Databases Reference
In-Depth Information
Master-Master with Replicas
A related configuration is to add one or more replicas to each co-master, as shown in
Figure 10-8
.
Figure 10-8. Master-master topology with replicas
The advantage of this configuration is extra redundancy. In a geographically distributed
replication topology, it removes the single point of failure at each site. You can also
offload read-intensive queries to the replicas, as usual.
If you're using a master-master topology locally for fast failover, this configuration is
still useful. Promoting one of the replicas to replace a failed master is possible, although
it's a little more complex. The same is true of moving one of the replicas to point to a
different master. The added complexity is an important consideration.
Ring Replication
The dual-master configuration is really just a special case
11
of the ring replication con-
figuration, shown in
Figure 10-9
. A ring has three or more masters. Each server is a
replica of the server before it in the ring, and a master of the server after it. This topology
is also called
circular replication
.
Rings don't have some of the key benefits of a master-master setup, such as symmetrical
configuration and easy failover. They also depend completely on every node in the ring
being available, which greatly increases the probability of the entire system failing. And
if you remove one of the nodes from the ring, any replication events that originated at
that node can go into an infinite loop: they'll cycle forever through the chain of servers,
because the only server that will filter out an event based on its server ID is the server
that created it. In general, rings are brittle and best avoided, no matter how clever you
are.
11. A slightly more sane special case, we might add.
