Why count ignores grouping by - sql

Question

I don't understand why my query doesn't group results of count by the column I specified. Instead it counts all occurrences of outcome_id in the 'un' subtable.
What am I missing there?
The full structure of my sample database and the query I tried are here:
https://www.db-fiddle.com/f/4HuLpTFWaE2yBSQSzf3dX4/4
CREATE TABLE combination (
combination_id integer,
ticket_id integer,
outcomes integer[]
);
CREATE TABLE outcome (
outcome_id integer,
ticket_id integer,
val double precision
);
insert into combination
values
(510,188,'{52,70,10}'),
(511,188,'{52,56,70,18,10}'),
(512,188,'{55,70,18,10}'),
(513,188,'{54,71,18,10}'),
(514,189,'{52,54,71,18,10}'),
(515,189,'{55,71,18,10,54,56}')
;
insert into outcome
values
(52,188,1.3),
(70,188,2.1),
(18,188,2.6),
(56,188,2),
(55,188,1.1),
(54,188,2.2),
(71,188,3),
(10,188,0.5),
(54,189,2.2),
(71,189,3),
(18,189,2.6),
(55,189,2)
with un AS (
SELECT combination_id, unnest(outcomes) outcome
FROM combination c JOIN
outcome o
on o.ticket_id = c.ticket_id
GROUP BY 1,2
)
SELECT combination_id, cnt
FROM (SELECT un.combination_id,
COUNT(CASE WHEN o.val >= 1.3 THEN 1 END) as cnt
FROM un JOIN
outcome o
on o.outcome_id = un.outcome
GROUP BY 1
) x
GROUP BY 1, 2
ORDER BY 1
Expected result should be:
510 2
511 4
512 2
513 3
514 4
515 4

Answer 1

Assuming, you have these PK constraints:
CREATE TABLE combination (
combination_id integer PRIMARY KEY
, ticket_id integer
, outcomes integer[]
);
CREATE TABLE outcome (
outcome_id integer
, ticket_id integer
, val double precision
, PRIMARY KEY (ticket_id, outcome_id)
);
and assuming this objective:
For each row in table combination, count the number of array elements in outcomes for which there is at least one row with matching outcome_id and ticket_id in table outcome - and val >= 1.3.
Assuming above PK, this burns down to a much simpler query:
SELECT c.combination_id, count(*) AS cnt
FROM combination c
JOIN outcome o USING (ticket_id)
WHERE o.outcome_id = ANY (c.outcomes)
AND o.val >= 1.3
GROUP BY 1
ORDER BY 1;
This alternative might be faster with index support:
SELECT c.combination_id, count(*) AS cnt
FROM combination c
CROSS JOIN LATERAL unnest(c.outcomes) AS u(outcome_id)
WHERE EXISTS (
SELECT
FROM outcome o
WHERE o.outcome_id = u.outcome_id
AND o.val >= 1.3
AND o.ticket_id = c.ticket_id -- ??
)
GROUP BY 1
ORDER BY 1;
Plus, it does not require the PK on outcome. Any number of matching rows still count as 1, due to EXISTS.
db<>fiddle here
As always, the best answer depends on the exact definition of setup and requirements.

Answer 2

A simpler version of #forpas answer:
-- You don't need to join to outcomes in the "with" statement.
with un AS (
SELECT combination_id, ticket_id, unnest(outcomes) outcome
FROM combination c
-- no need to join to outcomes here
GROUP BY 1,2,3
)
SELECT combination_id, cnt FROM
(
SELECT un.combination_id,
COUNT(CASE WHEN o.val >= 1.3 THEN 1 END) as cnt
FROM un
JOIN outcome o on o.outcome_id = un.outcome
and o.ticket_id = un.ticket_id
GROUP BY 1
)x
GROUP BY 1,2
ORDER BY 1
As others have pointed out, the expected result for 514 should be 3 based on your input data.
I'd also like to suggest that using full field names in the group by and order by clauses makes queries easier to debug and maintain going forward.

Answer 3

You need to join on ticket_id also:
with un AS (
SELECT c.combination_id, c.ticket_id, unnest(c.outcomes) outcome
FROM combination c JOIN outcome o
on o.ticket_id = c.ticket_id
GROUP BY 1,2,3
)
SELECT combination_id, cnt
FROM (SELECT un.combination_id, un.ticket_id,
COUNT(CASE WHEN o.val >= 1.3 THEN 1 END) as cnt
FROM un JOIN outcome o
on o.outcome_id = un.outcome and o.ticket_id = un.ticket_id
GROUP BY 1,2
) x
GROUP BY 1, 2
ORDER BY 1
See the demo.
Results:
> combination_id | cnt
> -------------: | --:
> 510 | 2
> 511 | 4
> 512 | 2
> 513 | 3
> 514 | 3
> 515 | 4