Question

The following query is taking more than 25 seconds to complete:

SELECT c.* , (c.age+ (UNIX_TIMESTAMP()-UNIX_TIMESTAMP(c.newdate))) AS ranking , IF(uc.id_user = 7,1,0) AS favorite
FROM c 
LEFT OUTER JOIN uc ON uc.id_cluster = c.id AND uc.id_user = '7' 
LEFT OUTER JOIN d ON d.id_cluster = c.id 
LEFT OUTER JOIN dt0 ON dt0.id_document = d.id 
LEFT OUTER JOIN t0 ON dt0.id_term = t0.id 
WHERE MATCH(t0.normalizacion) AGAINST ('term' IN NATURAL LANGUAGE MODE) 
GROUP BY c.id 
ORDER BY ranking ASC 
LIMIT 30

Indexes:

• c.id PRIMARY
• c.age INDEX
• c.newdate INDEX
• uc.id_user, uc.id_cluster PRIMARY
• d.id PRIMARY
• d.id_cluster INDEX
• dt0.id_document, dt0.id_term PRIMARY
• dt0.id_document INDEX
• dt0.id_term INDEX
• t0.id PRIMARY
• t0.normalizacion FULLTEXT

If I remove the ORDER BY clause, it takes only 2 seconds.

I’ve been searching and found that the same index has to be used for both GROUP BY and ORDER BY (tried ordering by c.id and took 2 or 3 seconds). How can I change the query to be faster?

Answer 1

(consolidating the discussion above)

You can’t order efficiently by a computation result. To make this query work fast, create a ranking column that contains c.age-UNIX_TIMESTAMP(c.newdate). Then create an index on id and ranking, i.e. CREATE INDEX id_ranking ON c (id, ranking) to make both GROUP BY and ORDER BY indexed.