[BUG] P2P links wrongly marked as 'multi-provider' LAN links

[jbemmel]

Lab topology

---
defaults.device: frr
groups:
 _auto_create: True
 core:
  members: [ c1, c2 ]
  provider: clab
 edge:
  members: [ e1, e2 ]
  provider: libvirt

links:
- c1-c2
- e1-c1
- e2-c2

Output

For host_vars/c1/topology.yml

af:
  ipv4: true
box: quay.io/frrouting/frr:10.0.1
clab:
  binds:
  - clab_files/c1/daemons:/etc/frr/daemons
  - clab_files/c1/hosts:/etc/hosts
  config_templates:
  - daemons:/etc/frr/daemons
  - hosts:/etc/hosts
  kind: linux
hostname: clab-Automation-c1
interfaces:
- bridge: Automation_1   <-- bridge added
  ifindex: 1
  ifname: eth1
  ipv4: 172.16.0.1/24
  linkindex: 1
  mtu: 1500
  name: c1 -> c2
  neighbors:
  - ifname: eth1
    ipv4: 172.16.0.2/24
    node: c2
  type: lan                      <-- p2p changed to 'lan'

Version: netlab dev branch

[jbemmel]

Potential fix in providers/__init__.py:

  """
  Generic provider pre-transform processing: Mark multi-provider links
  """
  def pre_transform(self,topology : Box) -> None:
    if not 'links' in topology:
      return

    for l in topology.links:
      providers = {}
      for intf in l.interfaces:
        node = topology.nodes[intf.node]
        provider = node.get('provider',topology.provider)
        providers[provider] = True                        # Take note of providers being used

      if len(providers)<=1:
        continue                                          # Not a multi-provider link - continue
      
      p_name = topology.provider                          # Get primary and secondary provider
      for s_name in providers.keys():                     # ... to make the rest of the code more readable
        if s_name!=p_name:
          l[p_name].provider[s_name] = True               # Collect secondary link provider(s)
          if 'uplink' in l[p_name]:                       # ... and copy primary uplink to secondary uplink
            l[s_name].uplink = l[p_name].uplink

[ipspace]

This has been discussed previously. The link type does not matter for containerlab, and the default prefix pool is selected based on the number of attached nodes not the link type since #1673.

While this issue is technically correct, it's also irrelevant at the moment.

[jbemmel]

This has been discussed previously. The link type does not matter for containerlab, and the default prefix pool is selected based on the number of attached nodes not the link type since #1673.

While this issue is technically correct, it's also irrelevant at the moment.

I disagree. The bridge needlessly breaks LACP between MLAG peers when using containers for devices like Dell OS10 or Arista EOS

[ipspace]

I disagree. The bridge needlessly breaks LACP between MLAG peers when using containers for devices like Dell OS10 or Arista EOS

You can disagree all you want, but the fact remains that a linux bridge is not used by containerlab configuration template when there are only two containers connected to a link.

[ipspace]

Now that the LAG link type is no longer used, the last vestiges of disagreements should be gone 🤪

[jbemmel]

We’ll find something else, don’t worry :)

We disagree, therefore we are