[ovirt-users] Ovirt resilience policy / HA

[Guillaume Penin]

Le 2015-03-18 11:18, Omer Frenkel a écrit :
> ----- Original Message -----
>> From: "Guillaume Penin" <guillaume at onlineacid.com>
>> To: users at ovirt.org
>> Sent: Monday, March 16, 2015 10:34:54 PM
>> Subject: [ovirt-users] Ovirt resilience policy / HA
>> 
>> Hi all,
>> 
>> I'm building a test ovirt (3.5.1) infrastructure, based on 3 ovirt 
>> nodes
>> and 1 ovirt engine.
>> 
>> Everything runs (almost) fine, but i don't exactly understand the
>> interaction between resilience policy (Cluster) and HA (VM).
>> 
>> => What I understand, in case of host failure :
>> 
>> - Setting resilience policy to :
>> 
>>      - Migrate Virtual Machines => All VMs (HA and non HA) will be
>> started on another host.
>>      - Migrate only Highly Available Virtual Machines => HA VMs only 
>> will
>> be started on another host.
>>      - Do Not Migrate Virtual Machines => HA and non HA VMs won't be
>> started on another host.
>> 
>> => In practice :
>> 
>>      - No matter what parameter i use in resilience policy, HA VMs 
>> only
>> will be started on another host in case of a host failure.
>> 
>> Is this the expected behaviour ? Am I misunderstanding the way it 
>> works
>> ?
>> 
> 
> there are 2 types of host failure:
> 1 - power/network/vdsm service.. - these leads to a state where engine
> and host cannot communicate, host will move to 'non-responsive' and
> will be fenced (power mgmt action)
> in this case, only HA vms will be restarted, once the engine is sure it 
> is safe.
> 
> 2 - software issues like storage connectivity - in this case, engine
> can communicate with the host, and decide it is not fulfilling the
> cluster/dc requirements
> the host will move to 'not operational' then the engine will
> live-migrate (not restart) vms according to your choice of resilience
> policy
> 
> so what did you do when you tested 'host failure'?
> according to your question i assume number 1, and this is what you got
> 
> let me know if it helps
> Omer.

Hi Omer,

Indeed, I only tested the "Non responsive" scenario with a power 
failure. The host was fenced and only HA VMs were restarted which was, 
as you said, the correct behaviour in this case.

I've just tested the "Not operational" scenario by "ifdowning" a network 
interface. Resilience policy was configured to "Migrate Virtual 
Machines" and as expected, all VMs have been migrated to another host.

Thank you very much for that clarification.

>> Kind regards,
>> _______________________________________________
>> Users mailing list
>> Users at ovirt.org
>> http://lists.ovirt.org/mailman/listinfo/users
>>