[virt-tools-list] live-migrating a VM whose current host's storage is fubar'ed

[Digimer]

On 16/04/14 02:17 PM, Cole Robinson wrote:
> On 04/15/2014 05:16 PM, Digimer wrote:
>> Hi all,
>>
>>    For those on the linux-cluster list, this is a similar question to my
>> post[1] there.
>>
>>    I recently had a client destroy their node's RAID array (they yanked hard
>> drives until the array was destroyed). The VMs running on the node continued
>> to operate for another hour, impressively. Their storage was based on DRBD, so
>> the VM's storage just started going to/from the other node's storage.
>>
>>    The problem was when I tried to migrate them off of the damaged node to a
>> healthy node. Without being able to write to the logs (I assume), the
>> migration failed. In the end, I had to force-off the damaged host and reboot
>> the VMs on the backup node. This caused an outage which, I think, could have
>> been avoided.
>>
>>    So my question is; Can VMs be migrated if/when the host's FS is
>> read-only/destroyed? If so, how?
>>
>> Thanks!
>>
>> 1. https://www.redhat.com/archives/linux-cluster/2014-April/msg00010.html
>>
>
> Probably a better question for libvirt lists, CCd, but I don't think anyone
> tests a scenario like that so there's bound to be problems.
>
> - Cole

Thanks for cc'ing.

Ya, I'm not surprised that I had problems. I guess now that I've hit it 
though, and given that it's not impossible to have this happen again, it 
would be worth looking in to a way to handle it. I suspect something 
like "--no-logging" to disable attempts to write to local disk during a 
migration would be enough.

-- 
Digimer
Papers and Projects: https://alteeve.ca/w/
What if the cure for cancer is trapped in the mind of a person without 
access to education?