Re: Azure Mysql NDB Cluster

[Mikael Ronström]

I am almost sure that what you should do is to
change the tables to use the READ BACKUP feature.
I've been blogging about the impact of this in a
number of blogs this week.

What happens here is that when you add the second
data node and don't have the READ BACKUP feature,
the reads will be spread among the VMs and thus
a lot of extra network jumps is caused.

You can read those new blogs to understand a bit more
about performance in various setups. There will be
more blogs coming next week.

It is possible to change an existing table to use
READ BACKUP as an ALTER TABLE that can be performed
online. I suggest you also add ndb_read_backup=on
to your my.cnf to ensure all tables use the READ BACKUP
feature. It is likely this will be default in some
future version of MySQL Cluster.