Hello Stefan,

the reason I'm asking, is, that I ran into a case, in which the
priorities were all equal (not set), but all VMs were tried to migrate
from two nodes to the same node of a 8 node cluster.

This by itself wouldn't really be a problem, but qemu managed to
allocate all the memory for the VMs, so no migration failed to start.
Then the VMs used the memory and the OOM started to kill VMs and other
processes.

In this special case, we have a temporary workaround in place:
- No swap
- vm.overcommit_memory = 2
- vm.overcommit_ratio = 99

This of course is not a nice solution at all.


Best regards,
Max

On Thu, Apr 14, 2022 at 12:40:35PM +0200, Stefan Sterz wrote:
> Hello,
> 
> the target node is chosen by the node priority of the HA group that a
> VM is in. You can read more about that in the manual [1]. Here is an
> example: Assuming you have a three node cluster (node0, node1, node2)
> and the HA group has the following priorities: node0:2, node1:1,
> node2:0. A VM running on node0 will be moved to node1 and then node2
> if each node fails one after another. If several nodes have the same
> priority, VMs will be distributed evenly.
> 
> Best regards,
> Stefan
> 
> [1]:
> https://pve.proxmox.com/pve-docs/pve-admin-guide.html#ha_manager_groups
> 
>