Fix unexpected interface store initialization for FlexibleIPAM uplink internal port (#7389) (#7432)

* Fix unexpected interface store initialization for FlexibleIPAM uplink internal port

Signed-off-by: Ran Gu <[email protected]>

Author: gran-vmv

pkg/agent/agent.go

@@ -339,15 +339,15 @@ func (i *Initializer) initInterfaceStore() error {
 		case interfacestore.AntreaIPsecTunnel:
 			intf = parseTunnelInterfaceFunc(port, ovsPort)
 		case interfacestore.AntreaHost:
-			if port.Name == i.ovsBridge {
-				// Need not to load the OVS bridge port to the interfaceStore
-				intf = nil
-			} else {
+			if i.nodeType == config.ExternalNode {
 				var err error
 				intf, err = externalnode.ParseHostInterfaceConfig(i.ovsBridgeClient, port, ovsPort)
 				if err != nil {
 					return fmt.Errorf("failed to get interfaceConfig by port %s: %v", port.Name, err)
 				}
+			} else {
+				// No need to load the OVS host Interface to the interfaceStore
+				intf = nil
 			}
 		case interfacestore.AntreaContainer:
 			// The port should be for a container interface.

pkg/agent/agent_test.go

@@ -96,7 +96,13 @@ func TestInitInterfaceStore(t *testing.T) {
 			interfacestore.NewContainerInterface("p2", uuid2, "pod2", "ns2", "eth0", "netns2", p2NetMAC, []net.IP{p2NetIP}, 0),
 		)),
 	}
-	initOVSPorts := []ovsconfig.OVSPortData{ovsPort1, ovsPort2}
+	uuid3 := uuid.New().String()
+	ovsPortUplinkInternal := ovsconfig.OVSPortData{UUID: uuid3, Name: "eth-antrea-test-1", IFName: "eth-antrea-test-1", OFPort: 100,
+		ExternalIDs: map[string]string{
+			interfacestore.AntreaInterfaceTypeKey: interfacestore.AntreaHost,
+		},
+	}
+	initOVSPorts := []ovsconfig.OVSPortData{ovsPort1, ovsPort2, ovsPortUplinkInternal}
 
 	mockOVSBridgeClient.EXPECT().GetPortList().Return(initOVSPorts, ovsconfig.NewTransactionError(fmt.Errorf("Failed to list OVS ports"), true))
 	initializer.initInterfaceStore()

test/e2e/antreaipam_test.go

@@ -279,8 +279,10 @@ func TestAntreaIPAM(t *testing.T) {
 	t.Run("testAntreaIPAMOVSRestartSameNode", func(t *testing.T) {
 		skipIfNotIPv4Cluster(t)
 		skipIfHasWindowsNodes(t)
-		testOVSRestartSameNode(t, data, testAntreaIPAMNamespace)
-		testOVSRestartSameNode(t, data, testAntreaIPAMNamespace11)
+		testOVSRestartSameNode(t, data, testAntreaIPAMNamespace, false)
+		testOVSRestartSameNode(t, data, testAntreaIPAMNamespace11, false)
+		testOVSRestartSameNode(t, data, testAntreaIPAMNamespace, true)
+		testOVSRestartSameNode(t, data, testAntreaIPAMNamespace11, true)
 		checkIPPoolsEmpty(t, data, ipPools)
 	})
 	t.Run("testAntreaIPAMOVSFlowReplay", func(t *testing.T) {

test/e2e/connectivity_test.go

@@ -60,7 +60,7 @@ func TestConnectivity(t *testing.T) {
 	t.Run("testOVSRestartSameNode", func(t *testing.T) {
 		skipIfNotIPv4Cluster(t)
 		skipIfHasWindowsNodes(t)
-		testOVSRestartSameNode(t, data, data.testNamespace)
+		testOVSRestartSameNode(t, data, data.testNamespace, false)
 	})
 	t.Run("testOVSFlowReplay", func(t *testing.T) {
 		skipIfHasWindowsNodes(t)
@@ -425,7 +425,7 @@ func testPodConnectivityAfterAntreaRestart(t *testing.T, data *TestData, namespa
 // testOVSRestartSameNode verifies that datapath flows are not removed when the Antrea Agent Pod is
 // stopped gracefully (e.g. as part of a RollingUpdate). The test sends ARP requests every 1s and
 // checks that there is no packet loss during the restart.
-func testOVSRestartSameNode(t *testing.T, data *TestData, namespace string) {
+func testOVSRestartSameNode(t *testing.T, data *TestData, namespace string, force bool) {
 	workerNode := workerNodeName(1)
 	t.Logf("Creating two toolbox test Pods on '%s'", workerNode)
 	podNames, podIPs, cleanupFn := createTestToolboxPods(t, data, 2, namespace, workerNode)
@@ -469,8 +469,24 @@ func testOVSRestartSameNode(t *testing.T, data *TestData, namespace string) {
 	time.Sleep(3 * time.Second)
 
 	t.Logf("Restarting antrea-agent on Node '%s'", workerNode)
-	if _, err := data.deleteAntreaAgentOnNode(workerNode, 30 /* grace period in seconds */, defaultTimeout); err != nil {
-		t.Fatalf("Error when restarting antrea-agent on Node '%s': %v", workerNode, err)
+	if !force {
+		if _, err := data.deleteAntreaAgentOnNode(workerNode, 30 /* grace period in seconds */, defaultTimeout); err != nil {
+			t.Fatalf("Error when restarting antrea-agent on Node '%s': %v", workerNode, err)
+		}
+	} else {
+		signalAgent := func(nodeName, signal string) {
+			cmd := fmt.Sprintf("pkill -%s antrea-agent", signal)
+			if testOptions.providerName != "kind" {
+				cmd = "sudo " + cmd
+			}
+			rc, stdout, stderr, err := data.RunCommandOnNode(nodeName, cmd)
+			if rc != 0 || err != nil {
+				t.Errorf("Error when running command '%s' on Node '%s', rc: %d, stdout: %s, stderr: %s, error: %v",
+					cmd, nodeName, rc, stdout, stderr, err)
+			}
+		}
+		// Send "KILL" signal to antrea-agent.
+		signalAgent(workerNode, "KILL")
 	}
 
 	if err := <-resCh; err != nil {