Why freezes VM (windows server 2016) on qemu/kvm (libvirtd)?

Question

linux

Host:
i9-11900K
nvme disk. lvm2.
OS Arch Linux.
There are three virtual machines on it: two for debian, the third one is Windows Server 2016 with RDP.
Once a day, the Windows virtual machine freezes for 1-4 minutes. At this moment, it is impossible to connect to it by any means in virt-manager is also impossible. She doesn't even ping.
Htop on the host shows at this time the loading of all 8 cores allocated to this virtual machine. Then everything returns to normal, no processes launched by the user before the freeze require them to be restarted.
At the same time, due to the fact that, just as in a black hole, you can’t look beyond the event horizon, I also don’t know during the freeze how to see which process is eating percent at this moment.
I began to keep the system monitor always running on the server. And now, after another freeze and the resumption of server availability. I went in, and there is such a picture:



That is, during this frieze, the VM does not even write anything to the logs - the graph is empty. In the literal sense, as if she freezes completely.
The question arises that perhaps the problem is in the settings of the virtual machine. But everything is classic there: virtio disk. Proc host. Another such moment is that if no one works on a virtual machine, for example, at night, or on weekends, there are no friezes. That is, it turns out that all the same, the frieze is initially initiated by the user load.
In short, tell someone, please, how to debug host resources? Or what settings of the VM or libvirtd itself should I pay attention to?

Answer 1

how to debug with host media

Configure VNC for this virtual machine at the KVM level. This does not use virtual machine resources, only host resources. And you need to log in through this VNC in advance and open the task manager, etc., so that during the freeze you only look at the VNC window.