加入4个表的MYSQL查询优化

Question

Need help to optimize this query

The task is to select my friends from friends table, to check if any of them is in ignore table, to get their info from users table, also to get the last message date from messages table. My friends also can be group's.

My query:

SELECT
        `u`.`id`, `u`.`login`, `u`.`name`, `u`.`surname`,
        CONCAT(`u`.`name`, ' ', `u`.`surname`) AS `namesurname`, `u`.`avatar`,
        `u`.`status`, `u`.`status_text`, `u`.`last_active`, `u`.`type`,
        `f`.`status` AS `friend_state`, `i`.`date` AS `ignore_date`,
        MAX(`m`.`date`) AS `last_message_date` 
    FROM `friends` AS f 
    INNER JOIN `users` AS u ON `u`.`id`=`f`.`friend_id`
    LEFT JOIN `ignore` AS `i` ON `i`.`uid`='10' AND `i`.`ignore_id`=`f`.`friend_id`
    LEFT JOIN `messages` AS `m` ON `m`.`date`>DATE_ADD(NOW(), INTERVAL -1 DAY) AND 
    (
        (`m`.`fromid`=`u`.`id` AND `m`.`toid`='10') OR 
        (`m`.`toid`=`u`.`id` AND `m`.`fromid`='10') OR
        (`m`.`toid`=`u`.`id` AND `m`.`toid` IN (16,22,27))
    ) 
    WHERE `f`.`uid`='10'
    GROUP BY `u`.`id`
    ORDER BY `last_message_date` DESC, `namesurname`

For 170k messages this query tooks 2.2-2.5 seconds.

if I remove

(`m`.`toid`=`u`.`id` AND `m`.`toid` IN (16,22,27))

the it tooks 0.55 seconds

10 - is my users ID

16,22,27 - id of groups

I need this line ( m . toid = u . id AND m . toid IN (16,22,27)) because in group chats it shouldn't depend on who sent the last message.

Thanks in advance

Answer 1

You probably don't need LEFT . Assuming that, I would turn the ORs into UNIONs and start with such:

Since you have not provided SHOW CREATE TABLE , I will guess that messages has PRIMARY KEY(mid) .

SELECT ...
    FROM ( ( SELECT mid FROM messages WHERE fromid = u.id AND toid = 10 )
           UNION ALL 
           ( SELECT mid FROM messages WHERE toid = u.id AND fromid = 10 )
           UNION ALL 
           ( SELECT mid FROM messages WHERE toid = u.id AND toid IN (16,22,27) ) AS um
    JOIN users AS u  ON ...
    JOIN 

Oops, instead I will approach it from trying to do as much as possible in messages first, assuming the 1-day limitation is a significant filter. (Notice how an estimated 170K rows need to be scanned.)

Start by checking this to see if it is 'fast' and delivers not 'too many' rows:

SELECT COUNT(*) FROM (
        ( SELECT toid AS either_id, MAX(date) AS last_date FROM messages
             WHERE toid IN (10, 16,22,27) AND date > NOW() + INTERVAL 1 DAY )
        UNION ALL   -- or, if you get dups, UNION ALL
        ( SELECT fromid, MAX(date) FROM messages
             WHERE fromid = 10 AND date > NOW() + INTERVAL 1 DAY )
                     ) um;

Assuming that is OK, then work on

SELECT ...
    FROM ( ... ) AS um   -- the derived table above
    JOIN users AS u  ON um.either_id = u.id
    JOIN ...   -- the rest of the stuff

Indexes...

messages:  (toid, date)
messages:  (fromid, date)

(It's a really messy query, so I don't have confidence that I have helped.)