Finding and Fixing Slow WordPress Database Queries

Avatar of Andy Adams
Andy Adams on (Updated on )

Slow SQL queries can crush your WordPress site’s performance. Sometimes, slow queries are a result of poorly-formed SQL that never should have been done that way. And sometimes, slow queries were actually fast queries at one point—but as the site grew older, the query got slower and slower, unable to keep up with the expanding database.

Regardless of how your SQL got slow, let’s take a look at a few ways to find and fix problem queries in WordPress.

Finding Slow Queries

Finding the source of slow queries involves 2 steps:

  1. Identifying which queries are actually the slow ones.
  2. Finding the code that is generating and executing them.

Let’s look at two plugins and one SaaS that can help us find slow queries.

Query Monitor

Query Monitor is a plugin that provides a ton of information about the current page. In addition to a whole bunch of information about the internal workings of WordPress, it gives a detailed breakdown of:

  • How many queries happened on this request
  • Which queries on the page took the longest
  • Which functions spent the most time on SQL queries
  • Whether those queries came from plugins, themes or the WordPress core
Query Monitor

Query monitor even identifies slow queries with scary red text, which makes it super easy to pick out the problem SQL:

Red, scary SQL

Debug Bar

Another excellent tool for finding glacially-slow SQL is the old, trusty Debug Bar plugin. Debug Bar gives you information about the internal workings of WordPress when you load a page with things like:

  1. WP_Query parameters
  2. Request information (including rewrite rules matching)
  3. SQL queries generated by the current page

To enable #3 (SQL tracking) in Debug Bar, make sure to enable SAVEQUERIES somewhere on your site—probably in wp-config.php—like so:

if ( ! defined( 'SAVEQUERIES' ) ) {
  define( 'SAVEQUERIES', true );
}

Warning: SAVEQUERIES has an effect on the performance on your site, and probably shouldn’t be used on a production server. Use it on a development machine instead.

Finding slow SQL isn’t as easy with Debug Bar. For example, it doesn’t provide the sortable tables or highlight slow queries for you. What Debug Bar does provide is a function trace that points you exactly where to find the source of a query.

Debug Bar function list

This is a list of the files loaded and the functions that result in the query being executed. Most of the time, you’re interested in the very last entry on the list; this is where the slow query was executed and where you should begin your search. The handy thing about having the context of every single function that leads to this query is that it can shed light on why the SQL was executed in the first place.

NewRelic

NewRelic a service that measures and monitors the performance of your web app, including WordPress. The service provides a metric ton of information about the performance of your site. It’s easy to get lost in the data NewRelic gives you from detailed code execution to line-by-line breakdowns for SQL queries.

NewRelic UI

There are two major differences between NewRelic and the plugins we mentioned earlier:

  1. NewRelic gives a lot more detail about the performance of your PHP, down to the number of milliseconds spent in each function
  2. NewRelic tracks every request to your site in the background, so you can reference it later to find slow SQL. The plugins only give you the current page.

It’s worth noting that NewRelic has a free plan tier that gives general information about the performance of your site, but that you’ll need to upgrade to a paid plan for the bells and whistles for monitoring individual requests and finding slow queries.

Understanding a Slow Query with EXPLAIN

So far, we’ve covered tools for finding slow queries. Now let’s figure out why those queries are bogging things down.

The MySQL EXPLAIN keyword can help, er, explain what’s happening. Adding EXPLAIN to the start of a query shows how MySQL executes a query. For complicated queries, EXPLAIN can help identify slow points in your SQLs, such as slow subqueries or inefficient operations.

For example, if you had a query that looked like this:

SELECT slow_column FROM slow_table

You could EXPLAIN that query by simply running the following:

EXPLAIN SELECT slow_column FROM slow_table

Here’s what the output of EXPLAIN looks like in phpMyAdmin:

MySQL EXPLAIN

Heck, I admittedly don’t understand all of the internal workings of MySQL, but running EXPLAIN on queries still gives insights into how MySQL is executing my SQL. Is the query using an index? Is it scanning the whole table? Even for simple queries, EXPLAIN provides a smidgen of information to help understand what is going on.

You can run EXPLAIN from the MySQL command line or your favorite MySQL tool.

Fixing Slow Queries

Now that we know our query is slow, and EXPLAIN has given told us why it is, let’s look at a few options for fixing those sluggish issues.

