RANK OVER PARTITION BY 在聚合函数上

Question

Here is a set of 3 tables:

Orders (ORDER_ID, CUSTOMER_ID, ORDER_DATE)

Order detail (ORDER_ID, PRODUCT_ID, PRICE, UNITS)

Customers (CUSTOMER_ID, SIGNUP_DATE, COUNTRY)

I need to identify the top 3 months, by revenue, for each country. This is the query that I thought of:

SELECT c.country
    , MONTH(o.order_date) as _Month
    , SUM(od.price * od.units) AS revenue
    , RANK() OVER (PARTITION BY country, _Month ORDER BY revenue) AS Rank  
FROM orders o 
LEFT JOIN customers c ON o.customer_id = c.customer_id  
JOIN order_detail od 
WHERE o.order_id = od.order_id GROUP BY country, Month
HAVING Rank <= 3

Will this query work because I am not sure if SUM(od.price * od.units) as revenue can be used in the RANK() OVER function or not since it's an aggregate function?

Answer 1

In databases that support window functions, this would normally be written as:

SELECT cm.*
FROM (SELECT c.country, MONTH(o.order_date) as Month, SUM(od.price * od.units) AS revenue,
             RANK() OVER (PARTITION BY country, MONTH(o.order_date) ORDER BY SUM(od.price * od.units) ) AS Rank 
      FROM orders o LEFT JOIN
           customers c
           ON o.customer_id = c.customer_id  JOIN
           order_detail od 
           ON o.order_id = od.order_id
      GROUP BY country, Month
     ) cm
HAVING Rank <= 3;

Answer 2

This is meant to be more Comment than Answer, since I can format a bit prettier down here, but the typical query logical order of operations goes something along the lines of:

FROM is built first. JOIN s will be applied. WHERE will cause the above tables to be filtered GROUP BY and aggregates will be applied. HAVING is applied. WINDOW FUNCTION s will be calculated. These operate outside of the ORDER BY SELECT will be applied. DISTINCT will be applied. ORDER BY will be applied. TOP / LIMIT will be applied.

In general, anything that happens in a step will be available to the steps following it (ie a SELECT alias can be used in the ORDER BY but not in the GROUP BY . But in MySQL, the HAVING can use aliases defined in the SELECT . Of course, the optimizer can always decide to do its own thing and may choose a different order, but this is what usually happens in most flavors of SQL.

A couple of references I have are MS SQL related, but, for the most part, still apply to other types:

https://blog.sqlauthority.com/2009/10/10/sql-server-download-logical-query-processing-poster/

Confused about Itzik Ben-Gan's Logical Query Processing order in his SQL Server 2005 book and SQL Server 2008 book

https://accessexperts.com/wp-content/uploads/2015/07/Logical-Query-Processing-Poster.pdf

https://www.brentozar.com/archive/2015/07/logical-query-processing/

Answer 3

I don't think Gordon Linoff's answer is right.

(1) You can not directly call the alias in the Rank() function, which has been mentioned by Gordon
(2) For your use case, you want to rank the months within each country, the rank part should be

RANK() OVER (PARTITION BY country ORDER BY SUM(od.price * od.units) ) AS Rank