Update E2E tests for Metal3Remediation with out-of-service taint

Signed-off-by: Carlo Lobrano <[email protected]>

test/e2e/common.go

@@ -52,6 +52,9 @@ const (
 	osTypeCentos           = "centos"
 	osTypeUbuntu           = "ubuntu"
 	ironicSuffix           = "-ironic"
+	// Out-of-service Taint test actions
+	oostAdded   = "added"
+	oostRemoved = "removed"
 )
 
 var releaseMarkerPrefix = "go://github.com/metal3-io/cluster-api-provider-metal3@v%s"

test/e2e/node_deletion_remediation.go

@@ -8,6 +8,7 @@ import (
 	infrav1 "github.com/metal3-io/cluster-api-provider-metal3/api/v1beta1"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
+	"golang.org/x/mod/semver"
 	corev1 "k8s.io/api/core/v1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -17,7 +18,9 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client"
 )
 
-type NodeDeletionRemediation struct {
+const minK8sVersionOutOfServiceTaint = "1.28"
+
+type NodeRemediation struct {
 	E2EConfig             *clusterctl.E2EConfig
 	BootstrapClusterProxy framework.ClusterProxy
 	TargetCluster         framework.ClusterProxy
@@ -27,22 +30,30 @@ type NodeDeletionRemediation struct {
 }
 
 /*
- * Node Deletion Remediation Test
+ * Node Remediation Test
  *
- * This test evaluates node deletion in reboot remediation feature added to CAPM3 Remediation Controller.
+ * This test evaluates node remediation (via deletion or use of out-of-service taint) in reboot remediation feature added to CAPM3 Remediation Controller.
  * issue #392: Reboot remediation is incomplete
  * PR #668: Fix reboot remediation by adding node deletion
- * This test evaluates the reboot remediation strategy with an enhancement of node deletion in the CAPM3 (Cluster API Provider for Metal3) Remediation Controller.
+ * This test evaluates the reboot remediation strategy with an enhancement in the CAPM3 (Cluster API Provider for Metal3) Remediation Controller
+ * consisting in:
+ * - node deletion (kubernetes server version < 1.28)
+ * - out-of-service taint on node (kubernetes server version >= 1.28)
  *
  * Tested Feature:
- * - Delete Node in Reboot Remediation
+ * - Manage Node in Reboot Remediation (deletion or out-of-service taint)
  *
  * Workflow:
  * 1. Retrieve the Metal3Machines associated with the worker nodes in the cluster.
  * 2. Identify the target worker machine node its associated BMH object corresponding to the Metal3Machine.
  * 3. Create a Metal3Remediation resource, specifying the remediation strategy as "Reboot" with a retry limit and timeout.
  * 4. Wait for the VM (Virtual Machine) associated with target BMH to power off.
- * 5. Wait for the target worker node to be deleted from the cluster.
+ * 5. Identify the kubernetes service version:
+ * 5a. if version < 1.28:
+ * - Wait for the target worker node to be deleted from the cluster.
+ * 5b. if version < 1.28:
+ * - Wait for the out-of-service taint to be set on target worker node.
+ * - Wait for the out-of-service taint to be removed from target worker node.
  * 6. Wait for the VMs to power on.
  * 7. Verify that the target worker node becomes ready.
  * 8. Verify that the Metal3Remediation resource is successfully delete
@@ -52,8 +63,8 @@ type NodeDeletionRemediation struct {
  * resiliency of the cluster by allowing workloads to be seamlessly migrated from unhealthy nodes to healthy node
  */
 
-func nodeDeletionRemediation(ctx context.Context, inputGetter func() NodeDeletionRemediation) {
-	Logf("Starting node deletion remediation tests")
+func nodeRemediation(ctx context.Context, inputGetter func() NodeRemediation) {
+	Logf("Starting node remediation tests")
 	input := inputGetter()
 	bootstrapClient := input.BootstrapClusterProxy.GetClient()
 	targetClient := input.TargetCluster.GetClient()
@@ -106,9 +117,18 @@ func nodeDeletionRemediation(ctx context.Context, inputGetter func() NodeDeletio
 	By("Waiting for VM power off")
 	waitForVmsState([]string{vmName}, shutoff, input.SpecName, input.E2EConfig.GetIntervals(input.SpecName, "wait-vm-state")...)
 
-	By("Waiting for node deletion")
-	interval := input.E2EConfig.GetIntervals(input.SpecName, "wait-vm-state")
-	waitForNodeDeletion(ctx, targetClient, workerNodeName, interval...)
+	k8sVersion := input.E2EConfig.GetVariable("KUBERNETES_VERSION")
+	if isOutOfServiceTaintSupported(k8sVersion) {
+		Byf("Waiting for Out of service taint on node to be added (kubernetes version %s)", k8sVersion)
+		interval := input.E2EConfig.GetIntervals(input.SpecName, "wait-vm-state")
+		waitForOutOfServiceTaint(ctx, targetClient, workerNodeName, oostAdded, interval...)
+		Byf("Waiting for Out of service taint on node to be removed (kubernetes version %s)", k8sVersion)
+		waitForOutOfServiceTaint(ctx, targetClient, workerNodeName, oostRemoved, interval...)
+	} else {
+		By("Waiting for node deletion")
+		interval := input.E2EConfig.GetIntervals(input.SpecName, "wait-vm-state")
+		waitForNodeDeletion(ctx, targetClient, workerNodeName, interval...)
+	}
 
 	By("Waiting for VM power on")
 	waitForVmsState([]string{vmName}, running, input.SpecName, input.E2EConfig.GetIntervals(input.SpecName, "wait-vm-state")...)
@@ -125,7 +145,7 @@ func nodeDeletionRemediation(ctx context.Context, inputGetter func() NodeDeletio
 		return apierrors.IsNotFound(err)
 	}, 2*time.Minute, 10*time.Second).Should(BeTrue(), "Metal3Remediation should have been deleted")
 
