Understanding hypervisor compatibility during live migration

asked 2020-08-06 09:59:40 -0600

Andrew Bogott gravatar image

I've recently added ceph shared storage to my cloud, and I'm experimenting with live migration. I have two groups of hypervisors, one running Debian Buster and one running Debian Stretch. In order to ensure migration compatibility I've set this on all compute nodes:

cpu_mode=custom cpu_model=IvyBridge-IBRS

Nevertheless, nova will migrate hosts from my Stretch to Buster hosts but if I try to migrate something back from Buster to Stretch it tells me 'The instance requires a newer hypervisor version than has been provided.'

How can I determine what it means by 'newer'? Surely it doesn't actually care about the host OS; Both are running nova 'rocky.' The libvirt versions are different due to the distro differences.

Stretch:

  • nova-compute 2:18.1.0-6+deb10u1~bpo9+1
  • nova-compute-kvm 2:18.1.0-6+deb10u1~bpo9+1
  • libvirt-clients 4.3.0-1~bpo9+1.1

Buster:

  • nova-compute 2:18.1.0-6+deb10u1~bpo10+1
  • nova-compute-kvm 2:18.1.0-6+deb10u1~bpo10+1
  • libvirt-clients 5.0.0-4+deb10u1

Can anyone advise me about how to support bi-directional migration? Or, failing that, help me understand what nova is upset about?

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
0

answered 2020-08-10 10:46:10 -0600

Nova is relying on libvirt to detect the correct cpu-type. So you should check which cpu-type libvirt detected. Sometimes (esp on newer server cpus) libvirt is unable to detect all features and/or is failing back to the desktop-cpu (which may not have all features)

There is some doc around on how to compile/generate the "least common cpu-type" with libvirt-cli.

edit flag offensive delete link more

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

Stats

Asked: 2020-08-06 09:59:40 -0600

Seen: 24 times

Last updated: Aug 06