[ovirt-users] oVirt 3.6 on CentOS 6.7 based HyperConverged DC : Upgradable?

[Yedidyah Bar David]

On Wed, Mar 29, 2017 at 5:16 PM, Nicolas Ecarnot <nicolas at ecarnot.net> wrote:
> Le 29/03/2017 à 15:54, Yedidyah Bar David a écrit :
>>
>> On Wed, Mar 29, 2017 at 4:35 PM, Nicolas Ecarnot <nicolas at ecarnot.net>
>> wrote:
>>>
>>> [Please ignore the previous msg]
>>>
>>> Hello,
>
>
> Hello Didi,
>
>>> One of our DC is a very small one, though quite critical.
>>> It's almost hyper converged : hosts are compute+storage, but the engine
>>> is
>>> standalone.
>>
>>
>> And you intend to keep it that way? You didn't mention below.
>
>
> I intend to keep it this way.
>
>>> At first glance, I would go this way (feel free to comment) :
>>> - upgrade the OS of the engine : 6.7 -> 7.3
>>
>>
>> How? This isn't supported, in principle, although it might work.
>
>
> Mmmm, yep, you're right. Many people documented it.
> I'm not very fond of playing for such a critical part.
>
>> The "official" way is using engine-backup to backup and restore it.
>> See also:
>> https://bugzilla.redhat.com/show_bug.cgi?id=1332463
>
>
> Ok,
> Sent you a question there.

Replied.

>
>>> During this upgrade, I have no constraint to keep everything running,
>>> total
>>> shutdown is acceptable.
>>
>>
>> So you can also do a full backup and restore.
>
>
> I could.
> But I may run out of hosts at present.
>
>> Create an NFS export domain
>> somewhere, export all your VMs there, recreate everything from scratch,
>> import the VMs. Will take much longer, but then you don't need to risk/
>> test/prepare for problems in upgrading gluster.
>
>
> Well, so your overall opinion is that I should stick to KISS, correct?

I didn't say that. I don't know gluster.

I just said that there *is* an option to KISS, if you want to.
Some people prefer to take the "smarter" path (upgrade in-place), and
it might work for them. If you want that, I simply suggest to spend
some more time both reading and testing gluster upgrades.

Personally I did both, and preferred backup/restore on more critical
systems, at least when I had the option. If you do your math, test,
and reach a conclusion that full backup/restore (with an export
domain) will simply take more time than your downtime window, this
is simply not an option... OTOH if you know that gluster upgrade
is bullet-proof, from both testing and reading other people's
reports, then go with it. Personally I'd still backup everything
beforehand, if I have the option :-)
-- 
Didi