Occasionally, Cilium's eBPF maps for L2 announcements can get "stale" after a configuration change (like switching from Local to Cluster). Perform a rolling restart of the Cilium pods to clear the state:

POD_NAME=$(kubectl get pod -l 'app.kubernetes.io/name'=cilium-agent -n cilium -o jsonpath='{.items[0].metadata.name}')
kubectl exec -n cilium  ${POD_NAME} -c cilium-agent -- cilium-dbg status

list load balancers ip pools: Required after changes

kubectl get CiliumLoadBalancerIPPool -A

Force helm:

helm repo add cilium https://helm.cilium.io/
helm upgrade cilium cilium/cilium \
--version 1.19.5 \
-f ./deploy/app/helm-values.yaml \
--namespace cilium