SM> It is the same. Replication works already with all table handlers,
SM> but all that remains to be solved is ROLLBACK.
What could be the problem is servers are recording in their
binary log only committed transactions ?
Why should ROLLBACK be implemented in replication ?
If only committed transactions are recorded in the binary log then
if a server crashes and it has to rollback it's uncommitted
transactions, this has no impact on the replicating scheme isn't it
Let me know if I'm wrong.