>>> Zygo Blaxell <[email protected]> schrieb am 15.12.2020 um
>>> 19:18 in
Nachricht <[email protected]>:
> On Fri, Dec 11, 2020 at 03:25:47PM +0100, Ulrich Windl wrote:
>> Hi!
>>
>> While configuring a VM environment in a cluster I had setup an SLES15 SP2
> test VM using BtrFS. Due to some problem with libvirt (or the VirtualDomain
> RA) the VM was active on more than one cluster node at a time, corrupting the
> filesystem beyond repair it seems:
[...]
> There are at least 3 different ways libvirt or other hosting
> infrastructure software on the VM host could have avoided passing the
> same physical device to multiple VM guests. I would suggest implementing
> some or all of them.
As I found out the problem is with (live) migration and pacemaker: Migration
fails for some reason still to find out, and pacemaker starts the VM on the
destination node while it's still active on the source node. Amusingly it
complains to "recover" from a VM running on two nodes when in fact it creates
the problem by doing so ("restart" the VM on the destination node where none is
running). Just for explanation...
[...]
Regards,
Ulrich
