优化此SQL查询

Question

This SQL query disgusts me. I didn't write it, but it's a massive cause of issues on our servers. I'm willing to split it up into multiple queries and do some of the processing via PHP (like, the RAND()).

$sql = "SELECT a.code, a.ad_id, a.position, a.type, a.image, a.url, a.height, a.width
    FROM " . AD_TABLE ." a, " . USER_GROUP_TABLE . " g
    WHERE (a.max_views >= a.views OR a.max_views = '0')
    AND (FIND_IN_SET(" .$forum_id. ", a.show_forums) > 0 OR a.show_all_forums = '1')
    AND g.user_id = " . $user->data['user_id'] . "
    AND FIND_IN_SET(g.group_id, a.groups)
    AND FIND_IN_SET(" . $user->data['user_rank'] . ", a.ranks)
    AND a.start_time < " . time() . "
    AND a.end_time > " . time() . "
    AND (a.clicks <= a.max_clicks OR a.max_clicks = '0')
    ORDER BY rand()";

Yeesh, I feel icky after pasting that...

EDIT: Below is the results of the "EXPLAIN" on a sample query in the above format, comma delimited:

"id","select_type","table","type","possible_keys","key","key_len","ref","rows","Extra"
1,"SIMPLE","g","ref","user_id","user_id","3","const",6,"Using temporary; Using filesort"
1,"SIMPLE","a","ALL","max_views","","","",10,"Using where"

It is

Answer 1

You have three major issues here:

1. FIND_IN_SET .

   It's not sargable, an index cannot make it faster. Create a many-to-many relationship table (or tables).

   • a.start_time < GETDATE() AND a.end_time > GETDATE()

   MySQL is not good in optimizing that. You can keep you timespans as geometry boxes and create a SPATIAL INDEX over them, this will be much faster (though less readable)

   • ORDER BY RAND()

   If you are using this to sample data (ie you don't need all rows but rather a small random subset), there is a more efficient way to do this, described in this article in my blog:

   • Selecting random rows

Answer 2

• You do not have an explicit join in this query between table a and table g : they are only related by find_in_set (g.group_id, a.groups)
• How large is your set in "a.groups" , ie does the csv string contain one group-id most of the times?
• If 99% of the cases contain only 1 group, then make a foreach loop over "a.groups" in php and execute a real join (or probably it might eliminate your user_group_table altogether) from the query.
  • For the minority of cases where you have more than 1 group membership, the query will still perform OK with an explicit join.

This will add some more code in your php class/function.

Answer 3

I am guessing your problem lies with the dates.

AND a.start_time < " . time() . "
AND a.end_time > " . time() . "

I would try putting indexes on these fields and see if that helps. Comparing dates causes the database to compare to each row in the table.

Answer 4

If you execute this a lot.. try and use bind variables (instead of string concatenation) .. so the query does not have to be parsed every single time..

edit: Sorry..didn't see the MYSQL tag. Unless it's changed recently I don't think MySQL likes prepared statements. ORACLE on the other hand LOVES them.