Oracle join a table using a variable field - sql

Question

I'm trying to get the rent rate for which a cage is to be charged with for each month. Here's the logic behind how to get it:
If the rate is defined in the cage table use that table's foreign key to link to the rate table.
If the rate is defined in the rack table use that table's foreign key to link to the rate table.
If the rate is defined in the room table use that table's foreign key to link to the rate table.
If all of the above is false return $0.00 (free rent).
So here are the tables for you to visually understand what I currently have.
Cages
id cage_number rate_id rack_id room_id
----------------------------------------------------
1 FG-12 600 1 1
2 FG-13 600 1 1
3 FG-14 NULL 2 1
4 FG-15 NULL 2 3
Racks
id rack_number rate_id
-----------------------------
1 BN-806 700
2 BN-971 NULL
Rooms
id room_number rate_id
-----------------------------
1 UM-100 800
2 UM-150 900
3 UM-200 NULL
Rates
id rate_id effective_on rate_value
------------------------------------------
1 600 01-Apr-2015 0.67
2 600 01-Oct-2016 0.80
3 700 01-Jan-2016 0.56
4 800 01-Jul-2012 0.61
5 800 01-Dec-2015 0.85
6 900 01-Feb-2015 0.75
7 900 01-Mar-2015 0.79
8 900 01-Apr-2015 0.90
So as you can see there are multiple rows for each rate_id because the price changes from time to time. I already have a query to get the correct rate for each rate_id by checking the effective_on field:
SELECT *
FROM rates r
WHERE r.id = (
SELECT MAX(rr.id) AS curr
FROM rates rr
WHERE rr.rent_id = r.rent_id
AND TRUNC(rr.effective_on) <= TRUNC(sysdate)
)
Let's go back to what my main problem is. I need to join the rate table using the rate_id located on 3 different tables.
Current Solution
My current approach is to join the rate table 3 times using 3 different aliases but since there could be multiple rows for the rate, all 3 instances could have multiple rows each which results in a lot of un-needed rows.
FROM cages c
LEFT JOIN racks rk ON rk.id = c.rack_id
LEFT JOIN rooms rm ON rm.id = c.room_id
LEFT JOIN rates cr ON cr.rate_id = c.rate_id
LEFT JOIN rates rkr ON rkr.rate_id = rk.rate_id
LEFT JOIN rates rmr ON rmr.rate_id = rm.rate_id
Attempted (but miserably failed)
I was thinking of a solution where I would only need to join the rate table once to fix that problem but when I tried it - it failed (as expected).
FROM cages c
LEFT JOIN racks rk ON rk.id = c.rack_id
LEFT JOIN rooms rm ON rm.id = c.room_id
CASE
WHEN c.rate_id IS NOT NULL
THEN LEFT JOIN rates r ON r.rate_id = c.rate_id
WHEN rk.rate_id IS NOT NULL
THEN LEFT JOIN rates r ON r.rate_id = rk.rate_id
WHEN rm.rate_id IS NOT NULL
THEN LEFT JOIN rates r ON r.rate_id = rm.rate_id
END

Answer 1

You can't use a case expression to decide which join condition to use, but you can use a case expression on the right-hand side of the condition, to decide what you're trying to match the rates table's ID against:
FROM cages c
LEFT JOIN racks rk ON rk.id = c.rack_id
LEFT JOIN rooms rm ON rm.id = c.room_id
LEFT JOIN rates r ON r.rate_id =
CASE
WHEN c.rate_id IS NOT NULL THEN c.rate_id
WHEN rk.rate_id IS NOT NULL THEN rk.rate_id
WHEN rm.rate_id IS NOT NULL THEN rm.rate_id
END
But it's simpler to use coalesce() for the outer join condition on the rates table:
LEFT JOIN rates r ON r.rate_id = COALESCE(c.rate_id, rk.rate_id, rm.rate_id)
Or more usefully on a subquery against the rates table that finds the most recent rate for each ID:
select cages.cage_number,
rates.rate_id as rate_id,
nvl(rates.rate_value, 0)
from cages
left join racks on racks.id = cages.rack_id
left join rooms on rooms.id = cages.room_id
left join (
select rate_id,
max(rate_value) keep (dense_rank last order by effective_on) as rate_value
from rates
where effective_on <= trunc(sysdate)
group by rate_id
) rates on rates.rate_id = coalesce(cages.rate_id, racks.rate_id, rooms.rate_id);
CAGE_ RATE_ID COALESCE(RATES.RATE_VALUE,0)
----- ---------- ----------------------------
FG-13 600 .8
FG-12 600 .8
FG-14 800 .85
FG-15 0
The subquery ignores rates that start in the future, as you were already doing; and uses keep dense_rank last to avoid the self-join.
That subquery is then used as an inline view (aliased back to 'rates' for simplicity). The join condition uses coalesce to join on the cage's rate ID if it's set; then the rack ID; then the room ID. Finally there's another coalesce to set the final selected value to zero if there was no match at all.
As #mathguy pointed out, this may end up doing unnecessary work - if the cage has a rate ID there's no point looking at the other tables for that cage. Depending on how selective the data is that may not matter too much, but if it does affect performance you could consider modifying the outer join conditions to eliminate some of them:
from cages
left join racks on racks.id = cages.rack_id
and cages.rate_id is null
left join rooms on rooms.id = cages.room_id
and cages.rate_id is null
and racks.rate_id is null
left join (
...

Answer 2

Here's a different way to get the proper rate_id for each cage. Depending on how many nulls you have in the different tables, it may be more efficient (faster) than doing the left joins on the entire tables. The idea is to collect the rate_id from the cages table when it is not null, and only when it is null to join to the racks table, and only when that doesn't work either, to join only the remaining rows to the rooms table.
I am not showing how to "connect" this to the rates table - either approach in Alex's answer will do. (I would use the second version, again for efficiency.)
with
prep ( id, cage_number, rate_id, room_id ) as (
select c.id, c.cage_number, rk.rate_id, c.room_id
from cages c left outer join racks rk on c.rack_id = rk.id
where c.rate_id is null
)
select id, cage_number, rate_id
from cages
where rate_id is not null
union all
select id, cage_number, rate_id
from prep
where rate_id is not null
union all
select p.id, p.cage_number, rm.rate_id
from prep p left outer join rooms rm on p.room_id = rm.id
where p.rate_id is null
;