Question

Hope everyone is doing well.

I have the following output…

+---------+--------------+--------------+-----------+---------+----------+
| ord_num | signoff_date | program_name | prod_desc | tx_comp | priority |
+---------+--------------+--------------+-----------+---------+----------+
| 1234567 | 2012-08-12   | ilearn       | run       |       1 |        1 |
| 1234567 | 2012-08-12   | ilearn       | plan      |       1 |        1 |
| 1234568 | 2012-08-12   | other        | run       |       1 |        1 |
| 1234569 | 2012-08-12   | other        | run       |       0 |        1 |
+---------+--------------+--------------+-----------+---------+----------+

What I would like to do is SUM the tx_comp column once per unique “ord_num”.

Now I cant use GROUP BY ord_num as I also do a sum on the type of tasks.

Its like I need to know what the previous ord_num was then sum if different?

Any ideas would be greatly appreciated.

Thanks.

* EDIT *

SELECT 
    signoff_date,
    SUM(IF(prod_desc = 'run', 1, 0)) AS run,
    SUM(IF(prod_desc = 'plan', 1, 0)) AS plan,
    SUM(tx_comp) AS tx_comp 
FROM
(
    SELECT 
        ord_num,
        signoff_date,
        program_name,
        prod_desc,
        tx_comp,
        priority 
    FROM
        test.orders 
        LEFT JOIN test.tx_comp USING (ord_num)
) AS grp

Obviously not the desired output

+--------------+------+------+---------+
| signoff_date | run  | plan | tx_comp |
+--------------+------+------+---------+
| 2012-08-12   |    3 |    1 |       3 |
+--------------+------+------+---------+

I am after…

+--------------+------+------+---------+
| signoff_date | run  | plan | tx_comp |
+--------------+------+------+---------+
| 2012-08-12   |    3 |    1 |       2 |
+--------------+------+------+---------+

Answer 1

If the value of tx_comp is always 1 or zero, then we can leverage COUNT(), which may give us more options. For instance, we can count the distinct ord_num where tx_comp is 1:

COUNT(distinct IF(tx_comp, ord_num, NULL))

Which gives me a final query of:

SELECT signoff_date,
  SUM(IF(prod_desc = 'run', 1, 0)) AS run,
  SUM(IF(prod_desc = 'plan', 1, 0)) AS plan,
  COUNT(distinct IF(tx_comp, ord_num, NULL)) as tx_comp
FROM 
  test.orders
  JOIN test.tx_comp USING (ord_num)
  GROUP BY signoff_date

And there is no need for the subquery in this case. (edit: updated for your JOIN)

I have tested this with your sample data; the only dependency is on the semantic nature of tx_comp. You have been saying “SUM”, and this assumes that the value will be at most 1 (I understand it to be a boolean flag, and in a comment on another answer you mentioned MAX(tx_comp) returning 1, so I think we’re good).