On Wed, 17 Mar 2010 10:08:19 -0500, "Savino, Candice" <CSavino@stripped> wrote:
> What would be the most stable way of moving data from a cluster to a read
> only db? If I did row based replication from ndb to an innodb engine is
> that stable? I know that statement based replication will not work with
> ndb. Is it more stable to do row based replication from a cluster to a read
> only cluster?
InnoDB can work, as since it is transactional you can recover from the
slave crashing. For engines such as MyISAM, this woudl not be possible.
The first challenge is getting the data there though. Typically, when
setting up a slave that's running NDB you can use ndb_restore to get to
a known state. Not so with innodb as the slave. One idea is to have an
NDB slave and stop replicating to it, do an sql dump, then resume
replication and use that SQL dump to bootstrap a InnoDB slave.
You cannot just use a SQL dump on the master if you have ongoing
transactions (as NDB does not support REPEATABLE_READ isolation level).