2021년 1월 24일 (일) 오전 1:22, Nir Soffer <nsoffer@redhat.com>님이 작성:
On Sat, Jan 23, 2021 at 5:18 PM Henry lol <pub.virtualization@gmail.com> wrote:
>
> Hi,
>
> according to HA VM documentation, a paused VM may be started on another host and later resumed on the original host.
> - https://www.ovirt.org/develop/ha-vms.html
> here, I'm assuming the HA VM was paused due to I/O error.
>
> but I'm wondering how it can happen because I guess the HA VM will be restarted on another host only after it's completely killed from the original host.

This is true for normal VMs, but not for HA VMs. These can be started
on another host even if we don't know if the VM is still running on the
original host. An example use case is host becoming disconnected from
the management network, or host having a hardware issue.

It means the normal VMs can be restarted by user's intervention so that it may result in the split brain??

> can you give the split brain scenario?

HA VM is using a storage lease, so it cannot have a split brain.

When VM is paused, it releases the lease. When VM is resumed,
it tries to acquire the lease before resuming, and resume will fail
if the lease is owned by another host.

If you start the HA VM on another host, the other host will acquire
the storage lease.  Resuming the original paused VM will fail.

Nir


Specifically, I assumed the PM setup and the HA VM not using the storage lease because I'm going to use MBS which doesn't support it yet AFAIK.
so, I was wondering in which case the split brain can occur under this circumstance.

I guess it doesn't occur as long as the user/admin doesn't manually restart VM, right?

Sincerely,