[Bucardo-general] Multi master-master replication
Greg Sabino Mullane
greg at endpoint.com
Mon Feb 25 15:11:04 UTC 2008
On Thu, 2008-02-14 at 02:55 +0100, Matic Meznar wrote:
> My project has a special need, it needs a peer-to-peer (multi
> master-master) replication between the database servers involved in
> providing the data storage for my application. This web application is
> spread among several servers, each at a different location, all over
> the world. It is essential that data is replicated between them within
> 5 seconds and that each node supports read and write operations (act
> on its own).
> While reading the documentation for Bucardo, I have found out that it
> only supports 2 nodes.
> When (3 months, 6 months, a year?) is multi master-master replication
> going to be supported?
> Are there any alternatives to Bucardo that would provide this kind of
Not for Postgres that I am aware of. As far as when Bucardo will support
it, it is hard to say. Expanding it to more than two master nodes is
quite possible in theory, and most of the infrastructure is there.
Whether or not it gets added depends on demand, available time, if a
company decides to sponsor the feature, etc.
> Oracle and MSSQL have what I need, but they are either too complex and
> resource intensive or require Windows to run (which is not an option).
> MySQL does not support multi master-master replication (circular
> replication is a joke).
How many nodes do you eventually envision? Bucardo should within a year
take advantage of asynchronous queries which will have its greatest
impact on master-master replication, but the five second rule may not be
practical with a large number of nodes (and may be most limited by
network speeds of course).
It may be that Bucardo may not even be the answer to what you are
looking for, which sounds very peer-to-peer: Bucardo will still need a
"central" Bucardo instance to coordinate between all of the masters, and
is not currently designed as a distributed multi-master system. It's
basically a central server that connects to all the others and swaps
data between them. It could certainly be set up with a heartbeat system
to have Bucardo failover (which should work well as no data is stored on
the central server, and the updates are atomic).
Greg Sabino Mullane greg at endpoint.com
End Point Corporation 610-983-9073
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 163 bytes
Desc: This is a digitally signed message part
Url : https://mail.endcrypt.com/pipermail/bucardo-general/attachments/20080225/ef5dffe4/attachment.bin
More information about the Bucardo-general