Question

I am running the following queries against a database:

CREATE TEMPORARY TABLE med_error_third_party_tmp
SELECT `med_error_category`.description AS category, `med_error_third_party_category`.error_count AS error_count
FROM
    `med_error_category` INNER JOIN `med_error_third_party_category` ON med_error_category.`id` = `med_error_third_party_category`.`category`
WHERE
    year = 2003
GROUP BY `med_error_category`.id;

The only problem is that when I create the temporary table and do a select * on it then it returns multiple rows, but the query above only returns one row. It seems to always return a single row unless I specify a GROUP BY, but then it returns a percentage of 1.0 like it should with a GROUP BY.

SELECT category, 
       error_count/SUM(error_count) AS percentage
  FROM med_error_third_party_tmp;

Here are the server specs:
Server version: 5.0.77
Protocol version: 10
Server: Localhost via UNIX socket

Does anybody see a problem with this that is causing the problem?

Answer 1

Standard SQL requires you to specify a GROUP BY clause if any column is not wrapped in an aggregate function (IE: MIN, MAX, COUNT, SUM, AVG, etc), but MySQL supports “hidden columns in the GROUP BY” — which is why:

SELECT category, 
       error_count/SUM(error_count) AS percentage
  FROM med_error_third_party_tmp;

…runs without error. The problem with the functionality is that because there’s no GROUP BY, the SUM is the SUM of the error_count column for the entire table. But the other column values are completely arbitrary – they can’t be relied upon.

This:

SELECT category, 
       error_count/(SELECT SUM(error_count)
                      FROM med_error_third_party_tmp) AS percentage
  FROM med_error_third_party_tmp;

…will give you a percentage on a per row basis — category values will be duplicated because there’s no grouping.

This:

  SELECT category, 
         SUM(error_count)/x.total AS percentage
    FROM med_error_third_party_tmp
    JOIN (SELECT SUM(error_count) AS total
            FROM med_error_third_party_tmp) x 
GROUP BY category

…will gives you a percentage per category of the sum of the categories error_count values vs the sum of the error_count values for the entire table.