Question

I have a MySql query that take a very long time to run (about 7 seconds). The problem seems to be with the OR in this part of the query: “(tblprivateitem.userid=?userid OR tblprivateitem.userid=1)”. If I skip the “OR tblprivateitem.userid=1” part it takes only 0.01 seconds. As I need that part I need to find a way to optimize this query. Any ideas?

QUERY:

SELECT 
    tbladdeditem.addeditemid,
    tblprivateitem.iitemid,
    tblprivateitem.itemid
FROM tbladdeditem 
INNER JOIN tblprivateitem 
    ON tblprivateitem.itemid=tbladdeditem.itemid 
        AND (tblprivateitem.userid=?userid OR tblprivateitem.userid=1)
WHERE tbladdeditem.userid=?userid

EXPLAIN:

id    select_type    table            type    possible_keys    key    key_len    ref                    rows    extra
1     SIMPLE         tbladdeditem     ref     userid           userid 4          const                  293     Using where
1     SIMPLE         tblprivateitem   ref     userid,itemid    itemid 4          tbladdeditem.itemid    2       Using where

TABLES:

tbladdeditem contains 1 100 000 rows:

CREATE TABLE `tbladdeditem` (
    `addeditemid` int(11) NOT NULL auto_increment,
    `itemid` int(11) default NULL,
    `userid` mediumint(9) default NULL,
    PRIMARY KEY  (`addeditemid`),
    KEY `userid` (`userid`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

tblprivateitem contains 2 700 000 rows:

CREATE TABLE `tblprivateitem` (
    `privateitemid` int(11) NOT NULL auto_increment,
    `userid` mediumint(9) default '1',
    `itemid` int(10) NOT NULL,
    `iitemid` mediumint(9) default NULL,
    PRIMARY KEY  (`privateitemid`),
    KEY `userid` (`userid`),
    KEY `itemid` (`itemid`) //Changed this index to only use itemid instead
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

Answer 1

UPDATE

I made my queries and schema match your original question exactly, multi-column key and all. The only possible difference is that I populated each table with two million entries. My query (your query) runs in 0.15 seconds.

delimiter $$
set @userid = 6
$$
SELECT 
    tbladdeditem.addeditemid,    tblprivateitem.iitemid,    tblprivateitem.itemid
FROM tbladdeditem 
INNER JOIN tblprivateitem 
    ON tblprivateitem.itemid=tbladdeditem.itemid 
        AND (tblprivateitem.userid=@userid or tblprivateitem.userid = 1)
WHERE tbladdeditem.userid=@userid

I have the same explain that you do, and with my data, my query return over a thousand matches without any issue at all. Being completely at a loss, as you really shouldn’t be having these issues — is it possible you are running a very limiting version of MySQL? Are you running 64-bit? Plenty of memory?

I had made the assumption that your query wasn’t performing well, and when mine was, assumed I had fixed you problem. So now I eat crow. I’ll post some of the avenues I went down. But I’m telling you, your query the way you posted it originally works just fine. I can only imagine your MySQL thrashing to the hard drive or something. Sorry I couldn’t be more help.

PREVIOUS RESPONSE (Which is also an update)

I broke down and recreated your problem in my own database. After trying independent indexes on userid and on itemid I was unable to get the query below a few seconds, so I set up very specific multi-column keys as directed by the query. Notice on tbladdeditem the multi-column query begins with itemid while on the tblprivateitem the columns are reversed:

Here is the schema I used:

CREATE TABLE `tbladdeditem` (
  `addeditemid` int(11) NOT NULL AUTO_INCREMENT,
  `itemid` int(11) NOT NULL,
  `userid` mediumint(9) NOT NULL,
  PRIMARY KEY (`addeditemid`),
  KEY `userid` (`userid`),
  KEY `i_and_u` (`itemid`,`userid`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

CREATE TABLE `tblprivateitem` (
  `privateitemid` int(11) NOT NULL AUTO_INCREMENT,
  `userid` mediumint(9) NOT NULL DEFAULT '1',
  `itemid` int(10) NOT NULL,
  `iitemid` mediumint(9) NOT NULL,
  PRIMARY KEY (`privateitemid`),
  KEY `userid` (`userid`),
  KEY `itemid` (`itemid`),
  KEY `u_and_i` (`userid`,`itemid`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

I filled each table with 2 million entries of random data. I made some assumptions:

• userid varies from 1 to 2000
• itemid varies between 1 and 10000

This gives each user about a thousand entries in each table.

Here are two versions of the query (I’m using workbench for my editor):

Version 1 – do all the filtering on the join.

Result: 0.016 seconds to return 1297 rows

delimiter $$
set @userid = 3
$$
SELECT 
    a.addeditemid,
    p.iitemid,
    p.itemid
FROM tblprivateitem as p
INNER JOIN tbladdeditem as a
    ON (p.userid in (1, @userid))
        AND p.itemid = a.itemid 
        AND a.userid = @userid
$$

Here’s the explain:

EXPLAIN: 
id select_type table type  key     ref  rows extra
1  SIMPLE      p     range u_and_i      2150 Using where; Using index
1  SIMPLE      a     ref   i_and_u      1    Using where; Using index

Version 2 – filter up front

Result: 0.015 seconds to return 1297 rows

delimiter $$
set @userid = 3
$$
SELECT 
    a.addeditemid,
    p.iitemid,
    p.itemid
from 
  (select userid, itemid, iitemid from tblprivateitem 
      where userid in (1, @userid)) as p
  join tbladdeditem as a on p.userid = a.userid and a.itemid = p.itemid;
where a.userid = @userid
$$

Here’s the explain:

id select_type table      type  key     ref               rows extra
1  PRIMARY     <derived2> ALL   null    null              2152
1  PRIMARY     a          ref   i_and_u p.itemid,const    1    Using where; Using index
2  DERIVED     p1         range u_and_i                   2150 Using where