On 2010-11-29 16:49, Bron Gondwana wrote: > So set the clocks on the machines out by a few more seconds and you'll > be fine. > > Seriously - the problem is that it works when you test manually because > they're not happening at the same time. "things die" because > replication is asynchronous, and the same UID gets allocated to messages > with different content! > > That said - it should be able to resolve the situation without crashing. > When we shut down one end, we bring up the other. Downtime is a couple > of seconds maximum. Everything gets delivered to postfix locally which > forwards to an LMTP proxy which knows which host is the master, and > delivers appropriately. > > Most MTAs are happy to spool locally generated mail until the server > is back up. Store and forward. > > Bron. Hello, First, I hope my comments were not taken as criticism of the Cyrus model, I like Cyrus enormously and am glad to deploy it ;) I am working instead on a second sendmail process that listens on localhost to forward all mail to the master in order to avoid this "issue". I might disable sendmail on the public IP of the slave entirely unless I need to temporarily use the slave and bring down the master for whatever reason. Glad its just architectural. Thanks again, Chris ---- Cyrus Home Page: http://www.cyrusimap.org/ List Archives/Info: http://lists.andrew.cmu.edu/pipermail/info-cyrus/
