On Tue, Jul 14, 2009 at 6:20 AM, Aleksander
Kmetec<aleksander.kmetec at intera.si> wrote:
> Replication is working fine now with the 6 hour cleanup interval, but I'd
> still like to know if this is a slony bug or if there's something strange
> going on with our cluster.

I don't have any multi-hour transactions, but it seems to me that
given slony uses the transactions to its advantage to enter rows into
the change log, that any rows in the change log wouldn't be visible to
any other transaction until the commit, and thus the cleanup action
could not see them to clean them until that time.  So it would really
have to be all or nothing for a transaction.

When that big transaction commits, how long does it take to replicate
that to the other node?  What if you cut your cleanup interval to just
under that amount of time and/or just over that time. Would you still
lose rows?