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 <gavinx@vmware.com>

pkg/agent/agent_windows.go

@@ -107,6 +107,11 @@ func (i *Initializer) prepareHNSNetworkAndOVSExtension() error {
 		return err
 	}
 	i.nodeConfig.UplinkNetConfig.DNSServers = dnsServers
+	Dhcp, err := util.GetDHCPByInterfaceIndex(adapter.Index)
+	if err != nil {
+		return err
+	}
+	i.nodeConfig.UplinkNetConfig.DHCP = Dhcp
 	// Save routes which are configured on the uplink interface, and configure them on the management virtual adapter
 	// if Windows host doesn't move the configuration automatically.
 	if err = i.saveHostRoutes(); err != nil {
@@ -117,7 +122,7 @@ func (i *Initializer) prepareHNSNetworkAndOVSExtension() error {
 	if subnetCIDR == nil {
 		return fmt.Errorf("failed to find valid IPv4 PodCIDR")
 	}
-	return util.PrepareHNSNetwork(subnetCIDR, i.nodeConfig.NodeTransportIPv4Addr, adapter, i.nodeConfig.UplinkNetConfig.Gateway, dnsServers, i.nodeConfig.UplinkNetConfig.Routes, i.ovsBridge)
+	return util.PrepareHNSNetwork(subnetCIDR, i.nodeConfig.NodeTransportIPv4Addr, adapter, i.nodeConfig.UplinkNetConfig.Gateway, dnsServers, i.nodeConfig.UplinkNetConfig.Routes, i.ovsBridge, i.nodeConfig.UplinkNetConfig.DHCP)
 }
 
 func (i *Initializer) prepareVMNetworkAndOVSExtension() error {

pkg/agent/config/node_config.go

@@ -115,6 +115,7 @@ type AdapterNetConfig struct {
 	Routes     []interface{}
 	// OFPort is the OpenFlow port number of the uplink interface allocated by OVS.
 	OFPort uint32
+	DHCP   bool
 }
 
 type WireGuardConfig struct {

pkg/agent/util/net_windows.go

@@ -447,7 +447,7 @@ func ConfigureLinkAddresses(idx int, ipNets []*net.IPNet) error {
 }
 
 // PrepareHNSNetwork creates HNS Network for containers.
-func PrepareHNSNetwork(subnetCIDR *net.IPNet, nodeIPNet *net.IPNet, uplinkAdapter *net.Interface, nodeGateway string, dnsServers string, routes []interface{}, newName string) error {
+func PrepareHNSNetwork(subnetCIDR *net.IPNet, nodeIPNet *net.IPNet, uplinkAdapter *net.Interface, nodeGateway string, dnsServers string, routes []interface{}, newName string, dhcp bool) error {
 	klog.InfoS("Creating HNSNetwork", "name", LocalHNSNetwork, "subnet", subnetCIDR, "nodeIP", nodeIPNet, "adapter", uplinkAdapter)
 	hnsNet, err := CreateHNSNetwork(LocalHNSNetwork, subnetCIDR, nodeIPNet, uplinkAdapter)
 	if err != nil {
@@ -461,7 +461,7 @@ func PrepareHNSNetwork(subnetCIDR *net.IPNet, nodeIPNet *net.IPNet, uplinkAdapte
 		}
 	}()
 
-	adapter, ipFound, err := adapterIPExists(nodeIPNet.IP, uplinkAdapter.HardwareAddr, ContainerVNICPrefix)
+	adapter, ipFound, err := adapterIPExists(nodeIPNet.IP, uplinkAdapter.HardwareAddr, ContainerVNICPrefix, dhcp)
 	if err != nil {
 		return err
 	}
@@ -521,29 +521,44 @@ func PrepareHNSNetwork(subnetCIDR *net.IPNet, nodeIPNet *net.IPNet, uplinkAdapte
 
