Paul Dekkers schrieb am Thu, Mar 07, 2002 at 02:11:11PM +0100: > Hi > > I'm looking for a redundant mail storage. Not just a seconday MX where mail > is temporarily stored on a second server, but just that when one server > crashes another server goes on with its normal tasks. > For mail I think it would be nice if all new messages go to both servers, > and when a message is read, it is removed from both servers. During downtime > then one server goes on with the normal tasks, and after some time the > server sends the server that is back up the data it missed. > > Using a single mailfile this is not possible I think, unless you write > something with a procmail script that delivers to two mailfiles, and hack > the pop3-daemon to delete files from both hosts. > With maildir however, this looks much better. Something like unison would be > nice then, since it seems to support bidirectional synchronisation; and > since there is no central database of the messages stored, it uses a diff to > change the other server. (Rsync is worse, because of the unidirectional > synchronisation I think... I might be wrong.) > Maybe also coda is a solution, but I never looked at it soo close, and it > looked a bit complex at first.
You have two possible solution that I know workingi in reality. It depends on your money you'd like to spend which one you choose. First you can "cluster" two systems together via single SCSI bus and then use www.missioncritical.com's kimberlite to set up services that'll be switched upon failure. Data will reside on one partition which is mounted by the machine that deems itself active. This is a very smart solution, I have tested it with collegues. Anyway, it's somewhat expensive. Second (esp. if you cannot afford the hardware described above) you may set up the heartbeat package from www.linux-ha.org. It is very easy to create failover mechnisms with heartbeat, but there is still the problem of data synchronisation (heartbeat does not deal with that). I myself use heartbeat together with rsync, which is surely unidirectional but can be turned into somewhat "bidirectional" in combination with heartbeat: The "active" machine will permanently send its data to the backup machine (simple as that: I use a cronjob that tests whether we are active or not). Upon failfailover, the situation will reverse, which leaves data intact and up-to-date. However, note that rsync is slow, if you have lots of data, you should definitely think over the first alternative. Regards, Birger
