Mysql – Speeding up conversion of MyISAM to InnoDB

mysql-5storage-engine

I've got a mysql 5.1 server with a database of approximately 450 tables, taking up 4GB . The vast majority of these tables (all but 2) are MyIsam. This has been fine for the most part (don't need transactions), but the application has been gaining traffic and certain tables have been impacted because of table-locking on updates. That's the reason 2 of the tables are InnoDB now.

The conversion on the smaller tables (100k rows) don't take long at all, causing minimal downtime. However a few of my tracking tables are approaching 50 million rows. Is there a way to speed up an ALTER TABLE...ENGINE InnoDB on large tables? And if not, are there other methods to convert minimizing downtime on these write-heavy tables?

Best Answer

Let me start by saying, I hate ALTER. It's evil, IMHO.

Say, this is your current table schema - 

CREATE TABLE my_table_of_love (
id INT UNSIGNED NOT NULL AUTO_INCREMENT,
my_value VARCHAR(40),
date_created DATE,
PRIMARY KEY(id)
) ENGINE=MyISAM CHARSET=utf8;

Here's the path I recommend -

Create a new table object that will replace the old one:

CREATE TABLE my_table_of_love_NEW (
id INT UNSIGNED NOT NULL AUTO_INCREMENT,
my_value VARCHAR(40),
date_created DATE,
PRIMARY KEY(id)
) ENGINE=InnoDB CHARSET=utf8

Insert all the rows from the old table by name into the new table:

INSERT INTO my_table_of_love_NEW (id,my_value,date_created)
SELECT id,my_value,date_created FROM my_table_of_love;

Smoke test your migration:

SELECT COUNT(*) FROM my_table_of_love_NEW;
SELECT COUNT(*) FROM my_table_of_love;
SELECT a.id,a.my_value,a.date_created FROM my_table_of_love_NEW a
LEFT JOIN my_table_of_love b ON (b.id = a.id)
WHERE a.my_value != b.my_value;

Swap table names so that you can maintain a back up in case you need to rollback.

RENAME TABLE my_table_of_love TO my_table_of_love_OLD;
RENAME TABLE my_table_of_love_NEW TO my_table_of_love;

Proceed to regression testing.

This approach becomes more and more preferable with tables with multiple indexes and millions of rows.

Thoughts?

Related Question