Merge pull request #30935 from ngopalak-redhat/ngopalak/system_compressible

OCPNODE-3725: Add System compressible test

Author: openshift-merge-bot[bot]

test/extended/node/node_sizing.go

@@ -9,6 +9,7 @@ import (
 	g "github.com/onsi/ginkgo/v2"
 	o "github.com/onsi/gomega"
 	corev1 "k8s.io/api/core/v1"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/kubernetes/test/e2e/framework"
@@ -62,7 +63,9 @@ var _ = g.Describe("[Suite:openshift/disruptive-longrunning][sig-node][Disruptiv
 			// Use JSON patch to remove the label atomically
 			patchData := []byte(fmt.Sprintf(`{"metadata":{"labels":{%q:null}}}`, testNodeMCPLabel))
 			_, updateErr := oc.AdminKubeClient().CoreV1().Nodes().Patch(cleanupCtx, nodeName, types.MergePatchType, patchData, metav1.PatchOptions{})
-			if updateErr != nil {
+			if apierrors.IsNotFound(updateErr) {
+				// Node already deleted, nothing to clean up
+			} else if updateErr != nil {
 				framework.Logf("Failed to remove label from node %s: %v", nodeName, updateErr)
 				return
 			}
@@ -135,7 +138,9 @@ var _ = g.Describe("[Suite:openshift/disruptive-longrunning][sig-node][Disruptiv
 			g.By("Cleaning up custom MachineConfigPool")
 			cleanupCtx := context.Background()
 			deleteErr := mcClient.MachineconfigurationV1().MachineConfigPools().Delete(cleanupCtx, testMCPName, metav1.DeleteOptions{})
-			if deleteErr != nil {
+			if apierrors.IsNotFound(deleteErr) {
+				// MachineConfigPool already deleted, nothing to clean up
+			} else if deleteErr != nil {
 				framework.Logf("Failed to delete MachineConfigPool %s: %v", testMCPName, deleteErr)
 			}
 		}
@@ -180,14 +185,18 @@ var _ = g.Describe("[Suite:openshift/disruptive-longrunning][sig-node][Disruptiv
 			g.By("Cleaning up KubeletConfig")
 			cleanupCtx := context.Background()
 			deleteErr := mcClient.MachineconfigurationV1().KubeletConfigs().Delete(cleanupCtx, kubeletConfigName, metav1.DeleteOptions{})
-			if deleteErr != nil {
+			if apierrors.IsNotFound(deleteErr) {
+				// KubeletConfig already deleted, nothing to clean up
+			} else if deleteErr != nil {
 				framework.Logf("Failed to delete KubeletConfig %s: %v", kubeletConfigName, deleteErr)
 			}
 
 			// Wait for custom MCP to be ready after cleanup
 			g.By("Waiting for custom MCP to be ready after KubeletConfig deletion")
 			waitErr := waitForMCP(cleanupCtx, mcClient, testMCPName, 5*time.Minute)
-			if waitErr != nil {
+			if apierrors.IsNotFound(waitErr) {
+				// MachineConfigPool already deleted, nothing to wait for
+			} else if waitErr != nil {
 				framework.Logf("Failed to wait for custom MCP to be ready: %v", waitErr)
 			}
 		}

test/extended/node/node_swap.go

@@ -118,11 +118,9 @@ var _ = g.Describe("[Jira:Node][sig-node] Node non-cnv swap configuration", func
 		g.By("Getting initial machine config resourceVersion")
 		// Get the initial resourceVersion of the worker machine config before creating KubeletConfig
 		workerMC, err := mcClient.MachineconfigurationV1().MachineConfigs().Get(ctx, workerGeneratedKubeletMC, metav1.GetOptions{})
-		initialResourceVersion := ""
-		if err == nil {
-			initialResourceVersion = workerMC.ResourceVersion
-			framework.Logf("Initial %s resourceVersion: %s", workerGeneratedKubeletMC, initialResourceVersion)
-		}
+		o.Expect(err).NotTo(o.HaveOccurred(), "Failed to get initial machine config %s", workerGeneratedKubeletMC)
+		initialResourceVersion := workerMC.ResourceVersion
+		framework.Logf("Initial %s resourceVersion: %s", workerGeneratedKubeletMC, initialResourceVersion)
 
 		g.By("Creating a KubeletConfig with swap settings")
 		kcName := fmt.Sprintf("test-swap-override-%d", time.Now().UnixNano())
@@ -190,10 +188,9 @@ var _ = g.Describe("[Jira:Node][sig-node] Node non-cnv swap configuration", func
 
