how to simplify this multiple-CTE solution to this sql question? - sql

Question

DDL:
create table transactions
(
product_id int,
store_id int,
quantity int,
price numeric
);
DML:
insert into transactions values
(1, 1, 10, 2),
(2, 1, 5, 2),
(1, 2, 5, 4),
(2, 2, 2, 4),
(2, 3, 1, 20),
(1, 3, 1, 8),
(2, 4, 2, 10),
(1, 5, 2, 5),
(2, 5, 1, 3),
(2, 6, 4, 8);
I'm trying to find the top 3 products of the top 3 stores, both are based on sale amount. The solution I have is to use cte as below:
with cte as
(
select store_id, rank_store
from
(select
*,
dense_rank() over(order by sale desc) as rank_store
from
(select
store_id, sum(quantity * price) as sale
from transactions
group by 1) t) t2
where
rank_store <= 3
),
cte2 as
(
select
a.store_id, a.product_id,
sum(a.quantity * a.price) as sale_store_product
from
transactions as a
join
cte as b on a.store_id = b.store_id
group by
1, 2
order by
1, 2
),
cte3 as
(
select
*,
dense_rank() over (partition by store_id order by sale_store_product desc) as rank_product
from
cte2
)
select *
from cte3
where rank_product <= 3;
Here is the expected result:
Basically, the first cte is to get the top 3 stores based on sale amount, I use dense_rank() window function to handle tie cases. then the 2nd cte is to get the top 3 stores' products and their total sale amount. The last cte is to use dense_rank() window function again to rank the products in each stores based on their sale amount. then my last query is to get the top 3 products in each store based on the sale amount.
I'm wondering if this can be improved a bit since I feel three CTEs is kind of too complicated. Appreciate for sharing any solutions and ideas. Thanks.

Answer 1

I'm trying to find the top 3 products of the top 3 stores
How can this be done without aggregating the data twice -- once for store/products and once for stores? This is possible using window functions along with aggregation:
select sp.*
from (select sp.*,
dense_rank() over (order by store_sales, store_id) as store_seqnum
from (select t.store_id, t.product_id,
sum(quantity * price) as sp_sales,
sum(sum(quantity * price)) over (partition by store_id) as store_sales,
row_number() over (partition by t.store_id order by sum(quantity * price)) as sp_seqnum
from transactions t
group by t.store_id, t.product_id
) sp
) sp
where store_seqnum <= 3 and sp_seqnum <= 3;
The inner subquery calculates the product/store information. The next level ranks the stores -- notes that ties are broken using store_id.
Here is a db<>fiddle.