Add DHCP IP Retries in PrepareHNSNetwork

To address the potential race condition issue where acquiring a DHCP IP address may fail after CreateHNSNetwork,
we added a retry mechanism to wait for an available IP. If the DHCP IP cannot be acquired within three seconds,
a warning message will be returned.

Signed-off-by: Shuyang Xin <[email protected]>

ci/jenkins/test.sh

@@ -295,6 +295,13 @@ function wait_for_antrea_windows_processes_ready {
         done
         sleep 10
     done
+    WIN_BRINT_DHCP=$(ssh -o StrictHostKeyChecking=no -n administrator@${WINIP} "powershell.exe (Get-NetIPInterface -InterfaceAlias br-int -AddressFamily IPv4).Dhcp")
+    if [[ $WIN_DHCP == $WIN_BRINT_DHCP ]]; then
+        echo "New created uplink DHCP status is consistent with the original adapter."
+    else
+        echo "New created uplink DHCP status is different with the original adapter."
+        exit 1
+    fi
 }
 
 function clean_ns {
@@ -652,6 +659,9 @@ function deliver_antrea_windows_containerd {
             ssh -o StrictHostKeyChecking=no -n Administrator@${IP} "gzip -d antrea-windows.tar.gz && ctr -n k8s.io images import antrea-windows.tar"
         fi
     done
+    # Add Windows interface DHCP check using CI script to obtain the original interface status.
+    WINIP=$(kubectl get nodes -o wide --no-headers=true | awk '$1 ~ /win-0/ {print $6}')
+    WIN_DHCP=$(ssh -o StrictHostKeyChecking=no -n administrator@${WINIP} "powershell.exe (Get-NetIPInterface -InterfaceAlias "vEthernet0 2" -AddressFamily IPv4).Dhcp")
     rm -f antrea-windows.tar
     echo "==== Finish building and delivering Windows containerd images ===="
 }

pkg/agent/util/net_windows.go

@@ -460,9 +460,28 @@ func PrepareHNSNetwork(subnetCIDR *net.IPNet, nodeIPNet *net.IPNet, uplinkAdapte
 			hnsNetworkDelete(hnsNet)
 		}
 	}()
-
-	adapter, ipFound, err := adapterIPExists(nodeIPNet.IP, uplinkAdapter.HardwareAddr, ContainerVNICPrefix)
+	var adapter *net.Interface
+	var ipFound bool
+	err = wait.PollImmediate(time.Second, 3*time.Second, func() (done bool, err error) {
+		apt, ipExist, err := adapterIPExists(nodeIPNet.IP, uplinkAdapter.HardwareAddr, ContainerVNICPrefix)
+		if err != nil {
+			return false, err
+		}
+		adapter = apt
+		if !ipExist {
+			return false, nil
+		}
+		ipFound = true
+		return true, nil
+	})
 	if err != nil {
+		if err == wait.ErrWaitTimeout {
+			dhcpStatus, err := InterfaceIpv4Dhcp(uplinkAdapter.Name)
+			if err != nil {
+				klog.ErrorS(err, "Failed to get Ipv4 DHCP status on the network adapter", "adapter", uplinkAdapter.Name)
+			}
+			klog.Warningf("Timeout acquiring IP for the adapter, DHCP status: %t", dhcpStatus)
+		}
 		return err
 	}
 	vNicName, index := adapter.Name, adapter.Index
@@ -647,6 +666,15 @@ func HostInterfaceExists(ifaceName string) bool {
 	return true
 }
 
+// InterfaceIpv4Dhcp returns the Ipv4 DHCP status on the specified interface.
+func InterfaceIpv4Dhcp(ifaceName string) (bool, error)  {
+	adapter, err := getAdapterInAllCompartmentsByName(ifaceName)
+	if err != nil {
+		return false, err
+	}
+	return adapter.ipv4Dhcp, nil
+}
+
 // SetInterfaceMTU configures interface MTU on host for Pods. MTU change cannot be realized with HNSEndpoint because
 // there's no MTU field in HNSEndpoint:
 // https://github.com/Microsoft/hcsshim/blob/4a468a6f7ae547974bc32911395c51fb1862b7df/internal/hns/hnsendpoint.go#L12
@@ -1085,6 +1113,7 @@ type updateIPInterfaceFunc func(entry *antreasyscall.MibIPInterfaceRow) *antreas
 type adapter struct {
 	net.Interface
 	compartmentID uint32
+	ipv4Dhcp bool
 }
 
 func (a *adapter) setMTU(mtu int, family uint16) error {
@@ -1238,6 +1267,10 @@ func getAdaptersByName(name string) ([]adapter, error) {
 		case windows.IF_TYPE_ATM:
 			ifi.Flags |= net.FlagBroadcast | net.FlagPointToPoint | net.FlagMulticast // assume all services available; LANE, point-to-point and point-to-multipoint
 		}
+		ipv4Dhcp := false
+		if aa.Flags&0x00000004 != 0{
+			ipv4Dhcp = true
+		}
 		if aa.Mtu == 0xffffffff {
 			ifi.MTU = -1
 		} else {
@@ -1250,6 +1283,7 @@ func getAdaptersByName(name string) ([]adapter, error) {
 		adapter := adapter{
 			Interface:     ifi,
 			compartmentID: aa.CompartmentId,
+			ipv4Dhcp: ipv4Dhcp,
 		}
 		adapters = append(adapters, adapter)
 	}