Merge pull request #2685 from okozachenko1203/release-0.11-fix-panic-on-server-deletion

k8s-ci-robot · web-flow · commit fc4946cbf93e · 2025-09-11T23:20:10.000-07:00
🐛 [release-0.11] fix: fix panic on server deletion
diff --git a/api/v1beta1/conditions_consts.go b/api/v1beta1/conditions_consts.go
@@ -44,6 +44,9 @@ const (
 	OpenStackErrorReason = "OpenStackError"
 	// DependencyFailedReason indicates that a dependent object failed.
 	DependencyFailedReason = "DependencyFailed"
+
+	// ServerUnexpectedDeletedMessage is the message used when the server is unexpectedly deleted via an external agent.
+	ServerUnexpectedDeletedMessage = "The server was unexpectedly deleted"
 )
 
 const (
diff --git a/controllers/openstackmachine_controller.go b/controllers/openstackmachine_controller.go
@@ -377,6 +377,12 @@ func (r *OpenStackMachineReconciler) reconcileNormal(ctx context.Context, scope
 		return ctrl.Result{}, err
 	}
 
+	if instanceStatus == nil {
+		conditions.MarkFalse(openStackMachine, infrav1.InstanceReadyCondition, infrav1.InstanceDeletedReason, clusterv1.ConditionSeverityError, infrav1.ServerUnexpectedDeletedMessage)
+		openStackMachine.SetFailure(capierrors.UpdateMachineError, errors.New(infrav1.ServerUnexpectedDeletedMessage)) //nolint:stylecheck // This error is not used as an error
+		return ctrl.Result{}, nil
+	}
+
 	instanceNS, err := instanceStatus.NetworkStatus()
 	if err != nil {
 		return ctrl.Result{}, fmt.Errorf("get network status: %w", err)
@@ -393,22 +399,20 @@ func (r *OpenStackMachineReconciler) reconcileNormal(ctx context.Context, scope
 	})
 	openStackMachine.Status.Addresses = addresses
 
-	result := r.reconcileMachineState(scope, openStackMachine, machine, machineServer)
-	if result != nil {
-		return *result, nil
-	}
+	if util.IsControlPlaneMachine(machine) {
+		err = r.reconcileAPIServerLoadBalancer(scope, openStackCluster, openStackMachine, instanceStatus, instanceNS, clusterResourceName)
+		if err != nil {
+			return ctrl.Result{}, err
+		}
 
-	if !util.IsControlPlaneMachine(machine) {
-		scope.Logger().Info("Not a Control plane machine, no floating ip reconcile needed, Reconciled Machine create successfully")
-		return ctrl.Result{}, nil
+		conditions.MarkTrue(openStackMachine, infrav1.APIServerIngressReadyCondition)
 	}
 
-	err = r.reconcileAPIServerLoadBalancer(scope, openStackCluster, openStackMachine, instanceStatus, instanceNS, clusterResourceName)
-	if err != nil {
-		return ctrl.Result{}, err
+	result := r.reconcileMachineState(scope, openStackMachine, machine, machineServer)
+	if result != nil {
+		return *result, nil
 	}
 
-	conditions.MarkTrue(openStackMachine, infrav1.APIServerIngressReadyCondition)
 	scope.Logger().Info("Reconciled Machine create successfully")
 	return ctrl.Result{}, nil
 }
diff --git a/controllers/openstackserver_controller.go b/controllers/openstackserver_controller.go
@@ -433,9 +433,18 @@ func (r *OpenStackServerReconciler) getOrCreateServer(ctx context.Context, logge
 
