On Wed, Mar 14, 2012 at 10:49 AM, Miles Fidelman <mfidel...@meetinghouse.net> wrote: > Hi Folks,
Hi. > > I'm currently running a pretty basic high-availability configuration for our > mail server (postfix) - it simply runs in a Xen virtual machine, with > mirrored disks across two machines (DRBD), and failover of the VM if > something goes wrong (pacemaker). > > I'm thinking about migrating the failover host to a 2nd datacenter - which > makes disk mirroring and VM migration a bit trickier, and I really don't > like how brittle all that infrastructure is, so I'm starting to think about > application layer redundancy - two mailservers, at remote locations, > multiple DNS records, and doing something to replicate ques, configurations, > and local delivery. The goal is the same: keep processing mail if a > machine goes down, and don't lose any data to machine or disk crashes. > > Which leads to a question: Are any of you running such a configuration? If > so, can you describe what you're doing? And.. are there any good Well, first question here: how much traffic are you going to handle? And now, my experience (please, postfix-list purists, stop reading now, this is more related to DRBD than it is to postfix): I have a HA cluster with two nodes on two locations, on softlayer, due that softlayer provides unlimited inter-server connectivity (please, if someone knows another hosting company that does this -unlimited communication between servers in different DCs-, let me know: softlayer is quite expensive), I'm just using the private network (that use to run at 200~500Mbps) to replicate the DRBD volume. I had several issues, but I suggest you try, and then post on the corresponding lists (DRBD, pacemaker, corosync, heartbeat, ....). I have VM-level failover here, but it is pretty much the same to setup service-level failover. About multiple DNS records, etc... I just used low TTL DNS, and a dynamic DNS setup, so that the VM updates the DNS record on failover. On a side note: I personally believe that service-level HA configuration is better than VM-level. > references, presentations, etc. that anybody knows about re. building > high-availability, scalable, distributed mail processing infrastructure? You can use postfix's mail routing capabilities to have distributed mail processing, ie: have some users on one server, and others at the other server... it is neat. Sincerely, Ildefonso Camargo
