PostgreSQL left join query object array aggregate

select
    person_id,
    jsonb_agg(to_jsonb(pet) - 'owner_id'),
    person_name
from person
left join pet on person_id = owner_id
group by person_id;

 person_id |                                 jsonb_agg                                  | person_name 
-----------+----------------------------------------------------------------------------+-------------
         1 | [{"pet_id": 1, "pet_name": "Fluffy"}, {"pet_id": 2, "pet_name": "Buster"}] | John
         2 | [{"pet_id": 3, "pet_name": "Doggy"}]                                       | Jill
         3 | [null]                                                                     | Mary
(3 rows)

Db<>fiddle.

Postgres' built in JSON and aggregation functions can handle your requirement:

select
    json_agg(row_to_json(row(p2.pet_id::text, p2.pet_name))) as json,
    p1.person_name
from person p1
left join pet p2
    on p1.person_id = p2.owner_id
group by
    p1.person_name;

Use LEFT JOIN LATERAL and aggregate in the subquery:

SELECT p.person_id, COALESCE(pet.pets, '[]') AS pets, p.person_name
FROM   person p
LEFT   JOIN LATERAL (
   SELECT json_agg(json_build_object('pet_id', pet.pet_id
                                   , 'pet_name', pet.pet_name)) AS pets
   FROM   pet
   WHERE  pet.owner_id = p.person_id
   ) pet ON true
ORDER  BY p.person_id;  -- optional, Q suggests ordered results

db<>fiddle here

This way you do not need to aggregate results from the outer query. Simpler and cleaner when your outer query is more complex than the example in the question. When aggregating multiple related tables, it even becomes a necessity:

• Multiple array_agg() calls in a single query
• Two SQL LEFT JOINS produce incorrect result

It is also typically much faster when there are selective predicates on the outer table person - which is the typical use case.

Make sure there is an index on pet(owner_id) to make it fast.
Or even one on pet(owner_id, pet_id, pet_name) or pet(owner_id) INCLUDE (pet_id, pet_name) in Postgres 11 or later, if your row isn't wide like in your example, and if you get index-only scans out of it.

Oh, and use json_build_object() to preserve attribute names for arbitrary selections:

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

Related:

• What is the difference between a LATERAL JOIN and a subquery in PostgreSQL?

demo:db<>fiddle

select
    COALESCE(
        json_agg(row_to_json(row(p2.pet_id::text, p2.pet_name))) FILTER (WHERE pet_id IS NOT NULL), 
       '[]'
    ) as json,
    p1.person_name
from person p1
left join pet p2
    on p1.person_id = p2.owner_id
group by
    p1.person_name;

1. FILTER clause to filter out NULL values. That creates a NULL value for Mary.
2. If you want to add an empty JSON array: Use COALESCE, which replaces NULL with a default value