Hi,

this is probably a bug. Can you open a new ticket in Bugzilla?
https://bugzilla.redhat.com/enter_bug.cgi?product=ovirt-engine

As a workaround, if you are sure that the VM's NUMA configuration is compatible with the host's NUMA configuration, you could create a custom cluster scheduling policy and disable the "NUMA" filter. In Administration -> Configure -> Scheduling Policies.

Regards,
Andrej

On Thu, 13 Jun 2019 at 12:49, Matthias Leopold <matthias.leopold@meduniwien.ac.at> wrote:
> Hi,
>
> I'm having trouble using vNUMA and hugepages at the same time:
>
> - hypervisor host hast 2 CPU and 768G RAM
> - hypervisor host is configured to allocate 512 1G hugepages
> - VM configuration
> * 2 virtual sockets, vCPUs are evenly pinned to 2 physical CPUs
> * 512G RAM
> * 2 vNUMA nodes that are pinned to the 2 host NUMA nodes
> * custom property "hugepages=1048576"
> - VM is the only VM on hypervisor host
>
> when I want to start the VM I'm getting the error message
> "The host foo did not satisfy internal filter NUMA because cannot
> accommodate memory of VM's pinned virtual NUMA nodes within host's
> physical NUMA nodes"
> VM start only works when VM memory is shrunk so that it fits in (host
> memory - allocated huge pages)
>
> I don't understand why this happens. Can someone explain to me how this
> is supposed to work?
>
> oVirt engine is 4.3.3
> oVirt host is 4.3.4
>
> thanks
> matthias