Hi, fellow admins. We'd like to configure a backup of our primary e-mail server, which runs qmail-ldap 20031101a on a 2 power supply, 2 CPU, 3 eth, RAID-5 SCSI Intel server (so it should never totally collapse and send all data to /dev/null, but we're playing being paranoid here).
The idea is having an as-exact-as-possible replica of the primary server (same distro, same qmail version, same config...) on a lesser machine, so that in the event of a total catastrophe on the primary one, we can manually (or automatically by means of mon, heartbeat or whatever) move all our mail services to the backups server in a matter of minutes. The problem of course lies in keeping all the contents of both servers synchronized. The LDAP users database isn't a problem thanks to slurpd. The real problems are: - the mail queues - the mailboxes We think that keeping the queues in synch isn't worth it (too complex): messages in the out-queue for too long will be (likely) discarded anyway after a copule of days, messages which were just arriving in the moment of the crash will be re-sent by the other-end servers, and messages just arrived but still undelivered to the local mailbox... well, I guess we'll have to cope with that. Ideas we have for syncing the mailboxes: - running rsync evey n minutes: really easy, but suboptimal (messages received since the last sync but not downloaded will be lost, messages downloaded since the last sync before the crash will be downloaded again). - sending all incoming e-mails via qmqp to the secondary server. Optimal as all incoming mails will be immediately sent to the backup, but how do we delete mails as the users read them from the primary server? With rsync again? Will they have the exact same name on both servers? - abusing in some way the clustering capabilities in qmail-ldap? Not suer about this one as I've never used qmail-ldap clustering before. Has any of you something like this running? Any other ideas or suggestions, besides the classical "NFS server for the mailboxes + n SMTP/POP heads"? Is this all too complicated, or over-paranoid? Any idea will be really appreciated. Thanks in advance. -- Vicente Aguilar <[EMAIL PROTECTED]> Departamento de Sistemas Tlf.: 965 98 71 92 Recursos en la Red, S.L.U. http://www.renr.es