Option 1: Change the Query

On CSS-Tricks, we had a query that was dragging the Edit Post screen to a snail’s pace. The query was part of the Custom Fields meta box. Here’s the SQL:

SELECT meta_key
FROM wp_postmeta
GROUP BY meta_key
HAVING meta_key NOT LIKE '\\_%'
ORDER BY meta_key
LIMIT 100

This particular bit of SQL is fetching a list of meta_keys from the wp_postmeta table that don’t start with an underscore (_). The GROUP BY statement means each result is unique.

Running this query 5 times, here’s how long it takes:

1.7146 sec
1.7912 sec
1.8077 sec
1.7708 sec
1.8456 sec

Could we write a different query to get the same result? We need to select unique meta_keys. Unique is a synonym for distinct, which just so happens to be a SQL statement!

Using the DISTINCT statement, we can do the following:

SELECT DISTINCT meta_key 
FROM wp_postmeta 
WHERE meta_key NOT LIKE '\\_%' 
ORDER BY meta_key

Running our rewritten query a handful of times gives the following results:

0.3764 sec
0.2607 sec
0.2661 sec
0.2751 sec
0.2986 sec

This is hardly a scientific comparison but does reveal a significant improvement!

Option 2: Add an Index

When you run a SQL query on a standard MySQL table, MySQL has to scan the entire table to figure out which rows are relevant to this particular query. When your table gets really big, that scanning starts taking a long time.

That’s where MySQL indexes come in. Indexes take the data in a table and organizes it in a way that makes data much easier to locate. By organizing the data a particular way, indexes help reduce the amount of scanning MySQL does for each query.

Indexes can be added to single columns or across multiple columns. The syntax looks like this:

CREATE INDEX wp_postmeta_csstricks ON wp_postmeta (meta_key)

With an index on meta_key, the original SQL query time looks like this:

0.0042 sec
0.0024 sec
0.0031 sec
0.0026 sec
0.0020 sec

That’s really snappy!

A word of caution on indexes: Every time INSERT creates a row or UPDATE is used on an indexed table, the index is re-calculated, which can be a costly operation. Indexes make reading from the table faster, but writing to the table is slower. A well-placed index can make your queries fly, but don’t go index-crazy without monitoring the overall effects of the index on your database.

Option 3: Cache the Query Results

We know we have a slow query. Instead of changing the query, what if we just stored the results of the query? That way, we’d limit how often the query was performed, and we’d get a “free pass” most of the time.

To cache the query, we could use the WordPress Transients API. Transients are used for storing results of expensive operations, such as:

  • Requests to external websites (e.g. fetching recent Facebook posts)
  • Slow chunks of processing (e.g. searching big strings with a regular expression)
  • Slow database queries!

Storing a query result with transients looks something like this:

if ( false === ( $results = get_transient( 'transient_key_name' ) ) ) {
  $results = ...; // Do the slow query to get the results here
  // 60 * 60 is the expiration in seconds - in this case, 3600 seconds (1 hour)
  set_transient( 'transient_key_name', $results, 60 * 60 ); 
}

Storing query results in a transient like so means the query will only be performed once every hour or so. That leads us to the BIG TRANSIENT WARNING: Be careful using transients for things that change a lot.

If you have a query with results that won’t change often, using transients is a nifty way to avoid hitting the database so frequently.

Picking an Approach

We’ve outlined three options, and there are probably 17 other ways to solve this slow query. Which approach do we take?

When working on code that isn’t my own, I prefer to follow the programmer’s maxim: “Do the simplest thing that can possibly work.”

Option 1 (rewriting the query) yielded excellent results, but what if the rewritten query doesn’t always have the same results? We could unknowingly bork our code with a slightly-off query.

Option 2 (adding an index) isn’t always possible depending on the table and columns used by the query. In the case of WordPress core tables, you’d need to worry about side-effects of indexes:

  • Does the core updating routine expect extra indexes?
  • Will adding an index slow down other queries, like INSERT and UPDATE?

Option 3 (caching the results via transients) has minimal impact—we’re not changing the original query and we don’t need to modify the database structure.

Most of the time I use Option 3. In your particular case, you might choose another option depending on the query you’re fixing or the particular site having SQL problems. There isn’t a one-size-fits-all answer to most performance problems, so you’re free to disagree with my choice—or, try all three at once!

Stay Tuned

We outlined a real problem here. The Custom Fields box on CSS-Tricks really was a culprit for some very slow database queries. We also outlined different paths toward potential solutions, but we didn’t actually give you a real coded solution. We’ll be doing a second post that covers that soon, and hopefully gives you the tools for fixing your own slow queries after you’ve discovered them.