I am currently trying to work my way our of a single point of failure
situation with the MySQL database structure that I manage.
Currently I have a single Master Server with 2 Slave Servers in read-
only mode. Round-Robin DNS shares the load between these.
Due to a desire to increase availability, I am looking to replicate
this setup in a second AmazonAWS datacenter and can see a case for
using 3 nodes at some point in the future.
Galera seems to offer a solution to my problem.
Is it possible to build a Master Replication cluster (the Master in
each datacenter) (so 2 or 3 masters) and then have each Master have 2
or more read-only Slaves doing "traditional" MySQL replication.
The purpose for wanting to do this is to reduce the amount of inter-
datacenter traffic.
It is possible. From each Galera node you could fan out a number of MySQL replication slaves.
However, now if your one galera node goes down, then that whole data center is disconnected. What is worse: if a node is disconnected long enough that it cannot do incremental state snapshot, then it will need to take a full copy of another galera node. Your replication slaves are now useless, you need to re-provision them too from the same backup.
If you don't plan to have dozens of read-only nodes, I would strongly consider using just galera clustering for all nodes. The simplicity is worth the cost of more traffic.
<david.b...@whizzeducation.com> wrote: > Hi there.
> I am currently trying to work my way our of a single point of failure > situation with the MySQL database structure that I manage.
> Currently I have a single Master Server with 2 Slave Servers in read- > only mode. Round-Robin DNS shares the load between these.
> Due to a desire to increase availability, I am looking to replicate > this setup in a second AmazonAWS datacenter and can see a case for > using 3 nodes at some point in the future.
> Galera seems to offer a solution to my problem.
> Is it possible to build a Master Replication cluster (the Master in > each datacenter) (so 2 or 3 masters) and then have each Master have 2 > or more read-only Slaves doing "traditional" MySQL replication.
> The purpose for wanting to do this is to reduce the amount of inter- > datacenter traffic.
> Thanks
> David
> -- > You received this message because you are subscribed to the Google Groups "codership" group. > To post to this group, send email to codership-team@googlegroups.com. > To unsubscribe from this group, send email to codership-team+unsubscribe@googlegroups.com. > For more options, visit this group at http://groups.google.com/group/codership-team?hl=en.
Henrik is right. Unfortunately your suspicions are correct too: if you have 3 Galera nodes at each datacenter, the traffic between datacenters is tripled. One day this will be fixed, but no ETA.
However replication traffic is normally way smaller than the client one, so there is a good chance it does not add much to out-of-datacenter traffic you already have.
> I am currently trying to work my way our of a single point of failure > situation with the MySQL database structure that I manage.
> Currently I have a single Master Server with 2 Slave Servers in read- > only mode. Round-Robin DNS shares the load between these.
> Due to a desire to increase availability, I am looking to replicate > this setup in a second AmazonAWS datacenter and can see a case for > using 3 nodes at some point in the future.
> Galera seems to offer a solution to my problem.
> Is it possible to build a Master Replication cluster (the Master in > each datacenter) (so 2 or 3 masters) and then have each Master have 2 > or more read-only Slaves doing "traditional" MySQL replication.
> The purpose for wanting to do this is to reduce the amount of inter- > datacenter traffic.
> Thanks
> David
-- Alexey Yurchenko, Codership Oy, www.codership.com Skype: alexey.yurchenko, Phone: +358-400-516-011
I mean, as far as I understand how Galera works, the data should flow from node to node only once, i.e. as I add additional node, the amount of time the data is replicated increase by one.
This means in 3 data-centers scenario, the data will flow from one node to another, i.e. 2 times rather then 3.
> I mean, as far as I understand how Galera works, the data should flow > from > node to node only once, i.e. as I add additional node, the amount of > time > the data is replicated increase by one.
I guess you have a typo there: "amount of data", not "amount of time".
> This means in 3 data-centers scenario, the data will flow from one > node to > another, i.e. 2 times rather then 3.
> Thanks.
You're right in your assumptions. However, I understood it so that the idea is to have 2 datacenters, 3 nodes each. This way external traffic triples.
