[ovirt-users] Two ovirt-engine manage one hypervisor

[Yedidyah Bar David]

On Thu, Jun 16, 2016 at 11:19 AM, Piotr Kliczewski
<piotr.kliczewski at gmail.com> wrote:
> You had to test it with 3.5. In 3.6+ we support more clients so the
> connection drop would be not the issue anymore.
>
> There are number of logical issues that would make your env not usable
> like Michal mentioned. Assuming at least two hosts it could happen
> that different hosts would become SPM for different engines.
> Engine is not designed to coexist with other engine and I suggest to
> not take this path.

Even two engines that something external (clustering tool) prevents them
from running in parallel? With same database?

> You may want to consider using hosted engine or
> use standalone engine which is able to recover after a crash.

Already mentioned, but I understand people that want to explore
other options. In principle it might provide faster reaction to
failure, depending on implementation.

>
> On Thu, Jun 16, 2016 at 10:17 AM, Yedidyah Bar David <didi at redhat.com> wrote:
>> On Thu, Jun 16, 2016 at 10:13 AM, Michal Skrivanek
>> <michal.skrivanek at redhat.com> wrote:
>>>
>>> On 16 Jun 2016, at 09:07, Yaniv Kaul <ykaul at redhat.com> wrote:
>>>
>>>
>>>
>>> On Thu, Jun 16, 2016 at 9:29 AM, Yedidyah Bar David <didi at redhat.com> wrote:
>>>>
>>>> On Thu, Jun 16, 2016 at 9:10 AM, Yaniv Kaul <ykaul at redhat.com> wrote:
>>>> >
>>>> >
>>>> > On Thu, Jun 16, 2016 at 8:53 AM, Yedidyah Bar David <didi at redhat.com>
>>>> > wrote:
>>>> >>
>>>> >> On Thu, Jun 16, 2016 at 5:33 AM, Sandvik Agustin
>>>> >> <agustinsandvik at gmail.com> wrote:
>>>> >> > Hi users,
>>>> >> >
>>>> >> > Good day, is it possible to configure two ovirt-engine to manage one
>>>> >> > hypervisor? My purpose for this is what if the first ovirt-engine
>>>> >> > fails,
>>>> >> > I
>>>> >> > still have the 2nd ovirt-engine to manage hypervisor.
>>>> >> >
>>>> >> > is this possible? or any suggestion similar to my purpose?
>>>> >>
>>>> >> The "normal" solution is hosted-engine, which has HA - the engine
>>>> >> runs in a VM, and HA daemons monitor it and the hosts, and if there
>>>> >> is a problem they can start it on another host.
>>>> >>
>>>> >> There were discussions in the past, which you can find in the list
>>>> >> archives,
>>>> >> about running two engines against a single database, and current bottom
>>>> >> line
>>>> >> is that it's not supported, will not work, and iiuc will require some
>>>> >> significant development investment to support.
>>>> >>
>>>> >> You might manage to have an active/passive solution - install an engine
>>>> >> on two machines, configure both to use the same remote database, but
>>>> >> make sure only one of them is active at any given time. Not sure if
>>>> >> that's
>>>> >> considered "fully supported", but might come close.
>>>
>>>
>>> even when you make it work when cert issues are sorted out, you need to be
>>> very careful not to bring both engines up managing a same host, they will
>>> fight over it and the monitoring is going to be received only by one of the
>>> engines, which in turn may cause HA VMs restart and split brains all over
>>> the place.
>>
>> And, IIRC from previous discussions, also internal caches etc.
>>
>> But this is not something specific to ovirt-engine - many services
>> have similar restrictions, and common clustering tools allow handling
>> them.
>>
>>>
>>>> >
>>>> >
>>>> > That's not enough - they need to share the same set of certificates...
>>>>
>>>> Best is to simply clone the machine after initial setup then change
>>>> what's needed, or backup/restore only files (engine-backup --mode=backup
>>>> --scope=files).
>>>>
>>>> Didn't check, but I do not think they actually need all the certs of
>>>> all hosts - that is, that it's not mandatory to keep /etc/pki synced
>>>> between them after initial setup. Didn't try that myself.
>>>
>>>
>>> I'm not sure what happens when you provision a host from Mgmt A, then move
>>> to Mgmt B and provision another from it:
>>> 1. Mgmt A won't be aware of that host, from cert req perspective. May not be
>>> such a big deal - donno.
>>>
>>> 2. Can Mgmt A provision another host? Need to ensure the certificate serial
>>> numbers are OK, etc.
>>>
>>> They really need to share the CA DB.
>>
>> Even keeping /etc/pki synced, or mounted from each one before
>> starting the engine and umounting when stopping, should not be
>> too hard.
>>
>>> The backup-restore sounds like good approach  to me.
>>> Y.
>>>
>>>>
>>>> > Y.
>>>> >
>>>> >>
>>>> >>
>>>> >> You can find on the net docs/resources about creating a redundant
>>>> >> postgresql cluster.
>>>> >>
>>>> >> Best,
>>>> >> --
>>>> >> Didi
>>>> >> _______________________________________________
>>>> >> Users mailing list
>>>> >> Users at ovirt.org
>>>> >> http://lists.ovirt.org/mailman/listinfo/users
>>>> >
>>>> >
>>>>
>>>>
>>>>
>>>> --
>>>> Didi
>>>
>>>
>>> _______________________________________________
>>> Users mailing list
>>> Users at ovirt.org
>>> http://lists.ovirt.org/mailman/listinfo/users
>>>
>>>
>>
>>
>>
>> --
>> Didi
>> _______________________________________________
>> Users mailing list
>> Users at ovirt.org
>> http://lists.ovirt.org/mailman/listinfo/users



-- 
Didi