Multisite and multiserver HA ZCS in XenServer Hey all,
I need some help on what is either very advanced ZCS setup, or a lack of understanding on my part.
We're currently in the very early stages of developing a new infrastructure, which includes replacing our nice stable and scalable in-house qmail based mail system with something with lots of flashy bells and shiny whistles. ZCS being the main contender for me.
The basics of this new infrastructure is likely to be 2 sites, each with 2 SANs, and a set of servers running Citrix XenServer. The sites will be active/backup for DR. There may even be a third backup-standby site with slightly different equipment.
For HA each site we will require 2 of each Zimbra service. Be that a single pair of servers running all services, or separate pairs of LDAP/MTA/etc servers, or a cluster running everything. At the moment we are not looking at having servers in each office, although it is not out of the question.
ZCS has some clustering features, but I'm a bit lost to how we could fit it into the above setup.
The docs suggest HA is only actually possible on physical machines, with direct access to the same LUN(s). But neither of these is possible with XenServer.
There can always be the possibility of exporting the ZCS LUN(s) as iSCSI and attaching them directly to the VMs, which should sort the shared storage bit out.
I do have experience of a 2 VM cluster with data on GFS on top of DRBD, inside Xen. But that was with Gentoo and plain heartbeat. It wasn't very reliable though.
Also, as I mentioned, there will be more than 1 physical site.
Are there any tools, or in-built functionality, to allow standby servers without shared storage access?
For the interactive, customer facing, services it's naturally critical that the service is up almost instantly a failure is detected and acted on. For this we will be having the VMs running and available at all times (requests will just not be directed to the standbys). Databases will be constantly rolling forward redo logs/WALs, and other local storage changes kept in sync with rsync.
ZCS isn't a interactive, customer facing, service so we could afford a delay while VMs boot and sort themselves out. I'd like to just announce a specific route from the standby DC into our internal routing domain and have traffic flow to ZCS automatically, but without data that no more than a few tens of minutes old it's not a great deal of use.
Thanks
Mike |