In an OCP node running on OCP 4.12.46 + RHCOS (RHEL 8.6) with metallb operator installed, after a reboot event, the following message has been reported for an load balancer service announced by the metallb operator with the layer 2 method: 


1608:2024-07-11T14:39:12.016231892+00:00 stdout F {"IPAdvertisement":{},"caller":"layer2_controller.go:142","ips":["102.0.0.4/25"],"level":"warn","localIfs":[],"msg":"the specified interfaces used to announce LB IP don't exist","op":"SetBalancer","pool":"<pool-name>","protocol":"layer2","service":"<namespace-name>/<service-name>","ts":"2024-07-11T14:39:12Z"}

The service is configured with the following paramters in the metallb perspective: 

# IPAddressPool #
apiVersion: metallb.io/v1beta1
kind: IPAddressPool
  labels:
    <label-name>: l2
spec:
  addresses:
  - x-ipv4-0000004687-x-x-ipv4-0000005944-x
  autoAssign: false
  avoidBuggyIPs: false

# L2Advertisement #

spec:
  ipAddressPoolSelectors:
  - matchExpressions:
    - key: <label-name>
      operator: In
      values:
      - l2
 
In the previous shared log, the localIFs parameter appears to be empty from the log message '"localIfs":[]'. Considering the current configuration, the node should reply arps from any interfaces and this did not happen

Additionally, the following event has been noticed:

  message: the interfaces specified by LB IP "x-ipv4-0000000041-x" doesn't exist in
    assigned node "xxxxxxxxxxxxxxxxxxxxx-infra-e021000004014a-az3-2tjrt" with protocol
    "layer2"

Only one service appeared to have the issues with the assignment

    Not easily

    1.
    2.
    3.
    

    The affected load balancer service is unavailable at the moment that 'the interface used does not exist' issue is reported

    The expected is that the announcement should be moved between the nodes without the outage issues.