How to optimize join when search in categories - mysql

Question

I have a table with items:
CREATE TABLE `ost_content` (
`uid` mediumint(8) unsigned NOT NULL AUTO_INCREMENT,
`type` enum('media','serial','season','series') NOT NULL,
`alias` varchar(200) NOT NULL,
`views` mediumint(7) NOT NULL DEFAULT '0',
`ratings_count` enum('0','1','2','4','5') NOT NULL DEFAULT '0',
`ratings_sum` mediumint(5) NOT NULL DEFAULT '0',
`upload_date` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',
`conversion_status` enum('converting','error','success','announcement') NOT NULL DEFAULT 'converting',
PRIMARY KEY (`uid`),
UNIQUE KEY `idx_uid_type` (`uid`,`type`),
KEY `idx_type` (`type`),
KEY `idx_upload_date DESC` (`upload_date`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;
And table, that connect items with categories:
CREATE TABLE `ost_categories2media` (
`categories2media_id` mediumint(6) unsigned NOT NULL AUTO_INCREMENT,
`categories2media_category_id` smallint(5) unsigned NOT NULL,
`categories2media_uid` mediumint(8) unsigned NOT NULL,
PRIMARY KEY (`categories2media_id`),
KEY `categories2media_media_id` (`categories2media_uid`),
KEY `categories2media_category_id` (`categories2media_category_id`)
) ENGINE=InnoDB AUTO_INCREMENT=501114 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
Than, I executing query:
SELECT
c1.uid,
c1.alias,
c1.type,
c1.views,
c1.upload_date,
c1.ratings_sum,
c1.ratings_count,
c1.conversion_status
FROM
ost_content c1
LEFT JOIN ost_categories2media c2m ON c2m.categories2media_uid = c1.uid
WHERE
c2m.categories2media_category_id = '53'
AND c1.conversion_status IN ('success', 'announcement')
AND c1.type IN ('serial', 'media')
ORDER BY
c1.upload_date DESC
LIMIT 16, 16
It executing slow, categories2media_category_id check many rows:
+----+-------------+-------+--------+--------------------------------------------------------+------------------------------+---------+---------------------------------+-------+----------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+--------+--------------------------------------------------------+------------------------------+---------+---------------------------------+-------+----------------------------------------------+
| 1 | SIMPLE | c2m | ref | categories2media_media_id,categories2media_category_id | categories2media_category_id | 2 | const | 32076 | Using where; Using temporary; Using filesort |
| 1 | SIMPLE | c1 | eq_ref | PRIMARY,idx_uid_type,idx_type | PRIMARY | 3 | uakino.c2m.categories2media_uid | 1 | Using where |
+----+-------------+-------+--------+--------------------------------------------------------+------------------------------+---------+---------------------------------+-------+----------------------------------------------+
How I can optimize or rewrite this query?

Answer 1

Mysql indexes are like cooks, too many of them aren't very useful because mysql uses only one index per table. Let's look at ost_categories2media,
That's three separate indexes on three columns. You are better off with two indexes like this.
PRIMARY KEY (`categories2media_id`),
KEY `categories2media_media_id` (`categories2media_uid`,`categories2media_category_id`)
Now mysql no longer has to decide between an index on categories2media_uid or categories2media_category_id it has an index that covers both!
Looking at your ost_content table we see
PRIMARY KEY (`uid`),
UNIQUE KEY `idx_uid_type` (`uid`,`type`),
KEY `idx_type` (`type`),
KEY `idx_upload_date DESC` (`upload_date`)
Some of these indexes are a bit redundant. Any query that filters on the uid field can use the PK while any query that filters on type can use idx_type that means idx_uid_type is there just to enforce the uniqueness. But we can make it more usefull like this:
PRIMARY KEY (`uid`),
UNIQUE KEY `idx_uid_type` (`type`,`uid`),
KEY `idx_upload_date DESC` (`upload_date`)
We've got rid of one index! that ought to make your indexes a lot faster. You still have an index on upload_date that isn't used in this particulary query. So how about a composite index for that?
PRIMARY KEY (`uid`),
UNIQUE KEY `idx_uid_type` (`type`,`uid`),
KEY `idx_upload_date DESC` (`uid`,`upload_date`)

Answer 2

First, the LEFT JOIN is not necessary. So, you can write the query as:
SELECT c.*
FROM ost_content c JOIN
ost_categories2media c2m
ON c2m.categories2media_uid = c.uid
WHERE c2m.categories2media_category_id = '53' AND
c.conversion_status IN ('success', 'announcement') AND
c.type IN ('serial', 'media')
ORDER BY c.upload_date DESC
LIMIT 16, 16;
Unfortunately, your conditions on the content table are not simple = conditions. If they were, and index on ost_content(conversion_status, type, uid) would be recommended. This might still be the better option.
Another option is to go the other way: An index on ost_categories2media(categories2media_category_id, categories2media_uid).
You might find that the first composite index and this query work best:
SELECT c.*
FROM ((SELECT c.*
FROM ost_content c JOIN
ost_categories2media c2m
ON c2m.categories2media_uid = c.uid
WHERE c2m.categories2media_category_id = '53' AND
c.conversion_status = 'success' AND
c.type IN ('serial', 'media')
) UNION ALL
(SELECT c.*
FROM ost_content c JOIN
ost_categories2media c2m
ON c2m.categories2media_uid = c.uid
WHERE c2m.categories2media_category_id = '53' AND
c.conversion_status = 'announcement' AND
c.type IN ('serial', 'media')
)
) c
ORDER BY c.upload_date DESC
LIMIT 16, 16;
This looks more complicated, but each subquery can take advantage of the index, so it might have improved performance.