> > No, I meant that stable has slony for 8.4, and testing (the next stable) > has slony for 9.1, which means someone upgrading from stable to the next > stable won't experience a regression, except that they will need to make > sure they use the repeatable read isolation level. > > Testing is currently frozen, so putting in a new major upstream release > is not possible. >
The problem is that slony 2.0.x (slon) opens up serializable transactions that caues the conflicts. This can happen even if user applications only perform repeatable read transactions. The transactions will be retried after they are aborted so they should eventually get through. If updating the upstream release isn't possible then it isn't possible, there isn't much that I can say to that. The feedback we received from sites that deployed the PG 9.1 + Slony 2.0.7 combination is what lead to the bug report and fix in 2.1 in the winter.