 		// Check if the machine config was created or updated (compare to initial resourceVersion captured earlier)
 		workerMC, err = mcClient.MachineconfigurationV1().MachineConfigs().Get(ctx, workerGeneratedKubeletMC, metav1.GetOptions{})
-		if err == nil {
-			o.Expect(workerMC.ResourceVersion).To(o.Equal(initialResourceVersion), "Machine config %s should not be updated when failSwapOn is rejected", workerGeneratedKubeletMC)
-			framework.Logf("Verified: %s was not updated (resourceVersion: %s)", workerGeneratedKubeletMC, workerMC.ResourceVersion)
-		}
+		o.Expect(err).NotTo(o.HaveOccurred(), "Failed to get machine config %s for verification", workerGeneratedKubeletMC)
+		o.Expect(workerMC.ResourceVersion).To(o.Equal(initialResourceVersion), "Machine config %s should not be updated when failSwapOn is rejected", workerGeneratedKubeletMC)
+		framework.Logf("Verified: %s was not updated (resourceVersion: %s)", workerGeneratedKubeletMC, workerMC.ResourceVersion)
 
 		g.By("Verifying worker nodes still have correct swap settings")
 		allWorkerNodes, err := getNodesByLabel(ctx, oc, "node-role.kubernetes.io/worker")

test/extended/node/node_swap_cnv.go

@@ -894,25 +894,34 @@ var _ = g.Describe("[Jira:Node/Kubelet][sig-node][Feature:NodeSwap][Serial][Disr
 
 		g.By("Checking initial OS-level swap status")
 		framework.Logf("Running: swapon -s")
-		initialSwapOutput, _ := ExecOnNodeWithChroot(oc, cnvWorkerNode, "swapon", "-s")
+		initialSwapOutput, err := ExecOnNodeWithChroot(oc, cnvWorkerNode, "swapon", "-s")
+		o.Expect(err).NotTo(o.HaveOccurred(), "Failed to check initial swap status on node %s: %v", cnvWorkerNode, err)
 		framework.Logf("Initial swapon -s output:\n%s", initialSwapOutput)
 		initialHasSwap := strings.TrimSpace(initialSwapOutput) != "" && initialSwapOutput != "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority"
 
 		// If swap is already enabled, disable it for this test
 		if initialHasSwap {
 			g.By("Disabling existing OS-level swap for test")
 			framework.Logf("Running: swapoff -a")
-			_, _ = ExecOnNodeWithNsenter(oc, cnvWorkerNode, "swapoff", "-a")
+			swapoffOutput, swapoffErr := ExecOnNodeWithNsenter(oc, cnvWorkerNode, "swapoff", "-a")
+			if swapoffErr != nil {
+				framework.Failf("Failed to disable swap on node %s: %v (output: %s)", cnvWorkerNode, swapoffErr, swapoffOutput)
+			}
 			framework.Logf("OS-level swap disabled")
 		}
 
