Hi,

Oracle Linux 9.7 was released recently and I was wondering if anybody else 
stumbled across some KVM Live Migration problems after the upgrade?
It seems some patches/enhancements to qemu-kvm where made in this release 
causing "guest CPU doesn't match specification: extra features: ht" errors, 
when trying to live migration VMs from Oracle Linux 9.6 hosts to a 9.7 host.

I'm not sure whether this is a bug or a long term change/enhancement and we 
should therefore restart all running VMs to deal with the change.

Log from the 9.7 host:

Nov 24 17:14:13 eb14 libvirtd[15172]: Domain id=22 name='i-37-594-VM' 
uuid=26c4b3df-fd55-43a0-a295-abf62ca9f822 is tainted: high-privileges
Nov 24 17:14:13 eb14 libvirtd[15172]: Domain id=22 name='i-37-594-VM' 
uuid=26c4b3df-fd55-43a0-a295-abf62ca9f822 is tainted: hook-script
Nov 24 17:14:13 eb14 systemd-machined[1464]: New machine qemu-22-i-37-594-VM.
Nov 24 17:14:13 eb14 systemd[1]: Started Virtual Machine qemu-22-i-37-594-VM.
Nov 24 17:14:13 eb14 libvirtd[15172]: operation failed: guest CPU doesn't match 
specification: extra features: ht
Nov 24 17:14:13 eb14 kernel: brbond0-1381: port 2(vnet25) entered disabled state
Nov 24 17:14:13 eb14 kernel: device vnet25 left promiscuous mode
Nov 24 17:14:13 eb14 kernel: brbond0-1381: port 2(vnet25) entered disabled state
Nov 24 17:14:13 eb14 NetworkManager[2598]: <info>  [1764000853.7324] device 
(vnet25): state change: activated -> unmanaged (reason 'unmanaged', 
managed-type: 'removed')
Nov 24 17:14:13 eb14 NetworkManager[2598]: <info>  [1764000853.7328] device 
(vnet25): released from controller device brbond0-1381
Nov 24 17:14:13 eb14 systemd[1]: 
machine-qemu\x2d22\x2di\x2d37\x2d594\x2dVM.scope: Deactivated successfully.
Nov 24 17:14:13 eb14 systemd-machined[1464]: Machine qemu-22-i-37-594-VM 
terminated.
Nov 24 17:14:14 eb14 libvirtd[15172]: no connection driver available for lxc:///
Nov 24 17:14:14 eb14 libvirtd[15172]: End of file while reading data: 
Input/output error


All hosts in the cluster are configured with the same cpu mode/model:

# grep cpu /etc/cloudstack/agent/agent.properties
guest.cpu.mode=custom
guest.cpu.model=Skylake-Server-noTSX-IBRS


As a quick fix I downgraded to the previous qemu-kvm packages (yum downgrade 
qemu-kvm-9.1.0-15.el9_6.9.x86_64) to regain the possibility of KVM live 
migration. But I'm not sure this is the right approach in the long-term.

Can anybody share some additional understanding about qemu? I know this is not 
exactly a cloudstack issue! :-)

Reply via email to