HNS loadbalancer backend pods malfunction

Hello,

I deployed a kuberenetes cluster with windows workers on openstack infrastructure.

What happen:

• We lost pods dns resolution on worker nodes.
• Pod can't reach de core dns service.
• We facing Windows Host Network Service (HNS) malfunction which cause loss of pods dns resolution on the windows server.
• It seems to be a malfunction between HNS loadbalancer and HNS pod backends.
• The issue disappears after restarting the windows Server.
• pod dns resolution work fine if i reach it with coredns pod ip address
• kubectl exec -it win-pod -- powershell.exe -Command resolve-dnsname kube-dns.kube-system.svc.cluster.local
  KO with timeout error : resolve-dnsname : kube-dns.kube-system.svc.cluster.local : This operation returned because the timeout period expired
• kubectl exec -it win-pod -- powershell.exe -Command resolve-dnsname kube-dns.kube-system.svc.cluster.local -Server 10.237.25.1
  OK (10.237.25.1 is coredns pod ip address)
• i followed this doc: https://argonsys.com/microsoft-cloud/library/troubleshooting-kubernetes-networking-on-windows-part-1/ and execute https://github.com/microsoft/SDN/blob/master/Kubernetes/windows/debug/collectlogs.ps1 (everything seems to be ok)

To reproduce the issue:

• Deploy/delete 5 pods on windows node each one minutes 20 times

Environment:
• Control plane (k8s Master Nodes):
o Linux CentOS-7.5
o Kuberenetes 1.18.3
o Docker 1.13.1
o CNI plugin Calico version: 3.17

• Windows server (k8s Worker Nodes):
o Windows Server 2019 Standard
o Version: 1809
o OS build: 17763.737
o Docker EE version: 19.3.13
o Networking : l2bridge with non-DSR mode

Start with the reported reproduction: deploy and delete five pods on a Windows node repeatedly, then compare resolve-dnsname through kube-dns with resolution through the CoreDNS pod IP. Review the referenced collectlogs.ps1 output and the HNS load balancer/backend state. Done means the HNS failure is reproducible and its cause or a verified fix is documented.