Getting the wrong values after SUM() query - sql

Question

I'm executing the following query:
SELECT
MAX(table1.id) as id,
clients.Username as account,
table1.clientid,
SUM(table1.symbols) as symbols,
SUM(table1.tickets) as tickets,
SUM(table1.cash) as cash,
(SUM(CASE WHEN table2.memo = 'Withdraw' THEN amount ELSE 0 END)) AS withdraw,
(SUM(CASE WHEN table2.memo = 'Depos' THEN amount ELSE 0 END)) AS depos,
FROM
table1
LEFT JOIN
(
clients
LEFT JOIN
table2
ON
clients.Fidx = table2.clientid
AND
table2.date >= '01-09-2016'
AND
table2.date <= '01-09-2017'
)
ON
clients.Fidx = table1.clientid
WHERE
table1.tradedate >= '01-09-2016'
AND
table1.tradedate <= '01-09-2017'
GROUP BY
table1.clientid, clients.Username, table2.clientid
ORDER BY
clients.Username;
And I want to get a simple result table combined of three tables:
+---------+--------+---------+
| account |withdraw| depos |
+---------+--------+---------+
| adaf | 300 | 0 |
| rich | 1000 | 355 |
| call | 0 | 45 |
| alen | 0 | 0 |
| courney| 0 | 106 |
| warren | 0 | 0 |
+---------+--------+---------+
What's the problem? - I'm getting the wrong values in the result table. Exactly in withdraw and depos. They're in 4 times more, than they should be. For example, for some client SUM(depos) should be 500, but in my result table this value gets 2000. I guess, the problem is in a GROUP BY method, cause when I'm executing the following query, the result seems to look OK:
SELECT clientid, SUM(case when memo = 'Withdraw' then amount else 0 end) as withdraw, SUM(case when memo = 'Depos' then amount else 0 end) as depos
from clients
LEFT JOIN
table2
ON
clients.Fidx = table2.clientid
WHERE table2.date >= '01-09-2016' and table2.date<='01-09-2017' GROUP BY clientid ORDER BY clientid;
What can be a reason of such a wrong result? I'm in trouble and need your help, guys.

Answer 1

You pretty much answered this yourself with the bottom query. You need to do the summing before the join, in a derived table or subquery, just like you have in the bottom query. This will ensure you join on a many-to-one relationship, instead of a many-to-many, which must be causing your current duplication (or 'multiplied' sums).
SELECT
MAX(table1.id) as id,
clients.Username as account,
table1.clientid,
SUM(table1.symbols) as symbols,
SUM(table1.tickets) as tickets,
SUM(table1.cash) as cash,
withdraw,
depos,
FROM
table1
LEFT JOIN
(SELECT clientid,
SUM(case when memo = 'Withdraw' then amount else 0 end) as withdraw,
SUM(case when memo = 'Depos' then amount else 0 end) as depos
FROM clients
LEFT JOIN table2 ON clients.Fidx = table2.clientid
AND table2.date >= '01-09-2016'
AND table2.date<='01-09-2017'
GROUP BY clientid) clients
ON
clients.clientID = table1.clientid
WHERE
table1.tradedate >= '01-09-2016'
AND
table1.tradedate <= '01-09-2017'
GROUP BY
table1.clientid, clients.Username, table2.clientid, clients.depos, clients.withdraw
ORDER BY
clients.Username;
Further example:
Table1
id | someInfo
1 | a
1 | b
1 | c
Table2
id | value
1 | 5
1 | 10
This query:
SELECT t1.id, SUM(t2.Value)
FROM table1 t1
JOIN table2 t2 on t1.id = t2.id --This will be many-to-many
GROUP BY t1.id
Will result in this:
Results
id | value
1 | 45 --sum of 45 because the `table2` values are triplicated from the join
Where this query:
SELECT DISTINCT t1.id, Value
FROM table1 t1
JOIN (SELECT id, SUM(Value) value
FROM table2
GROUP BY id) t2 on t1.id = t2.id --This will be many-to-one
Will result in this:
Results
id | value
1 | 15 --sum of 15 because the `table2` values are not triplicated from the join

Answer 2

Aggregate before joining:
select
t1.id,
c.Username as account,
c.clientid,
t1.symbols,
t1.tickets,
t1.cash,
coalesce(t2.withdraw, 0) as withdraw,
coalesce(t2.depos, 0) as depos
from clients c
join
(
select
clientid,
max(id) as id,
sum(symbols) as symbols,
sum(tickets) as tickets,
sum(cash) as cash
from table1
where date >= '20160901' and date <= '20170901'
group by clientid
) t1 on t1.clientid = c.fidx
left join
(
select
clientid,
sum(case when memo = 'Withdraw' then amount end) as withdraw,
sum(case when memo = 'Depos' then amount end) as depos
from table2
where date >= '20160901' and date <= '20170901'
group by clientid
) t2 on t2.clientid = c.fidx;