Windows Node fails to run Pod using the provided OVS binary

[wenyingd]

Describe the bug

When running Pods on Windows Node, I saw the Pods is always stucking in "RunContainerError" status. After describing Pod details, these events are got,

Events:
  Type     Reason  Age                  From     Message
  ----     ------  ----                 ----     -------
  Warning  Failed  22m (x12 over 100m)  kubelet  Error: context deadline exceeded
  Warning  Failed  8m (x23 over 106m)   kubelet  Error: context deadline exceeded
  Warning  Failed  2m (x16 over 108m)   kubelet  Error: context deadline exceeded

After checking antrea-agent logs, it looks antrea-agent has successfully allocated IP address to the Pod. HNSEndpoint, OVS port and OpenFlow entries are installed on the Node. But containerd runtime has no response to the call from kubelet.

After deleting the Pod with command "kubectl delete Pod xxx", Pod is stucking in "Terminating" status. Re-check HNS resources, it was found that the HNSEndpoint was not deleted. After manually deleting the HNS Endpoint, the powershell is stuck with no response. The setup is using the default OVS binary provided via this link (https://github.com/antrea-io/antrea/blob/main/hack/windows/Install-OVS.ps1#L35)

This is an OVS known issue which may block HNS services, and the fix is merged in the open source OVS (openvswitch/ovs#385). Hence, antrea provided OVS is supposed to include the fix.

To Reproduce

Expected

Actual behavior

Versions:

Additional context