libvirtd: failed to connect to socket after installation

[Carlos Bilbao]

Hello, made some progress and I think I see what the issue may be now...
and it looks like virt-install's fault.

 > > Regardless of all that, the client is trying to connect to virtqemud
 > > but you seem to have the monolithic libvirtd daemon running. I can't
 > > remember whether that's expected, but it's looks like a red flag to
 > > me. I think the client might have gotten confused about whether split
 > > daemons are in use.
 > >
 > > Please check out 
https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flibvirt.org%2Fdaemons.html&data=05%7C01%7Ccarlos.bilbao%40amd.com%7C50dc7866a7034346731108da90ecc0ac%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637981643769774692%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=ICnGUQ5%2BktKFZNx9kp2KvTCajLXKgzsL2uh3Fh%2FKBsY%3D&reserved=0 
and ensure you're
 > > in one of the two supported deployment scenarios and not a weird mix
 > > of the two.
 >

So, I started again from scratch. Deleted absolutely everything that had
to do with libvirt. I also had to unmask a bunch of service-related stuff
that were on masked state:

for ANNOYING_SERVICE in libvirtd virtlogd.socket virtlogd.service \
         virtlockd.socket virtlockd.service virtlogd-admin.socket 
virtlockd-admin.socket
do
         systemctl unmask $ANNOYING_SERVICE
done

So, after doing this, building and installing the tree, reloading the
daemon, and starting all services and sockets again, I finally get an
active libvirtd daemon:

# systemctl status libvirtd
● libvirtd.service - Virtualization daemon
      Loaded: loaded (/usr/local/lib/systemd/system/libvirtd.service; 
enabled; vendor preset: enabled)
      Active: active (running) since Thu 2022-09-08 15:51:39 UTC; 1min 
30s ago
        Docs: man:libvirtd(8)
              https://libvirt.org
    Main PID: 2429457 (libvirtd)
       Tasks: 20 (limit: 32768)
      Memory: 55.6M
         CPU: 28ms
      CGroup: /system.slice/libvirtd.service
              ├─   2760 /usr/sbin/dnsmasq 
--conf-file=/var/lib/libvirt/dnsmasq/default.conf --leasefile-ro 
--dhcp-script=/usr/lib/libvirt/libvirt_leaseshelper
              ├─   2761 /usr/sbin/dnsmasq 
--conf-file=/var/lib/libvirt/dnsmasq/default.conf --leasefile-ro 
--dhcp-script=/usr/lib/libvirt/libvirt_leaseshelper
              └─2429457 /usr/local/sbin/libvirtd --timeout 120

Sep 08 15:51:39 host systemd[1]: Starting Virtualization daemon...
Sep 08 15:51:39 host systemd[1]: Started Virtualization daemon.
#

Now, this is when things get interesting... The next step is testing
virt-install. Since I removed it, I proceed to install again:

# sudo apt install -y virtinst
(...)
Unpacking virtinst (1:4.0.0-1) ...
Setting up libvirt-glib-1.0-data (4.0.0-2) ...
Setting up libvirt0:amd64 (8.0.0-1ubuntu7.1) ...
Setting up libvirt-glib-1.0-0:amd64 (4.0.0-2) ...
Setting up python3-libvirt (8.0.0-1build1) ...
Setting up libvirt-clients (8.0.0-1ubuntu7.1) ...
Setting up libvirt-daemon-driver-qemu (8.0.0-1ubuntu7.1) ...
Setting up virt-viewer (7.0-2build2) ...
Setting up libvirt-daemon (8.0.0-1ubuntu7.1) ...
Setting up virtinst (1:4.0.0-1) ...

As you can see, virtinst also installs the libvirt daemon. I didn't think
this would be an issue because the libvirtd daemon running is still my
version. However, when I try to do virt-install I get:

# virt-install --connect qemu:///system --name ubuntu-sev \
--boot 
loader=/usr/share/OVMF/OVMF_CODE.fd,loader.readonly=yes,loader.type=pflash,nvram.template=/usr/share/OVMF/OVMF_VARS.fd,loader_secure=no 
\
--vcpus 8 --memory 4096 --memtune hard_limit=16777216 \
--disk pool=default,device=disk,size=32,format=raw \
--controller type=scsi,model=virtio-scsi  --network 
bridge=virbr0,model=virtio \
--controller type=virtio-serial --machine q35 --cpu host-passthrough \
--cdrom /var/lib/libvirt/images/ubuntu.iso --osinfo detect=on,require=on \
--launchSecurity sev,policy=0x00 --graphics none --tpm none
ERROR    Failed to connect socket to '/var/run/libvirt/virtqemud-sock': 
No such file or directory
#

So, back to square one. However, notice that virtqemud is one of the modular
daemons. So, I believe that what happens is that virt-install assumes
libvirt is using the new daemon mode. Is this something that can be
configured? I don't see that option in the man pages of virt-install.

 > >
 > > If you want to build a Debian package, I recommend using the one
 > > that's already in the archive as a starting point instead of using
 > > dh_make to create one from scratch. It's currently a couple of
 > > versions behind upstream, but should be pretty solid otherwise.
 > >
 >
 > Not sure what location you are referring to. Is this on a separate 
repo in
 > the GitLab? Also, I need to be able to test changes in the source code of
 > libvirt (e.g. src/qemu/qemu_capabilities.c). How would I go around that
 > with the prepared Debian package? Apologies if this is obvious, I have no
 > experience with dpkg-buildpackage.
 >
 > > dh_make is intended to create a very basic skeleton for a package,
 > > and you're expected to make significant changes to the files it
 > > produces before being able to build a working package.
 > >

Thanks,

Carlos