 		g.By("Verifying no OS-level swap is present")
 		framework.Logf("Running: swapon -s")
-		swapOutput, _ := ExecOnNodeWithChroot(oc, cnvWorkerNode, "swapon", "-s")
+		swapOutput, err := ExecOnNodeWithChroot(oc, cnvWorkerNode, "swapon", "-s")
+		o.Expect(err).NotTo(o.HaveOccurred(), "Failed to verify swap status on node %s: %v", cnvWorkerNode, err)
 		framework.Logf("swapon -s output:\n%s", swapOutput)
 		hasOSSwap := strings.TrimSpace(swapOutput) != "" && swapOutput != "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority"
 		if hasOSSwap {
-			framework.Logf("Warning: Could not disable OS swap, but continuing with test")
+			if initialHasSwap {
+				framework.Failf("OS-level swap still present on node %s after swapoff attempt. Output:\n%s", cnvWorkerNode, swapOutput)
+			} else {
+				framework.Failf("Unexpected OS-level swap found on node %s. Output:\n%s", cnvWorkerNode, swapOutput)
+			}
 		} else {
 			framework.Logf("Confirmed: No OS-level swap on node %s", cnvWorkerNode)
 		}
@@ -924,7 +933,7 @@ var _ = g.Describe("[Jira:Node/Kubelet][sig-node][Feature:NodeSwap][Serial][Disr
 		g.By("Creating LimitedSwap drop-in configuration")
 		framework.Logf("Creating drop-in file: %s", cnvDropInFilePath)
 		framework.Logf("Content:\n%s", loadConfigFromFile(cnvLimitedSwapConfigPath))
-		err := createDropInFile(oc, cnvWorkerNode, cnvDropInFilePath, loadConfigFromFile(cnvLimitedSwapConfigPath))
+		err = createDropInFile(oc, cnvWorkerNode, cnvDropInFilePath, loadConfigFromFile(cnvLimitedSwapConfigPath))
 		o.Expect(err).NotTo(o.HaveOccurred())
 		framework.Logf("Drop-in file created successfully")
 
@@ -1099,9 +1108,16 @@ var _ = g.Describe("[Jira:Node/Kubelet][sig-node][Feature:NodeSwap][Serial][Disr
 			}
 
 			g.By(fmt.Sprintf("Disabling any existing swap for %s test", swapSize.name))
-			framework.Logf("Running: swapoff -a")
-			ExecOnNodeWithNsenter(oc, cnvWorkerNode, "swapoff", "-a")
-			ExecOnNodeWithChroot(oc, cnvWorkerNode, "rm", "-f", swapFilePath)
+			framework.Logf("Running: swapoff -a on node %s", cnvWorkerNode)
+			swapoffOutput, swapoffErr := ExecOnNodeWithNsenter(oc, cnvWorkerNode, "swapoff", "-a")
+			if swapoffErr != nil {
+				framework.Failf("Failed to disable swap on node %s for %s test: %v (output: %s)", cnvWorkerNode, swapSize.name, swapoffErr, swapoffOutput)
+			}
+			framework.Logf("Running: rm -f %s on node %s", swapFilePath, cnvWorkerNode)
+			rmOutput, rmErr := ExecOnNodeWithChroot(oc, cnvWorkerNode, "rm", "-f", swapFilePath)
+			if rmErr != nil {
+				framework.Failf("Failed to remove swap file %s on node %s for %s test: %v (output: %s)", swapFilePath, cnvWorkerNode, swapSize.name, rmErr, rmOutput)
+			}
 
 			g.By(fmt.Sprintf("Creating %dMB swap file", swapSize.sizeMB))
 			framework.Logf("Running: dd if=/dev/zero of=%s bs=1M count=%d", swapFilePath, swapSize.sizeMB)

test/extended/node/node_utils.go

@@ -393,12 +393,12 @@ func installCNVOperator(ctx context.Context, oc *exutil.CLI) error {
 	framework.Logf("Checking MCP rollout status...")
 	mcClient, err := machineconfigclient.NewForConfig(oc.AdminConfig())
 	if err != nil {
-		framework.Logf("Warning: failed to create MC client for MCP check: %v", err)
-	} else {
-		err = waitForMCP(ctx, mcClient, "worker", 30*time.Minute)
-		if err != nil {
-			framework.Logf("Warning: MCP rollout check failed: %v", err)
-		}
+		return fmt.Errorf("failed to create MC client for MCP check: %w", err)
+	}
+
+	err = waitForMCP(ctx, mcClient, "worker", 30*time.Minute)
+	if err != nil {
+		return fmt.Errorf("MCP rollout failed after CNV installation: %w", err)
 	}
 
 	framework.Logf("CNV operator installed successfully")