You are viewing an older version of this section. View current production version.

Transitioning from MySQL to MemSQL min read


If you are running MemSQL Helios, use the admin endpoint to run the commands discussed in this topic.

Because MemSQL is wire-compatible with MySQL, transitioning your data and application from MySQL to MemSQL is very straightforward. Once you migrate your data via standard MySQL tools, all it takes to get your app running is to point it at MemSQL.

For simplicity, this guide assumes that both MySQL and MemSQL are running on your local machine and that MySQL is running on the standard port (3306) while MemSQL runs on 3307. Furthermore, it also assumes that both are accessible without a password by the root user.

Migrating Data with mysqldump

mysqldump is a popular tool packaged with the MySQL client infrastructure. It queries the database to produce a series of CREATE TABLE and INSERT statements that can be replayed to restore the database.

Find the mysqldump reference page here.

While moving your data from MySQL to MemSQL, there are a few considerations to keep in mind:

  • For most MySQL storage engines (MyISAM, InnoDB, etc), indexes are stored as B-trees. In MemSQL, indexes can be unidirectional lock-free skip lists (ascending or descending) or lock-free hash tables. Picking the right data structure for your index can have a significant impact on the performance of your application. While hash tables are optimized for key-value look ups, skip lists are extremely flexible for complex range scans and sorts (ORDER BY). While transferring your schema, you should audit your table definitions and investigate whether your indexes can be optimized for MemSQL. The default BTREE notation is converted into a skip list (see Skip List Indexes .
  • mysqldump will generate a few queries that are unsupported by MemSQL. For example, MemSQL does not support disabling UNIQUE_CHECKS. To make it easier to work with mysqldump, unsupported features are by default reported as warnings instead of errors. This functionality can be controlled by adjusting the warn_level variable. See Unsupported Features for more details. Some components of a CREATE TABLE statement might be blocked completely. If you run into this issue while loading a schema into MemSQL, you can manually massage the schema definition into something supported by MemSQL.
  • If the machine running MemSQL does not have enough memory to support the data you’re loading, the server will issue an error on offending INSERT statements indicating its out-of-memory state. In this case, you should upgrade your machine to one with more memory. If you copy your existing memsqlbin directory to the new machine, MemSQL will be able to reuse the schema definitions and INSERT statements that have already compiled.

You should separate your schema and data into separate files, so that you can easily review and modify your schema if necessary. To produce a dump of your database, run something like:

$ mysqldump -h -u root -B [database name] --no-data -r schema.sql
$ mysqldump -h -u root -B [database name] --no-create-info -r data.sql

You can then replay these files directly into MemSQL by running:

$ mysql -h -u root -P 3307 < schema.sql
$ mysql -h -u root -P 3307 < data.sql

While this step runs, you can observe the memsql.log file to see which unsupported features have been ignored. After the import is completed, you can connect to MemSQL and start querying the tables directly.

Transitioning Your Application

To transition your application, simply change the connection credentials to point to MemSQL.

One thing to keep in mind is MemSQL Code Generation. Code generation makes loading a schema into MemSQL for the first time is slower than with MySQL. The first time MemSQL encounters a table schema, it generates and compiles code that implements infrastructure around the table (memory allocation, inserts, deletes, iterations, etc.). Once a table is compiled, MemSQL will be able to reuse it for the lifetime of your application — even if you restart the server or drop (and recreate) the table. The INSERT queries generated by mysqldump also have to be compiled exactly once.

Here are a couple of recommendations when transitioning your application:

  • Even if you’re connecting to MemSQL locally, use the explicit host instead of localhost. Most MySQL clients will resolve localhost to use the global MySQL socket file and ignore the port setting. See the MemSQL FAQ for more information.
  • As you run through the application, monitor the memsql.log file. Queries that throw errors most likely correspond to unsupported syntax (see SQL Reference Overview for supported SQL surface area).