[VM Agent] Bugfix: antrea-agent failed to delete ExternalNode

The issue is seen on a RHEL 8.4 VM on azure cloud, which is configured with
dhclient to manage the network interface. The root cause is antrea-agent fails
to recover the IP/Routes from host internal interface to uplink after
ExternalNode is deleted, because the added IP/Routes is deleted or conflicted
with dhclient configuration. Then in the continuous retry with ExternalNode
deletion, antrea-agent is blocking at the precheck on the existentance of host
internal interface always returns true as the uplink's name is already
recovered.

Signed-off-by: wenyingd <[email protected]>

pkg/agent/externalnode/external_node_controller.go

@@ -588,6 +588,16 @@ func (c *ExternalNodeController) removeOVSPortsAndFlows(interfaceConfig *interfa
 		}
 	}()
 
+	// This is for issue #5111 (https://github.com/antrea-io/antrea/issues/5111), which may happen if an error occurs
+	// when moving the configuration back from host internal interface to uplink. This logic is run in the second
+	// try after the error is returned, at this time the host internal interface is already deleted, and the uplink's
+	// name is recovered. So the ips and routes in "adapterConfig" are actually read from the uplink and no need to
+	// move the configurations back. The issue was seen on VM with RHEL 8.4 on azure cloud.
+	if !hostInterfaceExists(uplinkIfName) {
+		klog.InfoS("Uplink is not existing on the host, return")
+		return nil
+	}
+
 	// Wait until the host interface created by OVS is removed.
 	if err = wait.PollImmediate(50*time.Millisecond, 2*time.Second, func() (bool, error) {
 		return !hostInterfaceExists(hostIFName), nil