Tuning MySQL: my.cnf, avoid this common pitfall!

Caution: Don’t overwrite my.cnf all at once. Backup data and config, make one or two changes per restart, test extensively.

It took me some time to decide the title for this article. MariaDB has been fast replacing MySQL as a growing number of Linux distributions now default to MariaDB over MySQL. MariaDB is an enhanced drop-in replacement for MySQL. Therein lies my largely self-made conundrum, MySQL or MariaDB? That said, the MySQL tuning advice below applies to both MySQL and MariaDB (and Percona). After tuning MySQL over the years I can safely say that the following pitfalls are the most common.

 

Avoid arbitrarily increasing the size of MySQL’s per-connection buffers

my.cnf basic config example

The my.cnf config file is well-known but also often grossly misconfigured. I’m not sure where or when it all started but it has become the norm to keep increasing the size and value of almost every setting in my.cnf without much reasoning behind those increases. Lets look at some important my.cnf parameters, where doing this, will not only hurt MySQL performance but also waste large chunks server memory and as a result reduce MySQL’s overall capacity and throughput.

Buffers such as join_buffer_sizesort_buffer_size, read_buffer_size and read_rnd_buffer_size are allocated per connection. Therefore, a setting of read_buffer_size=1M and max_connections=150 configures MySQL to allocate – from startup – 1MB per connection x 150 connections. For almost two decades the default read_buffer_sizeremains at 128KB. Increasing the default is not only a waste of server memory, but often does not help performance. In nearly all cases, its best to use the defaults by removing or commenting out these four buffer config lines. For a more gradual approach, reduce your current large values to free up wasted RAM, keep reducing them towards default values over time. I’ve actually seen improved throughput by reducing these buffers. Avoid arbitrarily increasing these!

 

Tuning MySQL join_buffer_size

The join_buffer_size is allocated for each full join between two tables. From MySQL’s documentation the join_buffer_size is described as: “The minimum size of the buffer that is used for plain index scans, range index scans, and joins that do not use indexes and thus perform full table scans.” It goes on to say: “Memory allocation time can cause substantial performance drops if the global size is larger than needed by most queries that use it.” The join buffer is allocated to cache table rows when the join can’t use an index. If your database(s) suffer from many joins performed without indexes it cannot be solved by just increasing join_buffer_size. The problem is “joins performed without indexes” and thus the solution for faster joins is to add indexes.

 

Tuning MySQL sort_buffer_size

Unless you have data indicating otherwise, you should avoid arbitrarily increasing the sort_buffer_size as well. Memory here is also assigned per connection! MySQL’s documentation warns: “On Linux, there are thresholds of 256KB and 2MB where larger values may significantly slow down memory allocation, so you should consider staying below one of those values.” Avoid increasing sort_buffer_size above 2M since there is a performance penalty which can eliminate any benefits.

 

Rule of thumb when tuning MySQL

A good rule of thumb is, if you can’t provide a valid reason for increasing any of these buffers, keep them set to the default values. (comment-out of config). Unfortunately, these four config options, often attract large increases when tuning MySQL. Hopefully you’ve found these quick MySQL tuning tips useful.

Also see: MySQL Query Cache Size and Performance -> Basically, if you are using mostly InnoDB tables, then you are better off just disabling query_cache completely. (MySQL query cache is deprecated as of MySQL 5.7.20, and is removed in MySQL 8.0.)

If you are using MariaDB you should also use Aria table types over MyISAM and Innodb in most cases.

Tags: , , , , ,