On 05/12/2014 12:43 AM, Paul Heinlein wrote:

On Sun, 11 May 2014, Itamar Heim wrote:

 I just inherited an oVirt cluster at work that's running 3.2 on
 Fedora 18 and would dearly love some direction about updating
 things without a system-wide downtime.

1. no downtime should happen, as you can upgrade hosts by moving
   them to maintenance, which will live migrate VMs running on them.

2. fedora 18 is EOL, so 'yum update' isn't going to help here.

3. re-install host should refresh vdsm and its dependent packages if
   the repo's configured have newer versions. when the host is in
   maint, you can simply run 'yum update' on the host, then
   re-activate it (of course reboot if kernel got updated, etc.)

I've followed those steps, and I can get VMs migrated *to* the updated
nodes. I have two worries:

  1. VMs migrated to the updated nodes cannot be migrated back to the
     old nodes. So once I start, it's all-in or nothing. Is that to be
     expected?

on fedora hosts this can happen, since there is no live migration compatibility between fedora versions.
on .el6 hosts this shouldn't happen[1]

  2. Once all the nodes are backed up, is there a fairly sure-fire way
     to update the master engine? Is that documented somewhere?

a. you should be able to upgrade engine or hosts regardless of each
   other.
b. the setup (upgrade) script takes a db backup just in case. if you
   happen to run it in a VM, snapshot/backing it up doesn't hurt as
   well.
c. if upgrading to a new minor version, note to change in the upgraded
   engine the cluster and DC level to benefit from new features.

Thanks!

[1] there is a known caveat is you have selinux disabled on some machines and enabled on others.