Question

Let’s say i have three tables: users, quizzes and questions. Each has an id (integer) field and a created_at (datetime) field. I want to be able to put all of these tables together, sort by created_at desc, and then get a ‘page’ of records (ie by manipulating offset) where a page is ten records for example. Ie, i want to get the ten most recently created records from either of those three tables.

My first thought would just be to do an outer join but this would result in a massive join table which really isn’t very efficient. It’s also fundamentally wrong – i don’t want to combinatorize all the records, i just want to put them all together into one list effectively.

I then tried just putting both tables into the from clause (for two tables just to test it out) and using comcat to order the tables:

select users.id, users.created_at, quizzes.id, quizzes.created_at, concat(users.created_at, quizzes.created_at) from users, quizzes order by concat(users.created_at, quizzes.created_at) desc limit 5;

But this doesn’t work either as, again, the two tables are combined and it takes a long time (10ish seconds) to do the query. Plus, concat isn’t the right approach i think.

Can anyone see what i’m trying to do and help me out? I feel like i’m not explaining it very well, monday morning syndrome perhaps. cheers, max

EDIT – here’s my solution, a tweaked version of Tomalak’s answer below. I’m using the updated_at field rather than created_at but it’s just another datetime field.

SELECT id, updated_at FROM ( 
  SELECT concat('user_',id) as id, updated_at FROM users 
  UNION SELECT concat('quiz_',id) as id, updated_at FROM quizzes 
  UNION SELECT concat('question_',id) as id, updated_at FROM questions ) 
  as everything ORDER BY updated_at DESC LIMIT 0, 10;

A problem i had with Tomalak’s answer was that you lose track of which table each record is from. So, above, i concatenate the table name to the start of the id, so you end up with something that looks like this:

+---------------+---------------------+
| id            | updated_at          |
+---------------+---------------------+
| user_1        | 2010-11-23 16:38:11 | 
| quiz_1306     | 2010-11-23 10:37:49 | 
| user_2        | 2010-11-22 11:57:08 | 
| user_401      | 2010-11-22 11:48:20 | 
| quiz_1303     | 2010-11-22 11:47:53 | 
| question_5163 | 2010-11-22 11:42:20 | 
| question_5162 | 2010-11-22 11:42:16 | 
| question_5161 | 2010-11-22 11:37:11 | 
| question_5160 | 2010-11-22 11:34:55 | 
| quiz_1304     | 2010-11-22 11:31:54 | 
+---------------+---------------------+
10 rows in set (0.16 sec)

Answer 1

Like this?

SELECT 
  tbl, id, updated_at 
FROM
  (
    SELECT 'users' AS tbl, id, updated_at FROM users
  UNION
    SELECT 'quizzes',      id, updated_at FROM quizzes
  UNION
    SELECT 'questions',    id, updated_at FROM questions
  ) as everything
ORDER BY
  updated_at DESC
LIMIT 10;

In terms of query efficiency, it might be that this variant runs faster:

SELECT 
  tbl, id, updated_at 
FROM
  (
    SELECT   'users' AS tbl, id, updated_at 
    FROM     users 
    ORDER BY updated_at DESC LIMIT 10
  UNION
    SELECT   'quizzes',      id, updated_at 
    FROM     quizzes
    ORDER BY updated_at DESC LIMIT 10
  UNION
    SELECT  'questions',     id, updated_at 
    FROM     questions 
    ORDER BY updated_at DESC LIMIT 10
  ) as everything
ORDER BY
  updated_at DESC
LIMIT 10;

You should have an index on (updated_at, id) on each individual table.

This way the query can be fulfilled entirely from looking at the index, which is already sorted and also contains the id field. The original table would not have to be looked at, unless you want to add more fields to the output. Additionally, the “outer” ORDER BY (which can’t use an index) would only have to sort 30 rows.