"default" network pool won't apply

[velis74]

I ran into an issue configuring the default network pool where qemu won't modify the subnet like I want.
The playbook runs just fine, but doesn't change anything.

- name: Basic provisioning of VM hosts
  hosts: vm_hosts
  roles:
    - role: stackhpc.libvirt-host
      libvirt_host_pools:
        - name: default
          type: dir
          path: /var/lib/libvirt/images
          mode: 755
          owner: libvirt-qemu
          group: kvm
      libvirt_host_networks:
        - name: default
          mode: nat
          bridge: virbr0
          domain: qemu.local
          ip: 192.168.123.254
          netmask: 255.255.255.0
          dhcp_start: 192.168.123.100
          dhcp_end: 192.168.123.250

Note the different subnet from default .122 and a different DHCP pool.

But if I change the name and bridge parameters, a new network will be created with my desired parameters.

I have found that stopping and deleting the default network from within Virtual Machine Manager GUI will leave qemu open for provisioning like I want (stopping alone is not enough). So this is currently work-around-able, clumsy-like.

It would seem that the task for provisioning the network only checks for its existence, but not its parameters.

I also tried changing virt_net command from define to modify, but that failed because it's not supported.

I think it should be possible (info, facts, list_nets?) to retrieve current configuration, determine that it doesn't match, then deactivate and delete the network. Only after that, create it anew from given parameters.

But, there's a catch:
I'm pretty sure this would be impossible with running guests (at least I HOPE qemu would not allow deleting a network in use), so this approach would only work while provisioning a host from scratch. I'm talking about scenario where the parameters would be found to differ, of course.

Still worth investigating / trying, IMO.

If you feel so too, let me know, I'll try to whip up a PR

[jovial]

Sounds like a good enhancement me. I guess we should just fail in the case where there are VMs running and a change is requested to the network.