I'm using vCPU tuning now.
It looks like this:
<cputune>
<shares>10000</shares>
</cputune>

That's the max amount of shares #libVirt / #VirtManager lets me allocate.

More shares = more priority (bigger fish).

The #PiHole #VM gets the max, because it runs DNS & DHCP - when it needs CPU cycles, it bloody well gets 'em!

#Nextcloud gets 5000, #HomeAssistant 4000, and #Jellyfin 1000 - they all can have the whole CPU, as long as there's no bigger fish around.

#Linux #Virtualisation #NAS

I think I've worked out what was wrong with my VMs on Fedora 44.

Looks like the firewall rules that Docker sets up do not play well with the ones libvirtd adds.

Forcing Docker to use nftables rather than iptables/firewalld appears to fix things

https://github.com/moby/moby/issues/52612

https://bugzilla.redhat.com/show_bug.cgi?id=2466836

#Fedora #Fedora44 #Docker #libvirt

Anybody else seeing network problems with Boxes/Libvirt VMs on Fedora 44?

I imported an existing disk image and tried to create a new from a Ubuntu ISO and neither get NAT network access.

If so please add any details from here:

https://bugzilla.redhat.com/show_bug.cgi?id=2466836

#Fedora #libvirt

I've imported a VM disk image into Fedora 44 boxes (I forgot to export the XML) and while it has got a IP address on the default network it doesn't appear to have access to the outside world (no NAT).

Anybody know how to enable it?

#Fedora44 #Fedora #libvirt

Ansible roles: proxy_env, ssh, etc_hosts, libvirt released

https://blog.wagemakers.be/blog/2026/05/03/ansible-proxy_env-ssh-etc_hosts-libvirt_released/

Made some time to do some work for a few #ansible roles that I maintain. You’ll find the new releases in this blog post.

* stafwag.proxy_env 2.1.0
* stafwag.ssh 1.1.1
* stafwag.libvirt 2.1.0
* stafwag.etc_hosts 1.1.1

#ansible, #libvirt #ssh #linux #freebsd

#stafwag @stafwag

PSA for anyone using #QEMU #KVM for #SingleGPUPassthrough
Guides are all over the net suggesting hooks scripts invoking pci devices, unloading/loading kernel modules, and other unnecessary things. Here's my current hook script for starting the VM:

That's it. And the reverse for teardown:

#IOMMU groups still apply, and you need to pass the correct PCI devices to your VM, but everything else is handled automatically.

Disclaimer that this is how it works currently for my AMD card. I did have a working setup with my NVIDIA card that did unload/load kernel modules, however it seems things have come a long way since I set that up.

#virtualization #libvirt #vfio

Asking the same question here as well.

When using libvirtd/qemu+kvm what is your preferred way for sharing a folder with a huge dataset >50TB into multiple guests especially in the case where you want the VM to be network isolated.

Are you using:
* the filesystem-"device": virtiofs, virtio-9p, mtp
* the disk-device: with dir-source
* link-local network with: NFS, SMB, ...
* a different architecture entirely?

See full question: https://www.reddit.com/r/HomeServer/comments/1skp5u5/storage_pass_through_without_to_libvirtqemu_vm/

#qemu #kvm #libvirt #NixOS #Homelab

Was messing around with the ansible libvirt collection to deploy some molecule instances. Turns out the virt_install module forgot to add the #cloud-config header to user-data when converting it from dictionary to string. Therefor my cloudinit user never got created. Currently preparing a pull request but I am stuck a bit on the test setup.

#ansible #molecule #python #libvirt #automation #cloudinit