am working with a wordpress website that is performing the following query, but I see this query is doing many inner joins and the website takes long to load and goes down a lot, and I have been trying to create a query that produces the same result but with no success yet
I would like to know what could be a better way to do this
SELECT *
FROM wp_posts
INNER JOIN wp_postmeta color ON wp_posts.ID = color.post_id
INNER JOIN wp_postmeta transmission ON wp_posts.ID = transmission.post_id
INNER JOIN wp_postmeta model ON wp_posts.ID = model.post_id
INNER JOIN wp_postmeta brand ON wp_posts.ID = brand.post_id
AND color.meta_key = 'color'
AND color.meta_value = 'red'
AND transmission.meta_key = 'transmission'
AND transmission.meta_value = 'auto'
AND model.meta_key = 'model'
AND model.meta_value = 'model'
AND brand.meta_key = 'brand'
AND brand.meta_value = 'brand'
AND wp_posts.post_status = 'publish'
AND wp_posts.post_type = 'car'
ORDER BY wp_posts.post_title
Here’s the explain output.
+----+-------------+-----------+--------+-----------------------------+----------+---------+------------------------+------+----------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-----------+--------+-----------------------------+----------+---------+------------------------+------+----------------------------------------------+
| 1 | SIMPLE | color | ref | post_id,meta_key | meta_key | 768 | const | 629 | Using where; Using temporary; Using filesort |
| 1 | SIMPLE | wp_posts | eq_ref | PRIMARY,type_status_date,ID | PRIMARY | 8 | tmcdb.color.post_id | 1 | Using where |
| 1 | SIMPLE | brand | ref | post_id,meta_key | post_id | 8 | tmcdb.wp_posts.ID | 4 | Using where |
| 1 | SIMPLE | transmission | ref | post_id,meta_key | post_id | 8 | tmcdb.color.post_id | 4 | Using where |
| 1 | SIMPLE | model | ref | post_id,meta_key | post_id | 8 | tmcdb.transmission.post_id | 4 | Using where |
+----+-------------+-----------+--------+-----------------------------+----------+---------+------------------------+------+----------------------------------------------+
It seems you are trying to obtain a result set with one row per post of type
car
. It seems you want to display various attributes of each car in the post, and those are stashed away inpostmeta
.Pro tip: Never use
SELECT *
in software unless you absolutely know why you’re doing it. Especially with queries containing lots ofJOIN
operations,SELECT *
returns lots of pointless and redundant columns.There’s a query design trick to know for the WordPress
postmeta
table. If you want to get a particular attribute, do this:It’s super-important to understand this pattern when doing what you’re trying to do. This pattern is required because
postmeta
is a peculiar type of table called a key-value or entity-attribute-value store. What’s going on here? A few things:posts
table and a particular attribute from thepostmeta
table.LEFT JOIN
ing thepostmeta
table so you still get a row if the attribute is missing.postmeta
table. Here it’spostmeta AS color
.meta_key
(here it’s'color' = color.meta_key
) in theON
condition of the join.SELECT
clause to present thepostmeta.meta_value
item with an appropriate column name. Here it’scolor.meta_value AS color
.Once you get used to employing this pattern, you can stack it up, with a cascade of
LEFT JOIN
operations, to get lots of different attributes, like so.I’ve done a bunch of indenting on this query to make it easier to see the pattern. You may prefer a different indenting style.
It’s hard to know why you were having performance problems with the query in your question. It’s possibly because you were getting a combinatorial explosion with all the
INNER JOIN
operations that was then filtered. But at any rate the query you showed was probably returning no rows.If you are still having performance trouble, try creating a compound index on
postmeta
on the(post_id, meta_key, meta_value)
columns. If you’re creating a WordPress plugin, that’s probably a job to do at plugin installation time.This is a WordPress database, and you might be reluctant to make extensive changes to the schema, because it could break other parts of the application or complicate upgrades in the future.
The difficulty of this query shows one of the downsides to the entity-attribute-value design. That design is flexible in that it allows for new attributes to be created at runtime, but it makes a lot of queries against such data more complex than they would be with a conventional table.
The schema for WordPress has not been optimized well. There are some naive indexing mistakes, even in the most current version 4.0.
For this particular query, the following two indexes help:
The
bk1
index helps to look up exactly the right meta key and value.The
bk2
index helps to avoid the filesort.These indexes can’t be covering indexes, because
post_title
andmeta_value
areTEXT
columns, and these are too long to be fully indexed. You’d have to change them toVARCHAR(255)
. But that risks breaking the application, if it’s depending on storing longer strings in that table.Speeding up
wp_postmeta
is detailed here: http://mysql.rjweb.org/doc.php/index_cookbook_mysql#speeding_up_wp_postmetaAnd Why are references to wp_postmeta so slow?
To resolve performance issue with 10+ joins SQL queries on innodb tables using utf8 charset, create a new index on postmeta :
Backup database first. Reduce
[wp_]postmeta.meta_key
length to 191 to avoid “Specified key was too long; max key length is 767 bytes” error.ALTER TABLE wp_postmeta MODIFY meta_key VARCHAR(191);
Create index
CREATE INDEX wpm_ix ON wp_postmeta (post_id, meta_key);
For performance try:
Be explicit on the columns you want to pull.
See what indexes you may or may not need.
Limit the amount of rows being pulled.
is this better?
If its still slow, which it probably will be, try adding this index,
You could also try adding this index to
wp_post
The more you can limit the select list, (the bit between
SELECT
andFROM
,) the better. There is no point returning lots of data you won’t use. You’ll get the best performance if the whole select list is “covered” by an index.Assuming you can actually change the code that handles the results, I would make it a much simpler query and use the code to filter the results.
Or, you could do something like…
In WordPress there’s a good query tool the WP_Query. To search in post meta values you can use this code:
For more information about the query API, visit this site, there’re numerous example:
http://codex.wordpress.org/Class_Reference/WP_Query