Terry hey <recreationh@gmail.com> writes:

> Dear Milan,
> Today, i just found that oVirt 4.2 support iLO5 and power management was
> set on all hosts (hypervisor).
> I found that if i choose VM lease and shutdown iSCSI network, the VM was
> shutdown.
> Then the VM will migrate to another host if the iSCSI network was resumed.

If the VM had been shut down then it was probably restarted on rather
than migrated to another host.

> If i just choose enable HA on VM setting, the VM was successfully migrate
> to another hosts.

There can be a special situation if the storage storing VM leases is
unavailable.

oVirt tries to do what it can in case of storage problems, but it all
depends on the overall state of the storage – for how long it remains
unavailable, if it is available at least on some hosts, and which parts
of the storage are available; there are more possible scenarios here.
Indeed, it's a good idea to experiment with failures and learn what
happens before real problems come!

> But i want to ask another question, what if the management network is down?
> What VM and hosts behavior would you expect?

The primary problem is that oVirt Engine can't communicate with the
hosts in such a case.  Unless there is another problem (especially
assuming storage is still reachable from the hosts) the hosts and VMs
will keep running, but the hosts will be displayed as unreachable and
VMs as unknown in Engine.  And you won't be able to manage your VMs from
Engine of course.  Once the management network is back, things should
return to normal state sooner or later.

Regards,
Milan

> Regards
> Terry Hung
>
> 2018-02-28 22:29 GMT+08:00 Milan Zamazal <mzamazal@redhat.com>:
>
>> Terry hey <recreationh@gmail.com> writes:
>>
>> > I am testing iSCSI bonding failover test on oVirt, but i observed that VM
>> > were paused and did not migrate to another host. Please see the details
>> as
>> > follows.
>> >
>> > I have two hypervisors. Since they are running iLO 5 and oVirt 4.2 cannot
>> > support iLO 5, thus i cannot setup power management.
>> >
>> > For the cluster setting, I set "Migrate Virtual Machines" under the
>> > Migration Policy.
>> >
>> > For each hypervisor, I bonded two iSCSI interface as bond 1.
>> >
>> > I created one Virtual machine and enable high availability on it.
>> > Also, I created one Virtual machine and did not enable high availability
>> on
>> > it.
>> >
>> > When i shutdown one of the iSCSI interface, nothing happened.
>> > But when i shutdown both iSCSI interface, VM in that hosts were paused
>> and
>> > did not migrate to another hosts. Is this behavior normal or i miss
>> > something?
>>
>> A paused VM can't be migrated, since there are no guarantees about the
>> storage state.  As the VMs were paused under erroneous (rather than
>> controlled such as putting the host into maintenance) situation,
>> migration policy can't help here.
>>
>> But highly available VMs can be restarted on another host automatically.
>> Do you have VM lease enabled for the highly available VM in High
>> Availability settings?  With a lease, Engine should be able to restart
>> the VM elsewhere after a while, without it Engine can't do that since
>> there is danger of resuming the VM on the original host, resulting in
>> multiple instances of the same VM running at the same time.
>>
>> VMs without high availability must be restarted manually (unless storage
>> domain becomes available again).
>>
>> HTH,
>> Milan
>>