> From http://bucardo.org/slides/b5_multi_master/ >
> Multi-master conflict = same primary key changed on two or more servers
> Built-in method: latest wins (all txns across all databases)
> Custom methods: you decide, but there can be only one winning database
> Not needed the great majority of the time!
> Exceptions and conflict handlers are two examples
> Can also run before the sync, before the copy, after copy, after sync
> After-sync is handy for cache invalidation
> Before-sync can even tell the sync not to run: like an advanced trigger
That's all good to know but doesn't help with what's going on.
The core problem here is that sequences are out of sync.
Either locally or across the replication connection.
I'd need the results of the queries I posted in order to be able
to formulate a more informed guess as to which scenario applies.
Replicating sequences sure sounds like a requirement since Marc
indicated he'll be writing on both ends of the replication concurrently.