This is an overview of query optimizer feature development in recent versions of MariaDB (5.5, 10.0) and MySQL (5.6).

To make things easier to understand, let's first partition the optimizer into areas:

optimizer-areas-r2

Let's look at things in historical order. There is no known evidence of the use of SQL query optimization by ancient Egyptians. In April 2012, MariaDB 5.3 and MariaDB 5.5 became stable. (5.5 was a merge of 5.3 with MySQL 5.5). The 5.3/5.5 release included a number of new major query optimizer features:

optimizer-areas-mariadb-55-r2

The size of the bubble roughly corresponds to the impact of the feature (in opinion of the author of this article). "Hash join" is in quotes, because this feature is not what is generally meant by hash join in the world of databases. Smaller features and bugfixes are not shown.

In February, 2013, MySQL 5.6 had its first stable release. It had a number of new major optimizer features. Some of them had common ancestry with MariaDB's features, and some of them were new development:

optimizer-areas-mysql-56-r2

In April, 2014, MariaDB 10.0 became stable. As compared to MariaDB 5.5, it had some features that were backported from MySQL 5.6, and also some new features were implemented:

optimizer-areas-mariadb-100-r2

One can see that

  • MariaDB has features that MySQL doesn't have
  • However, some MySQL 5.6 features were not ported into MariaDB 10.0.

The MariaDB team is frequently asked if all features (or feature X) from MySQL 5.6 are available in MariaDB, so here's our position on the features that have not [yet] been backported:

  • EXPLAIN FORMAT=JSON - This is available in MariaDB Server 10.1. Please read the documentation to know more.
  • Optimizer trace - we have doubts whether the feature is usable in its current form. For simple queries, it's easier to analyze using other means, for complex queries the size of the trace grows beyond any limit. We're looking for input re. how optimizer trace was (or could be) useful to somebody in a real-world setting.
  • Faster optimization for eq_ref - we believe that the optimization is useful but its implementation in MySQL uses a wrong approach.

See also

Comments

Comments loading...