This content has been marked as final. Show 1 reply
I would try to size your production environment as N+1 such that one server could be down and you could still fully support production. Obviously you should be doing regular backups but I would make sure you have really good backups and likely name a particular backup and flag it for retention before any upgrade.
Ideally your test environment would be identical to Production & would be ready to be cannibalized if Prod had a problem but understandably some people can't afford that. Assuming your Prod environment is a HA configuration, at minimum I would make sure your Test was at least two servers in HA so you could model & support prod. I would also make sure Test has test VMs on them that are suitable for regressioning the Prod VMs.
Once that environment is in place I believe through OVM 3 you can take an OVS offline & upgrade it and then bring it back online via the OVM manager. Obviously I would do that first to your Test environment and afterwards make sure your Test VMs are fully functional before going to Prod.
Once you do it in Test then for Prod you leverage the N+1 capacity to be able take down and upgrade each server one at at time... I would still shut down what VMs I could and run it on weekend when most of my users are gone but it is good to know that in theory I shouldn't have to do that.
My first OVS 3.1.1 cluster was actually setup with all 3 in Virtual Box and I modeled the above. I ran into an issue with the upgrade of the OVM Manager itself and left it there. My new env was fresh install of 3.2.1 & I haven't tried to upgrade yet ...
On a related note I am thinking about running my OVM 3.x Manager in Oracle Virtual Box Enterprise fully separate from OVM as I believe it is fully supported now and would allow me to make easy full cold backups and give me snapshots outside of OVS... I am working toward having the following setup:
Prod OVS: Physical N+1 (meaning at least 2 less than half loaded or or 3 no more than 2/3 loaded etc);
Prod OVM: Oracle Virtual Box VM (with paid full/support);
Test OVS: Physical N+1 like Prod (I run Test, Dev, Patch, etc here); If you can't do this I would at least have 2 physical OVS servers in HA with ideally the same motherboard, NICs, etc as Prod though maybe less cores/RAM if budget doesn't allow...
Test OVM: Oracle Virtual Box VM;
Dev OVS: Oracle Virtual Box OVS HA cluster;
Dev OVM: Oracle Virtual Box VM;
When upgrading, I would start in DEV and work my way up the stack.
My rule is no production VMs without a corresponding test VM though presently they are different IPs and different software stacks that I try to have installed identically.... Once I finish migrating all my ovs 2 servers to ovs 3, I plan to figure out a way to have a segregated subnet with the same IPs as my production subnet that I can NAT into and actually have my test VMs be exact copies of my production VMs. That is where I want to get...
Edited by: user11391721 on Apr 4, 2013 4:40 PM to fix typos