Xtradb cluster plus master/slave replication

Hi everybody,

we currently have a 3 nodes xtradb cluster. In order not to “disturb” these nodes with specific tasks like backup,
or heavy calculation, we want to set up slave nodes, the master being the Xtradb cluster.

Do you think it is a good architecture ? By good, i mean : robust, scalable, maintainable.

Thanks very much for any help,

Hi ath;

As long as the cluster node you are going to hang the slave off of has log_slave_updates enabled, then you can hook up to it via standard asynchronous replication. For what you are looking to do, this would be a standard solution to that problem. However, it is not necessarily robust/scalable/maintainable, as it is using standard replication and has many of the downsides of why you probably went to a cluster to start with.

The main issue is that if the one cluster node the slave is hooked up to fails, then that will break replication to the non-cluster slave, which will likely then have to be rebuilt. So if your application is hitting that non-cluster slave for reads, you’ll want to make sure it includes some logic to detect if it is out of date via a heartbeat style check of some sort (i.e. a table with a timestamp that gets replicated, and you can check how recent that time stamp is compared to the master).

Given that, you then have a single point of failure. This may be fine if you are using the non-cluster slave for non-critical items. Alternatively you could set up a few of these non-cluster slaves, attached to different cluster nodes, and use some sort of load balancer / proxy to manage reads to the non-cluster slaves. That way if one one of the non-cluster slaves goes down, you have another one that is attached to a different cluster node that is probably still working. However this gets you into higher complexity that may or may not defeat the purpose of the initial simple solution. =)

-Scott