Tino Wildenhain wrote:
the standard approach for master-master replicated data is to use
an UUID-Datatype (see http://en.wikipedia.org/wiki/UUID )

Good point, although we only have space for 64 bits, so the chance of collision would be higher than it usually is in a UUID system.

That is a serious issue that will lead to database corruption. The commit and pack locks need to be cluster-wide. Does MySQL have a way to do that?

Well, synchronous replication is a hard beast, master-master synchronous
even harder (and you have to ask what problem you really want to solve
with it, since the trade-offs are massive)

True. Master-master replication can be done, but it often turns out to perform *worse* than a single master, and the mere notion of a cluster-wide lock frightens little children who want their daddy to come home rather than debug clusters all night.

I'd rather look at master-slave replication with a little enhancement: RelStorage already has distinct load and store connections, so it could choose a slave when loading and the master when storing.


For more information about ZODB, see the ZODB Wiki:

ZODB-Dev mailing list  -  ZODB-Dev@zope.org

Reply via email to