Need help optimizing mysql query to get it to sort quickly by index

Question

Someone helped me come up with this query but its still too slow; The order by is slowing it down and I dont think its using my index I'm hoping someone can fix it for me :D Yes I read the manual page but I can't understand it.

Query:

 EXPLAIN SELECT u.id, u.url, u.title, u.numsaves
 FROM urls u
 JOIN tags t ON t.url_id = u.id
 AND t.tag = 'osx'
 ORDER BY u.numsaves DESC
 LIMIT 20 

 Showing rows 20 - 19 ( 20 total, Query took 1.5395 sec) [numsaves: 6130 - 2107]


 id     select_type     table   type    possible_keys   key     key_len     ref     rows    Extra
 1  SIMPLE  t   ref     tag_id  tag_id  767     const   49432   Using where; Using index; Using temporary; Using filesort
 1  SIMPLE  u   eq_ref  PRIMARY,id_numsaves_IX  PRIMARY     4   jcooper_whatrethebest_urls.t.url_id     1   

Database:

 CREATE TABLE `urls` (
 `id` int(11) NOT NULL AUTO_INCREMENT,
 `url` text NOT NULL,
 `domain` text,
 `title` text NOT NULL,
 `description` text,
 `numsaves` int(11) NOT NULL,
 `firstsaved` varchar(256) DEFAULT NULL,
 `md5` varchar(255) NOT NULL DEFAULT '',
 PRIMARY KEY (`id`),
 UNIQUE KEY `md5` (`md5`),
 KEY `id_numsaves_IX` (`id`,`numsaves`)
 ) ENGINE=InnoDB AUTO_INCREMENT=2958560 DEFAULT CHARSET=utf8

 CREATE TABLE `tags` (
 `url_id` int(11) DEFAULT NULL,
 `hash` varchar(255) NOT NULL,
 `tag` varchar(255) NOT NULL,
 UNIQUE KEY `tag_id` (`tag`,`url_id`)
 ) ENGINE=InnoDB DEFAULT CHARSET=utf8

Answer 1

I think the main problem with your query is your choice of indexes.

1) tags has a compound UNIQUE KEY on tag and url_id but no PRIMARY KEY .

If nothing else, you should make it primary - this may help a bit with performance. Also, you might want to take a close look if VARCHAR(255) is really necessary for your tags. It makes the index quite big.

2) add a separate index on numsaves since you're ordering by that. The compound index on id and numsaves is not going to help here.

3) EXPLAIN says that you have 49432 rows in tags that match "osx" . This is quite redundant. You may want to split your tags table into two, one containing the text while the other contains the N:M link to urls .