On Mar 26, 2007, at 5:10 , Sylvain Coutant wrote:
In fact, my comments were stupid. What happens is we go split brain in this case and half of the requests are executed on one controller and other half on the other. Finally, we end with backends that have different data stored in and *everything* is lost, we have to revert all databases to the last dump, losing several hours (if we were in production) of very valuable data (if they were not really valuable, they would not be served by such a replication solution ;-)
You could have a resync mechanism if split brain can be avoided. This is not easy to reach, but can be done using primary partitions. We are building a protocol that provides primary partitions, so maybe it can be used for this in the future.
Cheers, -- Nuno Carvalho University of Lisbon, Portugal http://dialnp.di.fc.ul.pt
_______________________________________________ Sequoia mailing list [email protected] https://forge.continuent.org/mailman/listinfo/sequoia
