I restarted libvirt and nova-compute, but nothing changes...

Reinstall glance and python-glance client in your controller. Then at /etc/glance/glance-api.cfg add this line enable_v2_api = True. Reinstall nova-api, nova-conductor and python-novaclient on controller.

On your compute node, uninstall nova-compute and then reinstall it as follow:

apt-get update -y && apt-get  dist-upgrade -y

apt-get install -y nova-compute sysfsutils

finally restart all above mentioned services and by running nova service-list, make sure all of them are up and running. Then you can lunch your vm and delete them without any problem.

some times, reinstalling some services may affect other services. So do not get panic if you get some weird error messages. reinstalling affected services solve your problem.

Restarting doesn't work (they are named openvswitch-switch and nova-compute on my machine). The nova/neutron versions match between the controller and the compute (2.23.0/2.4.0)

try to restart all your nova services if this will solve or try to reboot the controller and the compute.

    for serverror in api conductor scheduler cert consoleauth compute novncproxy network; do
systemctl restart nova-$serverror
done

or check the /var/lib/nova/instances of compute