-	By("NODE DELETION TESTS PASSED!")
+	By("NODE REMEDIATION TESTS PASSED!")
 }
 
 func waitForNodeDeletion(ctx context.Context, cl client.Client, name string, intervals ...interface{}) {
@@ -137,3 +157,30 @@ func waitForNodeDeletion(ctx context.Context, cl client.Client, name string, int
 			return apierrors.IsNotFound(err)
 		}, intervals...).Should(BeTrue())
 }
+
+func waitForOutOfServiceTaint(ctx context.Context, cl client.Client, name, action string, intervals ...interface{}) {
+	Byf("Waiting for Out of service taint to Node '%s' to be %s", name, action)
+	var oostExpectedToExist bool
+	if action == oostAdded {
+		oostExpectedToExist = true
+	}
+	Eventually(
+		func() bool {
+			node := &corev1.Node{}
+			err := cl.Get(ctx, client.ObjectKey{Name: name}, node)
+			Expect(err).To(BeNil())
+			for _, t := range node.Spec.Taints {
+				if t.Key == "node.kubernetes.io/out-of-service" &&
+					t.Value == "nodeshutdown" &&
+					t.Effect == corev1.TaintEffectNoExecute {
+					return true == oostExpectedToExist
+				}
+			}
+			return false == oostExpectedToExist
+		}, intervals...).Should(BeTrue())
+}
+
+func isOutOfServiceTaintSupported(k8sVersion string) bool {
+	Byf("comparing current version %s with supported version %s", k8sVersion, minK8sVersionOutOfServiceTaint)
+	return semver.Compare(k8sVersion, minK8sVersionOutOfServiceTaint) >= 0
+}

test/e2e/remediation_based_feature_test.go

@@ -17,15 +17,19 @@ import (
  * These tests involve simulating failure scenarios, triggering the remediation process, and then verifying that the remediation actions successfully restore the nodes to the desired state.
  *
  * Test Types:
- * 1. Metal3Remediation Test: This test specifically evaluates the Metal3 Remediation Controller's node deletion feature in the reboot remediation strategy.
+ * 1. Metal3Remediation Test: This test specifically evaluates the Metal3 Remediation Controller's node management feature in the reboot remediation strategy.
  * 2. Remediation Test: This test focuses on verifying various annotations and actions related to remediation in the CAPM3 (Cluster API Provider for Metal3).
  *
- * NodeDeletionRemediation Test:
+ * NodeRemediation Test:
  * - Retrieve the list of Metal3 machines associated with the worker nodes.
  * - Identify the target worker Metal3Machine and its corresponding BareMetalHost (BMH) object.
  * - Create a Metal3Remediation resource with a remediation strategy of type "Reboot" and a specified timeout.
  * - Wait for the associated virtual machine (VM) to power off.
- * - Wait for the node (VM) to be deleted.
+ * - If kubernetes server version < 1.28:
+ *   - Wait for the node (VM) to be deleted.
+ * - If kubernetes server version >= 1.28:
+ *   - Wait for the out-of-service taint to be set on the node.
+ *   - Wait for the out-of-service taint to be removed from the node.
  * - Wait for the VM to power on.
  * - Wait for the node to be in a ready state.
  * - Delete the Metal3Remediation resource.
@@ -71,9 +75,9 @@ var _ = Describe("Testing nodes remediation [remediation] [features]", Label("re
 		targetCluster, _ = createTargetCluster(e2eConfig.GetVariable("KUBERNETES_VERSION"))
 
 		// Run Metal3Remediation test first, doesn't work after remediation...
-		By("Running node deletion remediation tests")
-		nodeDeletionRemediation(ctx, func() NodeDeletionRemediation {
-			return NodeDeletionRemediation{
+		By("Running node remediation tests")
+		nodeRemediation(ctx, func() NodeRemediation {
+			return NodeRemediation{
 				E2EConfig:             e2eConfig,
 				BootstrapClusterProxy: bootstrapClusterProxy,
 				TargetCluster:         targetCluster,

test/go.mod

@@ -12,6 +12,7 @@ require (
 	github.com/onsi/gomega v1.33.1
 	github.com/pkg/errors v0.9.1
 	golang.org/x/crypto v0.24.0
+	golang.org/x/mod v0.18.0
 	gopkg.in/yaml.v3 v3.0.1
 	k8s.io/api v0.29.5
 	k8s.io/apiextensions-apiserver v0.29.5
@@ -116,7 +117,6 @@ require (
 	go.opentelemetry.io/otel/trace v1.22.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
 	golang.org/x/exp v0.0.0-20230905200255-921286631fa9 // indirect
-	golang.org/x/mod v0.17.0 // indirect
 	golang.org/x/net v0.26.0 // indirect
 	golang.org/x/oauth2 v0.18.0 // indirect
 	golang.org/x/sync v0.7.0 // indirect

test/go.sum

@@ -305,8 +305,8 @@ golang.org/x/exp v0.0.0-20230905200255-921286631fa9/go.mod h1:S2oDrQGGwySpoQPVqR
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.6.0-dev.0.20220419223038-86c51ed26bb4/go.mod h1:jJ57K6gSWd91VN4djpZkiMVwK6gcyfeH4XE8wZrZaV4=
-golang.org/x/mod v0.17.0 h1:zY54UmvipHiNd+pm+m0x9KhZ9hl1/7QNMyxXbc6ICqA=
-golang.org/x/mod v0.17.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/mod v0.18.0 h1:5+9lSbEzPSdWkH32vYPBwEpX8KwDbM52Ud9xBUvNlb0=
+golang.org/x/mod v0.18.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=