Devstack stack run issue
[keeping originator in Cc since they don't seem to be subscribed] On 2021-10-26 16:54:49 +0000 (+0000), Venkata Krishna Reddy wrote:
While running stack on latest devstack, ended up with the following error message: [...] Oct 26 16:36:49 ubuntu nova-compute[337228]: ERROR oslo_service.service nova.exception.InvalidCPUInfo: Configured CPU model: Nehalem is not compatible with host CPU. Please correct your config and try again. Unacceptable CPU info: CPU doesn't have compatibility.
A week ago, the stack was successful and it is failing now. [...]
This is due to https://review.opendev.org/815020 which merged a week ago to update the default CPU model in DevStack with one which will work for CentOS 9. See also this mailing list post and its replies which go into greater detail on the situation: http://lists.openstack.org/pipermail/openstack-discuss/2021-October/025500.h... You can probably override LIBVIRT_CPU_MODEL in your configuration setting the old default of "none" if you want the old behavior for now, though there are likely better long-term solutions. -- Jeremy Stanley
On Tue, Oct 26, 2021, at 10:05 AM, Jeremy Stanley wrote:
[keeping originator in Cc since they don't seem to be subscribed]
On 2021-10-26 16:54:49 +0000 (+0000), Venkata Krishna Reddy wrote:
While running stack on latest devstack, ended up with the following error message: [...] Oct 26 16:36:49 ubuntu nova-compute[337228]: ERROR oslo_service.service nova.exception.InvalidCPUInfo: Configured CPU model: Nehalem is not compatible with host CPU. Please correct your config and try again. Unacceptable CPU info: CPU doesn't have compatibility.
A week ago, the stack was successful and it is failing now. [...]
This is due to https://review.opendev.org/815020 which merged a week ago to update the default CPU model in DevStack with one which will work for CentOS 9. See also this mailing list post and its replies which go into greater detail on the situation:
http://lists.openstack.org/pipermail/openstack-discuss/2021-October/025500.h...
You can probably override LIBVIRT_CPU_MODEL in your configuration setting the old default of "none" if you want the old behavior for now, though there are likely better long-term solutions.
You would need to set LIBVIRT_CPU_MODE to "none" then LIBVIRT_CPU_MODEL should be ignored (note MODE vs MODEL). If that doesn't work it is a bug and we should fix it. The other thing that might be worth double checking on is what CPU you are running devstack on. We expected that this would work for any reasonably new x86 based system. Aarch64 has an explicit override for the MODEL later in the code. I suppose it is possible we broke this for powerpc and need to add an explicit override similar to aarch64? If you are on x86 more details about the specific CPU might be useful to determine if there are problems with the expected Nehalem compatibility. Clark
participants (3)
-
Clark Boylan
-
Jeremy Stanley
-
Venkata Krishna Reddy