[ovirt-users] VM running on multiple hosts

Wed Sep 20 19:18:18 UTC 2017

> On 20 Sep 2017, at 21:08, Yaniv Kaul <ykaul at redhat.com> wrote:
> 
> 
> 
> On Sep 20, 2017 9:50 PM, <support at jac-properties.com <mailto:support at jac-properties.com>> wrote:
> This matches about with what we were thinking, thank you!
> 
> To answer your questions
> 
> We do not have power management configured due to it causing a cascading failure early in our deployment.  The host was not fenced and "confirm host rebooted" was never used.  The VMs were powered on via virsh (this shouldn't have happened)
> 
> The way they were powered on is most likely why they were corrupted is our thought
> 
> 

yes. 
That’s why we put a basic password protection to the plain virsh access. Easy to circumvent, but then you’re on your own…

Hm, how exactly were they powered on by virsh? Normally this is not possible for oVirt VMs at all due to initial set up of host-specific things(disk paths), we also use transient libvirt domains so stopped VMs are not defined in libvirt once they stop. So I wonder how exactly was this done?
Unless they were in Paused state where you indeed can simply continue the execution. 

> 
> We'd be happy if you could share both engine and host logs, including vdsm.log, engine.log and /var/log/messages from both. 
> Y. 
> 
> 
> 
> Logan
> 
>> On September 20, 2017 at 12:03 PM Michal Skrivanek <michal.skrivanek at redhat.com <mailto:michal.skrivanek at redhat.com>> wrote:
>> 
>> 
>>> On 20 Sep 2017, at 18:06, Logan Kuhn <support at jac-properties.com <mailto:support at jac-properties.com>> wrote:
>>> 
>>> We had an incident where a VM hosts' disk filled up, the VMs all went unknown in the web console, but were fully functional if you were to login or use the services of one.
>> 
>> Hi,
>> yes, that can happen since the VM’s storage is on NAS whereas the server itself is non-functional as the management and all other local processes are using local resources
>> 
>>>   We couldn't migrate them so we powered them down on that host and powered them up and let ovirt choose the host for it, same as always. 
>> 
>> that’s a mistake. The host should be fenced in that case, you likely do not have a power management configured, do you? Even when you do not have a fencing device available it should have been resolved manually by rebooting it  manually(after fixing the disk problem), or in case of permanent damage (e.g. server needs to be replaced, that takes a week, you need to run those VMs in the meantime elsewhere) it should have been powered off and VM states should be reset by “confirm host has been rebooted” manual action.
>> 
>> Normally you should now be able to run those VMs while the status of the host is still Not Responding - was it not the case? How exactly you get to the situation that you were able to power up the VMs?

sorry, I meant "Normally you should not be able to run those VMs”

Thanks,
michal

>> 
>> 
>>> However the disk image on a few of them were corrupted because once we fixed the host with the full disk, it still thought it should be running the VM.  Which promptly corrupted the disk, the error seems to be this in the logs:
>> 
>> this can only happen for VMs flagged as HA, is it a case?
>> 
>> Thanks,
>> michal
>> 
>>> 
>>> 2017-09-19 21:59:11,058 INFO  [org.ovirt.engine.core.vdsbroker.monitoring.VmAnalyzer] (DefaultQuartzScheduler3) [36c806f6] VM '70cf75c7-0fc2-4bbe-958e-7d0095f70960'(testhub) is running in db and not running on VDS 'ef6dc2a3-af6e-4e00-aa4
>>> 0-493b31263417'(vm-int7)
>>> 
>>> We upgraded to 4.1.6 from 4.0.6 earlier in the day, I don't really think it's anything more than coincidence, but it's worrying enough to send to the community.
>>> 
>>> Regards,
>>> Logan
>>> _______________________________________________
>>> Users mailing list
>>> Users at ovirt.org <mailto:Users at ovirt.org>
>>> http://lists.ovirt.org/mailman/listinfo/users <http://lists.ovirt.org/mailman/listinfo/users>
>> 
> 
> _______________________________________________
> Users mailing list
> Users at ovirt.org <mailto:Users at ovirt.org>
> http://lists.ovirt.org/mailman/listinfo/users <http://lists.ovirt.org/mailman/listinfo/users>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20170920/2d1c90b8/attachment.html>