 // adapterIPExists finds the network adapter configured with the provided IP, MAC and its name has the given prefix.
 // If "namePrefix" is empty, it returns the first network adapter with the provided IP and MAC.
+// If DHCP is enabled on the original adapter, it will attempt to acquire an IP multiple times.
 // It returns true if the IP is found on the adapter, otherwise it returns false.
-func adapterIPExists(ip net.IP, mac net.HardwareAddr, namePrefix string) (*net.Interface, bool, error) {
+func adapterIPExists(ip net.IP, mac net.HardwareAddr, namePrefix string, dhcp bool) (*net.Interface, bool, error) {
 	adapters, err := netInterfaces()
 	if err != nil {
 		return nil, false, err
 	}
+	// To avoid the issue of insufficient time for the Windows adapter to obtain an dhcp IP after the creation of the HNS network,
+	// wait for up to three seconds and retry IP acquisition in case of failures.
 	ipExists := false
 	for idx, adapter := range adapters {
 		if bytes.Equal(adapter.HardwareAddr, mac) {
 			if namePrefix == "" || strings.Contains(adapter.Name, namePrefix) {
-				addrList, err := netInterfaceAddrs(&adapters[idx])
-				if err != nil {
-					return nil, false, err
-				}
-				for _, addr := range addrList {
-					if ipNet, ok := addr.(*net.IPNet); ok {
-						if ipNet.IP.Equal(ip) {
-							ipExists = true
-							break
+				for retry := 0; retry < 4; retry++ {
+					addrList, err := netInterfaceAddrs(&adapters[idx])
+					if err != nil {
+						return nil, false, err
+					}
+					for _, addr := range addrList {
+						if ipNet, ok := addr.(*net.IPNet); ok {
+							if ipNet.IP.Equal(ip) {
+								ipExists = true
+								break
+							}
 						}
 					}
+					if ipExists {
+						return &adapter, true, nil
+					}
+					if dhcp == false {
+						return &adapter, false, nil
+					} else if !ipExists && retry < 4 {
+						time.Sleep(1 * time.Second)
+					} else {
+						klog.Warning("Unable to find an available DHCP IP with MAC %s, IP %s, and name prefix %s after retries", mac.String(), ip.String(), namePrefix)
+						return &adapter, false, nil
+					}
 				}
-				return &adapter, ipExists, nil
 			}
 		}
 	}
@@ -619,6 +634,22 @@ func SetAdapterDNSServers(adapterName, dnsServers string) error {
 	return nil
 }
 
+// GetDHCPByInterfaceIndex returns the DHCP status on the specified interface.
+func GetDHCPByInterfaceIndex(ifIndex int) (bool, error) {
+	cmd := fmt.Sprintf("$(Get-NetIPInterface -InterfaceIndex %d -AddressFamily IPv4).Dhcp", ifIndex)
+	Dhcp, err := runCommand(cmd)
+	if err != nil {
+		return false, err
+	}
+	Dhcp = strings.TrimSpace(Dhcp)
+	if Dhcp == "Enabled" {
+		klog.Infof("IPv4 DHCP is Enabled for Interface %d", ifIndex)
+		return true, nil
+	}
+	klog.Infof("IPv4 DHCP is Disabled for Interface%d", ifIndex)
+	return false, nil
+}
+
 // ListenLocalSocket creates a listener on a Unix domain socket or a Windows named pipe.
 // - If the specified address starts with "\\.\pipe\",  create a listener on the a Windows named pipe path.
 // - Else create a listener on a local Unix domain socket.

pkg/agent/util/net_windows_test.go

@@ -415,7 +415,7 @@ func TestPrepareHNSNetwork(t *testing.T) {
 			defer mockHNSNetworkCreate(tc.hnsNetworkCreateErr)()
 			defer mockHNSNetworkDelete(nil)()
 			defer mockAntreaNetIO(&antreasyscalltest.MockNetIO{CreateIPForwardEntryErr: tc.createRowErr})()
-			gotErr := PrepareHNSNetwork(testSubnetCIDR, tc.nodeIPNet, testUplinkAdapter, "testGateway", tc.dnsServers, testRoutes, tc.newName)
+			gotErr := PrepareHNSNetwork(testSubnetCIDR, tc.nodeIPNet, testUplinkAdapter, "testGateway", tc.dnsServers, testRoutes, tc.newName, false)
 			assert.Equal(t, tc.wantErr, gotErr)
 		})
 	}