vs_port not working in CentOS 9 preview

if you still have this environment up, can you try the following DIFF

``` diff

index 5bb8526..eea7b71 100644
--- a/usr/share/openstack-puppet/modules/vswitch/lib/puppet/provider/vs_port/ovs_redhat.rb
+++ b/usr/share/openstack-puppet/modules/vswitch/lib/puppet/provider/vs_port/ovs_redhat.rb.old
@@ -15,7 +15,7 @@ Puppet::Type.type(:vs_port).provide(
     'ONBOOT' => 'yes',
     'BOOTPROTO' => 'dhcp',
     'PEERDNS' => 'no',
- 'NM_CONTROLLED' => 'yes',
+ 'NM_CONTROLLED' => 'no',
     'NOZEROCONF' => 'yes'
   }

@@ -23,8 +23,8 @@ Puppet::Type.type(:vs_port).provide(
   defaultfor :osfamily => :redhat

   commands :ip => 'ip'
- commands :ifdown => 'nmcli connection down'
- commands :ifup => 'nmcli connection up'
+ commands :ifdown => 'ifdown'
+ commands :ifup => 'ifup'
   commands :vsctl => 'ovs-vsctl'

   def initialize(value={})

```

I have a hypothesis that NM will still read IFCFG files in C9, even if network-scripts is not available and I'd like to see if the following change would "fix" the issue you're seeing?

sorry the diff was in reverse. but you can see the modification i'm making.

in my local test environment on C8, this seems to work.

``` shell
$ sudo puppet apply --modulepath=/usr/share/openstack-puppet/modules test.pp
Warning: /etc/puppet/hiera.yaml: Use of 'hiera.yaml' version 3 is deprecated. It should be converted to version 5
   (file: /etc/puppet/hiera.yaml)
Warning: Undefined variable '::deploy_config_name';
   (file & line not available)
Notice: Compiled catalog for undercloud.localdomain in environment production in 0.21 seconds
Notice: /Stage[main]/Main/Vs_bridge[br-test]/ensure: created
Notice: /Stage[main]/Main/Vs_port[loop2]/ensure: created
Notice: Applied catalog in 0.93 seconds
```

However, I don't have immediate access to a C9 lab at this time to confirm.

Yes this should work, in RHEL9 / CentOS Stream 9 default /etc/NetworkManager/NetworkManager.conf has
[main]
#plugins=keyfile,ifcfg-rh

which is coming from F33 feature https://fedoraproject.org/wiki/Changes/NetworkManager_keyfile_instead_of_ifcfg_rh

so by default EL9 NM will not create ifcfg* files but will be able to read them

So you could also reproduce this on Fedora >=33 if you like!

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/puppet-vswitch/+/793262

This is not working as expected in my c9 env. puppet apply succeeded but it's using ovs instead of ovs_redhat provider and it's not persisting the change. Note that command implies confine so it's probably looking for a 'nmcli connection down' command?

I can provide you with a c9 environment for your test if you want to experiment a bit.

NM can actually check ifcfg files as alan mentioned, but i'm not sure if understand the full syntax, tbh.

Alfredo, if you have a reproducer env and dont mind share (or providing a new one) I would greatly appreciate it. These are my keys https://github.com/cloudnull.keys

I spent some more time looking into this last week / end and sadly I've discovered that the rh-ifcfg plugin does not work for our use-case; it does not support any OVS device type. https://developer.gnome.org/NetworkManager/stable/nm-settings-ifcfg-rh.html

So we're going to need to develop nmcli integration into our module, or switch to something else.

So... I think there are many problems now...

1.
The vswitch calss used vswitch::params::provider to determine which class should be included but we don't have vswitch::ovs_redhat so this class does NEVER work in CentOS/RHEL.
(This is an independent problem, though)

2.
Currently we have multiple provider implementations for vs_port but these implementations are NEVER automatically selected but the first one(ovs.rb) is always used. That's why Alfredo experienced that ovs provider was used.
We need to add a new option to that ovs_port type to switch provider.
(I'm still looking for the way to implement this properly)

So what we need here is to fix 2 and add a new provider implementation relying on network manager...

Ignore what I mentioned in the second point. I totally missed defined defaultfor/confine.

If that switch is working then I think what we need is a new provider code relying on nmcli.

So I submitted very rough prototype of the new provider based on nmcli
 https://review.opendev.org/c/openstack/puppet-vswitch/+/795476

Because I've not yet tested this in actual setup, it might be broken and also some features are still missing now. However I think this can be used as our starting point to support CentOS/RHEL9.

Regarding 1 I mentioned in my comment 9, I submitted the fix.
 https://review.opendev.org/c/openstack/puppet-vswitch/+/795477

Change abandoned by "Kevin Carter <email address hidden>" on branch: master
Review: https://review.opendev.org/c/openstack/puppet-vswitch/+/793262