Setting mem_stats_period_seconds=0 should prevent the “Virtio memory balloon” driver from loading

[Albert Braden]

Setting mem_stats_period_seconds=0 in nova.conf should prevent the “Virtio memory balloon” driver from loading but it doesn't.

We are running Rocky installed with openstack-ansible. To reproduce the error:

1. In nova.conf set "mem_stats_period_seconds = 0" on controllers and hypervisors
2. Restart nova services on controllers and hypervisors
3. Build VM
4. Log into VM and type: lspci
5. lspci output will include "Red Hat, Inc. Virtio memory balloon"

For more information please see mailing list thread:

http://lists.openstack.org/pipermail/openstack-discuss/2020-February/012336.html

The problem this causes is that the Virtio memory balloon driver is not able to address large amounts of RAM. We encountered the problem when we built VMs with 1.4T RAM. The VM cannot boot because the driver fails:

"BUG: unable to handle kernel paging request at ffff988b19478000"

root@us01odc-dev2-ctrl1:~# dpkg -l | grep nova
ii nova-api 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute - API frontend
ii nova-common 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute - common files
ii nova-conductor 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute - conductor service
ii nova-novncproxy 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute - NoVNC proxy
ii nova-placement-api 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute - placement API frontend
ii nova-scheduler 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute - virtual machine scheduler
ii python-nova 2:18.2.3-0ubuntu1~cloud0 all OpenStack Compute Python 2 libraries
ii python-novaclient 2:11.0.0-0ubuntu1~cloud0 all client library for OpenStack Compute API - Python 2.7

root@us01odc-dev2-hv002:~# virsh --version
4.0.0

root@us01odc-dev2-hv002:~# qemu-system-x86_64 --version
QEMU emulator version 2.11.1(Debian 1:2.11+dfsg-1ubuntu7.21)

root@us01odc-dev2-hv002:~# nova --version
11.0.0

root@us01odc-dev2-hv002:~# openstack --version
openstack 3.16.1

[Balazs Gibizer (balazs-gibizer)]

I cannot reproduce the issue on master and the relevant code patch haven't change since rocky. Did you configure mem_stats_period_seconds = 0 config in the [libvirt] section of your nova.conf used by the nova-compute service? Did you restarted the nova-compute service after the reconfiguration? Also please note that such configuration change does not apply to the already running nova instances.

[Launchpad Janitor (janitor)]

[Expired for OpenStack Compute (nova) because there has been no activity for 60 days.]