Background The Skinny

Performing schema changes often requires extended downtime for applications. This is due to MySQL needing to rebuild tables for common schema change operations. Tools like pt-online-schema-change have been written to try to overcome the downtime associated with schema changes, however they are complex and put a high load on the database. Amazon’s Aurora improves some schema changes operations, but still requires a table rebuild for common operations like adding a column using before or after, or simply to add a column with a default value. Rebuilding a table with millions of rows can take hours and prevent writes to that table the entire time.

How Can Tungsten Clustering Keep Applications Running? How Does It All Work?

…

OK, let me start with saying that a table without a primary key shouldn't be something that a DBA should ever stumble into.  Sure, InnoDB will secretly add one for us without telling - but that's not something we can use at application or administration level, it's just to make the engine happy.

So let's suppose you find some tables that lack a primary key, of course you need to do something about it, right?  Now, put Galera Cluster in the mix - Galera does not support tables without a primary key and will, secretly again, make that table inconsistent at cluster level.

You need to fix the damn table(s)!! And this is where the fun begins...  as you can't afford downtime for the operation so you need to resort to an online schema change of some type.

Galera cluster is …