On Nov 12, 2009, at 6:41 PM, Rich Megginson wrote: > Robert Viduya wrote: > I don't see anything which indicates that someone initiated a > replica init from this master. I would expect to see a MOD on the > replication agreement entry, and I don't see that. That's the issue, no-one initiated a replica init. But somehow, that master saw one come over the wire. > Are you using a BigIP or some other type of load balancing device > between the two servers? If so, I wonder if that is somehow causing > problems. We are using a BIGIP to load-balance our customer's access, but obviously you can't load-balance replication. Our replication agreements point to the servers themselves, not to the advertised service on the load balancer. It does look like, from the logs, there were some communication errors, but we can't avoid those. The replication protocol should be robust enough to handle that.
