SQL 当没有 where 或 having 子句时，服务器查询聚合不正确

Question

Edit: I've tried the first two solutions, but am still having this issue of the query returning correct results when looking at a single customer with a WHERE clause, but incorrectly for the same customer without it. How could this be happening? What is going on under the hood that could lead to this?

I am building a query to join and aggregate customer information on a big table, so I am starting out building the query with a where clause for a single customer to make sure the logic is working before implementing it on the population of customers.

The tables I'm joining look something like this:

Table A:

| customer | order_id |
----------------------
| abc      | 1       |
| abc      | 2       |
| xyz      | 3       |
| xyz      | 4       |
| xyz      | 5       |
| xyz      | 6       |
...

Table B:

| order_id | return_date   |
----------------------------
| 1        |       Mon     |
| 3        |       Tues    |
| 5        |       Wed     |
...

I need to aggregate these by the customer name and essentially count the number of times their info appears in each table.

So the query looks something like this:

SELECT 
  a.customer as customer_name
  ,COUNT(DISTINCT(a.order_id)) as total_orders
  ,COUNT(DISTINCT(B.order_id)) as num_returns
FROM B

RIGHT JOIN (
  SELECT 
    customer
    order_id
  FROM A
  ) as a

ON B.order_id = a.order_id
WHERE customer = 'xyz'
GROUP BY a.customer

This works perfectly when the where clause is present (also works with a HAVING customer = 'xyz' after the group by instead) But when I remove the where clause to apply this to the population of customers, the results are completely incorrect. How can I fix this to work for the population?

Answer 1

This query should work:

SELECT a.customer as customer_name,
       COUNT(DISTINCT a.order_id) as total_orders,
       COUNT(DISTINCT B.order_id) as num_returns
FROM A LEFT JOIN
     B
     ON B.order_id = a.order_id
WHERE a.customer = 'xyz'
GROUP BY a.customer;

If xyz has no rows in A , then this returns no rows.

Answer 2

I would recommend pre-aggregation on b , and a left join :

select a.customer, count(*) total_orders, coalesce(sum(b.num_returns), 0) num_returns
from a
left join (
    select order_id, count(*) num_returns
    from b
    group by order_id
) b on b.order_id = a.order_id
group by a.customer

The results are consistent, regardless of whether a where clause is used or not. Note that this assumes no duplicate (customer_id, order_id) in a , as showned in your sample data.

A lateral join would also do:

select a.customer, count(*) total_orders, sum(b.num_returns) num_returns
from a
cross apply (
    select count(*) num_returns
    from b
    where b.order_id = a.order_id
) b
group by a.customer