Summing cost by id that appears on multiple rows - sql

Question

SOLUTION
I solved it by simple doing the following.
SELECT table_size, sum(cost) as total_cost, sum(num_players) as num_players
FROM
(
SELECT table_size, cost, sum(tp.uid) as num_players
FROM tournament as t
LEFT JOIN takes_part AS tp ON tp.tid = t.tid
LEFT JOIN users as u on u.uid = tp.tid
JOIN attributes as a on a.aid = t.attrId
GROUP BY t.tid
) as res
GROUP BY table_size
I wasn't sure it would work, what with the other aggregate functions that I had to use in my real sql, but it seems to be working ok. There may be problems in the future if I want to do other kind of calculations, for instance do a COUNT(DISTINCT tp.uid) over all tournaments. Still, in this case that is not all that important so I am satisfied for now. Thank you all for your help.
UPDATE!!!
Here is a Fiddle that explains the problem:
http://www.sqlfiddle.com/#!2/e03ff/7
I want to get:
table_size | cost
-------------------------------
5 | 110
8 | 80
OLD POST
I'm sure that there is an easy solution to this that I'm just not seeing, but I can't seem to find a solution to it anywhere. What I'm trying to do is the following:
I need to sum 'costs' per tournament in a system. For other reasons, I've had to join with lots of other tables, making the same cost appear on multiple rows, like so:
id | name | cost | (hidden_id)
-----------------------------
0 | Abc | 100 | 1
1 | ASD | 100 | 1
2 | Das | 100 | 1
3 | Ads | 50 | 2
4 | Ads | 50 | 2
5 | Fsd | 0 | 3
6 | Ads | 0 | 3
7 | Dsa | 0 | 3
The costs in the table above are linked to an id value that is not necessary selected in by the SQL (this depends on what the user decides at runtime). What I want to get, is the sum 100+50+0 = 150. Of course, if I just use SUM(cost) I will get a different answer. I tried using SUM(cost)/COUNT(*)*COUNT(tourney_ids) but this only gives correct result under certain circumstances. A (very) simple form of query looks like this:
SELECT SUM(cost) as tot_cost -- This will not work as it sums all rows where the sum appears.
FROM t
JOIN ta ON t.attr_id = ta.toaid
JOIN tr ON tr.toid = t.toid -- This row will cause multiple rows with same cost
GROUP BY *selected by user* -- This row enables the user to group by several attributes, such as weekday, hour or ids of different kinds.
UPDATE. A more correct SQL-query, perhaps:
SELECT
*some way to sum cost*
FROM tournament AS t
JOIN attribute AS ta ON t.attr_id = ta.toaid
JOIN registration AS tr ON tr.tourneyId = t.tourneyId
INNER JOIN pokerstuff as ga ON ta.game_attr_id = ga.gameId
LEFT JOIN people AS p ON p.userId = tr.userId
LEFT JOIN parttaking AS jlt ON (jlt.tourneyId = t.tourneyId AND tr.userId = jlt.userId)
LEFT JOIN (
SELECT t.tourneyId,
ta.a - (ta.b) - sum(c)*ta.cost AS cost
FROM tournament as t
JOIN attribute as ta ON (t.attr_id = ta.toaid)
JOIN registration tr ON (tr.tourneyId = t.tourneyId)
GROUP BY t.tourneyId, ta.b, ta.a
) as o on t.tourneyId = o.tourneyId
AND whereConditions
GROUP BY groupBySql
Description of the tables
tournament (tourneyId, name, attributeId)
attributes (attributeId, ..., gameid)
registration (userId, tourneyId, ...)
pokerstuff(gameid,...)
people(userId,...)
parttaking(userId, tourneyId,...)
Let's assume that we have the following (cost is actually calculated in a subquery, but since it's tied to tournament, I will treat it as an attribute here):
tournament:
tourneyId | name | cost
1 | MyTournament | 50
2 | MyTournament | 80
and
userId | tourneyId
1 | 1
2 | 1
3 | 1
4 | 1
1 | 2
4 | 2
The problem is rather simple. I need to be able to get the sum of the costs of the tournaments without counting a tournament more than once. The sum (and all other aggregates) will be dynamically grouped by the user.
A big problem is that many solutions that I've tried (such as SUM OVER...) would require that I group by certain attributes, and that I cannot do. The group by-clause must be completely decided by the user. The sum of the cost should sum over any group-by attributes, the only problem is of course the multiple rows in which the sum appears.
Do anyone of you have any good hints on what can be done?

Answer 1

Try the following:
select *selected by user*, sum(case rownum when 1 then a.cost end)
from
(
select
*selected by user*, cost,
row_number() over (partition by t.tid) as rownum
FROM t
JOIN ta ON t.attr_id = ta.toaid
JOIN tr ON tr.toid = t.toid
) a
group by *selected by user*
The row_number is used to number each row with the same tournament row. When suming the costs we only consider those rows with a rownum of 1. All other rows are duplicates of this one with regards to the costs.
In terms of the fiddle:
select table_size, sum(case rownum when 1 then a.cost end)
from
(
SELECT
table_size, cost,
row_number() over (partition by t.tid) as rownum
FROM tournament as t
LEFT JOIN takes_part AS tp ON tp.tid = t.tid
LEFT JOIN users as u on u.uid = tp.tid
JOIN attributes as a on a.aid = t.attrId
) a
group by table_size

Answer 2

As the repeated costs are the same each time you can average them by their hidden id and do something like this:
WITH MrTable AS (
SELECT DISTINCT hidden_id, AVG(cost) OVER (PARTITION BY hidden_id) AS cost
FROM stuff
)
SELECT SUM(cost) FROM MrTable;

Answer 3

(Updated) Given that the cost currently returned is the total cost per tournament, you could include a fractional value of cost on each line of an inner select, such that the total of all those values adds up to the total cost (allowing for the fact that each given tournament's values may be appearing multiple times), then sum that fractional cost in your outer select, like so:
select table_size, sum(frac_cost) as agg_cost from
(SELECT a.table_size , cost / count(*) over (partition by t.tid) as frac_cost
FROM tournament as t
LEFT JOIN takes_part AS tp ON tp.tid = t.tid
LEFT JOIN users as u on u.uid = tp.uid
JOIN attributes as a on a.aid = t.attrId) sq
GROUP BY table_size
SQLFiddle here.