How to aggregate based on other rows that share a key? - sql

Question

I have a table in the following format:
I feel like this should be simple but I'm struggling to come up with a performant query that can perform aggregations based on other rows with a shared key. For example, I want to sum the rows for a user with the key MediaLength but only if the rows with the key Score that share the event_id are greater than or equal to 3.
The result from a simple sum:
SELECT SUM(value::float) FROM data WHERE key = 'MediaLength' AND user_id = '9765f312-0d0b-4db0-b4c5-217eec81d7c3'
Result: 40
The result I am trying to achieve here is 15. In the table above you can see the rows are children of an event. I only want to sum the value column where key = 'MediaLength' and its sister row with key = 'Score' has value >= 3.
This is the query I have tried so far but it seems a bit messy and also doesn't work due to a more than one row returned by subquery error:
select
sum(value::float)
filter (where (
select d.value::float
from data d
where d.event_id = event_id
and d.key = 'Score'
) >= 3)
from data
where user_id = '9765f312-0d0b-4db0-b4c5-217eec81d7c3'
This is a simple example but in the future I would need to filter on potentially multiple other keys as well, so any advice on how to extend that is also hugely appreciated.

Answer 1

I only want to sum the value column where key = 'MediaLength' and its sister row with key = 'Score' has value >= 3.
SELECT sum(value::float) -- why the cast?
FROM data d
WHERE user_id = '9765f312-0d0b-4db0-b4c5-217eec81d7c3'
AND key = 'MediaLength'
AND EXISTS (
SELECT FROM data ds
WHERE ds.event_id = d.event_id
AND ds.user_id = d.user_id -- !
AND ds.key = 'Score'
AND ds.value >= 3
);
Here, rows with key = 'MediaLength' qualify if any sister passes the filter. (There may be more sisters failing the test.)
If there can only ever be a single qualifying sister row (enforced by a unique constraint / index?), a self-join is a bit simpler:
SELECT sum(value::float)
FROM data d
JOIN data ds USING (event_id, user_id)
WHERE d.user_id = '9765f312-0d0b-4db0-b4c5-217eec81d7c3'
AND d.key = 'MediaLength'
AND ds.key = 'Score'
AND ds.value >= 3;
The self-join would produce multiple result rows for multiple qualifying sister rows.
At its core, this can be cast as relational-division problem. Especially, since ...
in the future I would need to filter on potentially multiple other keys as well
See:
How to filter SQL results in a has-many-through relation