test(e2e): improve additional network interfaces e2e test reliability (#2290)

This PR improves the reliability and diagnostics of the VirtualMachineAdditionalNetworkInterfaces E2E test.

Signed-off-by: Dmitry Lopatin <dmitry.lopatin@flant.com>
Signed-off-by: Dmitry Lopatin <93423466+LopatinDmitr@users.noreply.github.com>
Co-authored-by: Roman Sysoev <36233932+hardcoretime@users.noreply.github.com>

Author: LopatinDmitr

test/e2e/internal/framework/dump.go

@@ -65,6 +65,7 @@ func (f *Framework) saveTestCaseDump() {
 	f.saveIntvirtvmiDescriptions(ft, tmpDir)
 	f.saveNodeAdditionalInfo(ft, tmpDir)
 	f.saveEvents(ft, tmpDir)
+	f.saveClusterNetworkInfo(ft, tmpDir)
 }
 
 // GetFormattedTestCaseFullText returns CurrentSpecReport().FullText(), formatted with the following rules:
@@ -303,3 +304,41 @@ func (f *Framework) saveEvents(testCaseFullText, dumpPath string) {
 		}
 	}
 }
+
+func (f *Framework) saveClusterNetworkInfo(testCaseFullText, dumpPath string) {
+	GinkgoHelper()
+
+	// Only for tests that use additional networks.
+	// We use the original full text for checking because testCaseFullText may be truncated.
+	if !strings.Contains(CurrentSpecReport().FullText(), "VirtualMachineAdditionalNetworkInterfaces") {
+		return
+	}
+
+	// Get all ClusterNetwork resources
+	cmd := f.Clients.Kubectl().RawCommand("get clusternetwork -o yaml", ShortTimeout)
+	if cmd.Error() != nil {
+		GinkgoWriter.Printf("Failed to get clusternetwork:\nCmdError: %v\nStderr: %s\n", cmd.Error(), cmd.StdErr())
+	}
+
+	fileName := fmt.Sprintf("%s/e2e_failed__%s__clusternetwork.yaml", dumpPath, testCaseFullText)
+	if len(cmd.StdOutBytes()) > 0 {
+		err := os.WriteFile(fileName, cmd.StdOutBytes(), 0o644)
+		if err != nil {
+			GinkgoWriter.Printf("Failed to write clusternetwork dump:\nFile: %s\nError: %v\n", fileName, err)
+		}
+	}
+
+	// Get CEP (Cilium Endpoints) for the namespace
+	cepCmd := f.Clients.Kubectl().RawCommand(fmt.Sprintf("get cep -n %s -o yaml", f.Namespace().Name), ShortTimeout)
+	if cepCmd.Error() != nil {
+		GinkgoWriter.Printf("Failed to get cep:\nCmdError: %v\nStderr: %s\n", cepCmd.Error(), cepCmd.StdErr())
+	}
+
+	cepFileName := fmt.Sprintf("%s/e2e_failed__%s__cep.yaml", dumpPath, testCaseFullText)
+	if len(cepCmd.StdOutBytes()) > 0 {
+		err := os.WriteFile(cepFileName, cepCmd.StdOutBytes(), 0o644)
+		if err != nil {
+			GinkgoWriter.Printf("Failed to write cep dump:\nFile: %s\nError: %v\n", cepFileName, err)
+		}
+	}
+}

test/e2e/vm/additional_network_interfaces.go

@@ -18,6 +18,7 @@ package vm
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"strings"
 	"time"
@@ -183,12 +184,10 @@ var _ = Describe("VirtualMachineAdditionalNetworkInterfaces", Label(precheck.NoP
 		)
 
 		const (
-			getLastInterfaceNameCmd = "ip -o link show | tail -1 | cut -d: -f2 | awk \"{print \\$1}\""
+			expectedLastInterfaceName = "eno3"
 		)
 
 		It("should preserve interface name after removing middle ClusterNetwork and rebooting", func() {
-			var lastInterfaceNameBeforeRemoval string
-
 			By("Create VM with Main network and two additional ClusterNetworks", func() {
 				ns := f.Namespace().Name
 
@@ -216,10 +215,7 @@ var _ = Describe("VirtualMachineAdditionalNetworkInterfaces", Label(precheck.NoP
 
 			By("Get last interface name via SSH", func() {
 				util.UntilSSHReady(f, vm, framework.LongTimeout)
-				output, err := f.SSHCommand(vm.Name, vm.Namespace, getLastInterfaceNameCmd)
-				Expect(err).NotTo(HaveOccurred())
-				lastInterfaceNameBeforeRemoval = strings.TrimSpace(output)
-				Expect(lastInterfaceNameBeforeRemoval).NotTo(BeEmpty(), "Failed to get last interface name")
+				checkLastInterfaceName(f, vm.Name, vm.Namespace, expectedLastInterfaceName)
 			})
 
 			By("Remove middle ClusterNetwork from VM spec", func() {
@@ -253,13 +249,7 @@ var _ = Describe("VirtualMachineAdditionalNetworkInterfaces", Label(precheck.NoP
 
 			By("Verify last interface name has not changed", func() {
 				util.UntilSSHReady(f, vm, framework.LongTimeout)
-				output, err := f.SSHCommand(vm.Name, vm.Namespace, getLastInterfaceNameCmd)
-				Expect(err).NotTo(HaveOccurred())
-				lastInterfaceNameAfterRemoval := strings.TrimSpace(output)
-				Expect(lastInterfaceNameAfterRemoval).NotTo(BeEmpty(), "Failed to get last interface name")
-
-				Expect(lastInterfaceNameAfterRemoval).To(Equal(lastInterfaceNameBeforeRemoval),
-					fmt.Sprintf("Interface name changed from %s to %s after removing middle ClusterNetwork", lastInterfaceNameBeforeRemoval, lastInterfaceNameAfterRemoval))
+				checkLastInterfaceName(f, vm.Name, vm.Namespace, expectedLastInterfaceName)
 			})
 		})
 	})
@@ -361,3 +351,33 @@ func checkResultSSHCommand(f *framework.Framework, vmName, vmNamespace, cmd, equ
 		return strings.TrimSpace(res), nil
 	}).WithTimeout(Timeout).WithPolling(Interval).Should(Equal(equal))
 }
+
+func checkLastInterfaceName(f *framework.Framework, vmName, vmNamespace, expected string) {
+	GinkgoHelper()
+	Eventually(func() (string, error) {
+		cmd := "ip -j link show"
+		result, err := f.SSHCommand(vmName, vmNamespace, cmd, framework.WithSSHTimeout(5*time.Second))
+		if err != nil {
+			return "", fmt.Errorf("failed to execute command: %w: %s", err, result)
+		}
+
+		var links IPLinks
+		err = json.Unmarshal([]byte(result), &links)
+		if err != nil {
+			return "", fmt.Errorf("failed to parse ip JSON output: %w", err)
+		}
+		if len(links) == 0 {
+			return "", fmt.Errorf("no network interfaces found")
+		}
+
+		return links[len(links)-1].IFName, nil
+	}).WithTimeout(Timeout).WithPolling(Interval).Should(Equal(expected))
+}
+
+// IPLinks represents the JSON output of ip -j link show command.
+type IPLinks []IPLink
+
+// IPLink represents a single network interface in the ip JSON output.
+type IPLink struct {
+	IFName string `json:"ifname"`
+}