Comments on: Using MMM to ALTER huge tables

By: Log Buffer #91: a Carnival of the Vanities for DBAs

Log Buffer #91: a Carnival of the Vanities for DBAs — Fri, 04 Apr 2008 16:43:34 +0000

[...] huge queries to huge tables. On the MySQL performance blog, Aurimas Mikalauskas walks us through using MMM to ALTER huge tables. He writes, “When it comes to changes that really require table to be rebuilt – [...]

By: peter

peter — Thu, 03 Apr 2008 03:37:41 +0000

Another note – besides ALTER this works for other similar type of statements. For example you can do online OPTIMIZE table this way.

By: peter

peter — Tue, 01 Apr 2008 04:01:10 +0000

Tobias,

In theory you can use this VIEW approach for complex migrations but in practice there can be many performance gotchas you can run into.

By: peter

peter — Tue, 01 Apr 2008 03:59:20 +0000

Keith,

This article is by Aurimas, lets give him a credit !

By: Simetrical

Simetrical — Fri, 28 Mar 2008 17:29:45 +0000

This is no different from if you use master-slave replication, surely? The only caveat there is that you need to promote one of the slaves to a master if you do this, but that shouldn’t be a problem. Wikipedia does essentially this to alter tables without downtime, and of course it uses slave replication, not master-master.

By: Tobias Petry

Tobias Petry — Fri, 28 Mar 2008 13:36:18 +0000

Great idea, but what do you do if you have to alter the structure due to better performance?
While rading this article i had an idea, i would be really happy if you could comment it:

At first you create in both Masters views which represent the new data structure (maybe you have to alter the relationship of tables). Then you will change application to use these new views instead of the old tables, because the views will later be tables.
Now you turn off one of the masters, and change the tables according to the views you have created. This will take a long long time…
When all changes are ready you can turn the master on, and it can load all changes from the master who was online. Now you have got one master with all data and the new structure. Now you can copy the data to the other master who was online all the time.

By: Keith Murphy

Keith Murphy — Fri, 28 Mar 2008 00:48:33 +0000

Nice article Peter. Thanks for the example!!

By: Mark Callaghan

Mark Callaghan — Fri, 28 Mar 2008 00:00:03 +0000

This is a great way to do it without having the complexity in the MySQL server from trying to do it online as Oracle might. I hope that MySQL doesn’t spend too much time trying to support that. The MMM approach works for most schema changes with far less complexity and risk. But if MySQL spends time on that, then this is an opportunity for MMM as it works right now. The problem gets more interesting as you increase the number of slaves on which the change must be done.

When can we expect the merger of MMM + Maatkit (== MMMaatkit)?

By: Chris

Chris — Thu, 27 Mar 2008 22:09:51 +0000

If you’re just doing an ALTER, a simple way to do it is run:

SET SQL_LOG_BIN=0;
ALTER TABLE …
SET SQL_LOG_BIN=1;

First on one server then on the other, switching masters each time.