Question

I have a table that tracks contact class state changes by date. The question that I am trying to answer is what is the current state of all contacts on a certain date.

DROP TABLE IF EXISTS `contact_class_state`;
CREATE TABLE `contact_class_state` (
    `id` int unsigned NOT NULL AUTO_INCREMENT,
    `contact_id` int unsigned DEFAULT NULL, -- the contact
    `contact_class` int unsigned,
    `state_date` date,
PRIMARY KEY (`id`),
INDEX (`contact_id`)
) DEFAULT CHARSET=utf8;

INSERT INTO `contact_class_state` (`contact_id`, `contact_class`, `state_date`) VALUES
(1, 1, '2011-01-01'),
(2, 1, '2011-01-01'),
(3, 1, '2011-01-01'),
(4, 1, '2011-01-01'),
(5, 1, '2011-01-01'),
(1, 2, '2011-02-01'),
(3, 2, '2011-02-01'),
(5, 2, '2011-02-01'),
(1, 1, '2011-02-15'),
(5, 3, '2011-03-01');

For example, the following query:

SELECT contact_id, contact_class, state_date
FROM contact_class_state
WHERE state_date <= '2011-02-27'
ORDER BY contact_id, state_date DESC

returns

+------------+---------------+------------+
| contact_id | contact_class | state_date |
+------------+---------------+------------+
|          1 |             1 | 2011-02-15 |
|          1 |             2 | 2011-02-01 |
|          1 |             1 | 2011-01-01 |
|          2 |             1 | 2011-01-01 |
|          3 |             2 | 2011-02-01 |
|          3 |             1 | 2011-01-01 |
|          4 |             1 | 2011-01-01 |
|          5 |             2 | 2011-02-01 |
|          5 |             1 | 2011-01-01 |
+------------+---------------+------------+

While this is technically correct, I only need the first (or last if sorted ASC) row for each contact_id as the latest date will always give me current state of the contact, per the below:

+------------+---------------+------------+
| contact_id | contact_class | state_date |
+------------+---------------+------------+
|          1 |             1 | 2011-02-15 |
|          2 |             1 | 2011-01-01 |
|          3 |             2 | 2011-02-01 |
|          4 |             1 | 2011-01-01 |
|          5 |             2 | 2011-02-01 |
+------------+---------------+------------+

I am pretty sure a sub or a complex query would do the trick but I am having a mental block with the SQL. I am also open to other approaches to solve this issue.

Thanks!

Answer 1

If your query is in fact what you want (except then grouped by contact_id), then do exactly that.

SELECT * FROM
(SELECT contact_id, contact_class, state_date
FROM contact_class_state
WHERE state_date <= '2011-02-27'
ORDER BY contact_id, state_date DESC) table1
GROUP BY contact_id

This is tested and works perfect.