How to capture rows that match an aggregate - sql

Question

Say I have a table with (pseudocode):
TABLE Order
(
orderid int,
type int,
price NUMERIC(18,2),
)
Now I want to list those orders whose price matches the maximum price for a particular order type.
I start with the following, giving me the max price per order type:
SELECT type, MAX(price)
FROM Order
GROUP BY type
Now I know the maximum price by type. However, I want to, as efficiently as possible, get a result set of the actual orders whose price is that maximum price, instead of just the type/MAX(price).
The table is very large with potentially tens of millions of rows, so efficiency is key here (assuming proper indexes are in place, of course, such as on the type column in this case).
I start with something like:
SELECT orderid, price
FROM Order AS O
WHERE O.price=(SELECT MAX(O2.price)
FROM Order AS O2
WHERE O2.type=O.type)
It's not particularly fast, but it does the job.
Then I realize that orders appear multiple times in this table, because it's actually a denormalized order history table and it really looks more like:
TABLE Order
(
id int, -- This is just an identity column - the surrogate key
orderid int, -- multiple records exist for the same
-- orderid with different update times
type int,
price NUMERIC(18,2),
updatetime DATETIME2(3)
)
So, what I want is actually the latest version of those orders based on updatetime whose price matches the maximum price for their particular type. This is my question.
Extending:
SELECT *
FROM Order AS O
WHERE O.price=(SELECT MAX(O2.price)
FROM Order AS O2
WHERE O2.type=O.type)
..., to handle the new requirement seems like a mess waiting to happen. So I was wondering a good efficient (and hopefully readable) solution to the new requirements would be.
Based on Gordon's suggestion of:
select o.*
from (select o.*,
row_number() over (partition by type, price order by updatetime desc) as seqnum
from (select o.*, max(o.price) over (partition by type) as maxprice,
from Orders o
) o
where price = maxprice
) o
where seqnum = 1;
I have come up with the following query, with comments added to describe my thought process. The comments should of course be read from the innermost
query outward:
SELECT *
FROM
(
-- We want the max price for each order type, but we only want to
-- use the latest version of each order (i.e., seqnum=1). So, we
-- partition by type/seqnum, calculate the max price for each
-- partition and the only use the max prices from the seqnum=1
-- partitions for each type via the WHERE clause in the outer query
SELECT *,
MAX(price) OVER (PARTITION BY type, seqnum) AS maxprice
FROM
(
-- We only want to examine the latest version of each order.
-- BTW, the order price can change between versions.
-- So, let's start by marking the latest version of each order
-- with seqnum=1 which we will use as a "filter in" clause later
SELECT *,
row_number() OVER (PARTITION BY orderid
ORDER BY updatetime DESC) AS seqnum
) AS O
WHERE seqnum=1; -- Discard all but the latest versions of orders
) AS O
WHERE price=maxprice
I am not sure if this is correct though, because it is quite complicated...

Answer 1

Use window functions. Your original query can be written as:
select o.*
from (select o.*, max(o.price) over (partition by type) as maxprice
from Orders o
) o
where price = maxprice;
If you want the most recent order for the price:
select o.*
from (select o.*, max(o.price) over (partition by type) as maxprice,
row_number() over (partition by type, price order by updatetime desc) as seqnum
from Orders o
) o
where price = maxprice and seqnum = 1;
EDIT:
This would be a bit more efficient with an index on Orders(type, price, updatetime). You can also try writing this as:
select o.*
from (select o.*,
row_number() over (partition by type, price order by updatetime desc) as seqnum
from (select o.*, max(o.price) over (partition by type) as maxprice,
from Orders o
) o
where price = maxprice
) o
where seqnum = 1;
This may greatly reduce the data being used for the second analytic function.