree. I don't bother to update all my old answers. But since this one needed an update anyway ... – Astragalus

this worked for my case: json_agg(to_json(items.*)) as "items" – Kailey

@ErwinBrandstetter Is there any way of doing this where the original type of a timestamp column is retained as a timestamp type rather than being converted to a string? If value_three in the example above is a timestamp, it gets converted to string in all of the options you suggested and the application needs to convert it back. While value_two in the example will be kept as a timestamp. Any way to have value_three handled like value_two? – Cypher

@user779159; No. JSON itself does not support timestamp as datatype. See: en.wikipedia.org/wiki/JSON#Data_types,_syntax_and_example and postgresql.org/docs/current/static/…. Your options are to store it as string (see: https://mcmap.net/q/28577/-ignoring-time-zones-altogether-in-rails-and-postgresql) or as number (see: https://mcmap.net/q/95341/-get-a-timestamp-from-concatenating-day-and-time-columns). – Astragalus

Hello,is there anyway to add DISTINCT in the following query:- json_agg( json_build_object ( 'id', ca.id, 'name',ca.name, 'url',ca.url) ) as authors – Heedful

@edwinBrandstetter, at the end of that series, a cast back to foo brings us full circle: SELECT val2, ( json_populate_recordset( NULL::foo /** cast as type / ,json_agg(to_json((SELECT t FROM (SELECT val3, val4) t)))) ).* / expand the composite record **/ FROM tbl GROUP BY val2; – Kennith

Answer 2

to_json with array_agg with composite type

begin;
create table  mytable(
id bigint, value_two text, value_three text, value_four  text);
insert into mytable(id,value_two, value_three,value_four)
values
 ( 1, 'a',       'A',           'AA'),
  (2, 'a'    ,     'A2'  ,       'AA2'),
  (3, 'b'  ,       'A3',         'AA3'),
 ( 4, 'a'   ,      'A4',          'AA4'),
  (5, 'b' ,        'A5',          'AA5');
commit;
create type mytable_type as (value_three text, value_four text);

---

select value_two,
       to_json( array_agg(row(value_three,value_four)::mytable_type))
from mytable
group by 1;

Jenness

Answer 3

use jsonb_agg and to_jsonb.

SELECT
    value_two,
    jsonb_agg(to_jsonb (t.*) - '{id,value_two}'::text[]) AS data
FROM
    mytable t
GROUP BY
    1
ORDER BY
    1;

based on manual reference

jsonb - text[] → jsonb

Deletes all matching keys or array elements from the left operand.

Jenness