 	if openStackServer.Status.InstanceID != nil {
 		instanceStatus, err = computeService.GetInstanceStatus(*openStackServer.Status.InstanceID)
-		if err != nil {
-			logger.Info("Unable to get OpenStack instance", "name", openStackServer.Name)
-			conditions.MarkFalse(openStackServer, infrav1.InstanceReadyCondition, infrav1.OpenStackErrorReason, clusterv1.ConditionSeverityError, "%s", err.Error())
+		if err != nil || instanceStatus == nil {
+			logger.Info("Unable to get OpenStack instance", "name", openStackServer.Name, "id", *openStackServer.Status.InstanceID)
+			var msg string
+			var reason string
+			if err != nil {
+				msg = err.Error()
+				reason = infrav1.OpenStackErrorReason
+			} else {
+				msg = infrav1.ServerUnexpectedDeletedMessage
+				reason = infrav1.InstanceNotFoundReason
+			}
+			conditions.MarkFalse(openStackServer, infrav1.InstanceReadyCondition, reason, clusterv1.ConditionSeverityError, msg)
 			return nil, err
 		}
 	}
@@ -450,11 +459,6 @@ func (r *OpenStackServerReconciler) getOrCreateServer(ctx context.Context, logge
 			logger.Info("Server already exists", "name", openStackServer.Name, "id", instanceStatus.ID())
 			return instanceStatus, nil
 		}
-		if openStackServer.Status.InstanceID != nil {
-			logger.Info("Not reconciling server in failed state. The previously existing OpenStack instance is no longer available")
-			conditions.MarkFalse(openStackServer, infrav1.InstanceReadyCondition, infrav1.InstanceNotFoundReason, clusterv1.ConditionSeverityError, "virtual machine no longer exists")
-			return nil, nil
-		}
 
 		logger.Info("Server does not exist, creating Server", "name", openStackServer.Name)
 		instanceSpec, err := r.serverToInstanceSpec(ctx, openStackServer)
@@ -468,6 +472,8 @@ func (r *OpenStackServerReconciler) getOrCreateServer(ctx context.Context, logge
 			openStackServer.Status.InstanceState = &infrav1.InstanceStateError
 			return nil, fmt.Errorf("create OpenStack instance: %w", err)
 		}
+		// We reached a point where a server was created with no error but we can't predict its state yet which is why we don't update conditions yet.
+		// The actual state of the server is checked in the next reconcile loops.
 		return instanceStatus, nil
 	}
 	return instanceStatus, nil
diff --git a/controllers/openstackserver_controller_test.go b/controllers/openstackserver_controller_test.go
@@ -31,8 +31,11 @@ import (
 	"github.com/gophercloud/gophercloud/v2/openstack/networking/v2/ports"
 	. "github.com/onsi/gomega" //nolint:revive
 	"go.uber.org/mock/gomock"
+	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/utils/ptr"
+	clusterv1 "sigs.k8s.io/cluster-api/api/v1beta1"
+	"sigs.k8s.io/cluster-api/util/conditions"
 
 	infrav1alpha1 "sigs.k8s.io/cluster-api-provider-openstack/api/v1alpha1"
 	infrav1 "sigs.k8s.io/cluster-api-provider-openstack/api/v1beta1"
@@ -548,3 +551,184 @@ func Test_OpenStackServerReconcileCreate(t *testing.T) {
 		})
 	}
 }
+
+func TestOpenStackServerReconciler_getOrCreateServer(t *testing.T) {
+	tests := []struct {
+		name            string
+		openStackServer *infrav1alpha1.OpenStackServer
+		setupMocks      func(r *recorders)
+		wantServer      *servers.Server
+		wantErr         bool
+		wantCondition   *clusterv1.Condition
+	}{
+		{
+			name: "instanceID set in status but server not found",
+			openStackServer: &infrav1alpha1.OpenStackServer{
+				Status: infrav1alpha1.OpenStackServerStatus{
+					InstanceID: ptr.To(instanceUUID),
+				},
+			},
+			setupMocks: func(r *recorders) {
+				r.compute.GetServer(instanceUUID).Return(nil, gophercloud.ErrUnexpectedResponseCode{Actual: 404})
+			},
+			wantErr: false,
+			wantCondition: &clusterv1.Condition{
+				Type:    infrav1.InstanceReadyCondition,
+				Status:  corev1.ConditionFalse,
+				Reason:  infrav1.InstanceNotFoundReason,
+				Message: infrav1.ServerUnexpectedDeletedMessage,
+			},
+		},
+		{
+			name: "instanceID set in status but server not found with error",
+			openStackServer: &infrav1alpha1.OpenStackServer{
+				Status: infrav1alpha1.OpenStackServerStatus{
+					InstanceID: ptr.To(instanceUUID),
+				},
+			},
+			setupMocks: func(r *recorders) {
+				r.compute.GetServer(instanceUUID).Return(nil, fmt.Errorf("error"))
+			},
+			wantErr: true,
+			wantCondition: &clusterv1.Condition{
+				Type:    infrav1.InstanceReadyCondition,
+				Status:  corev1.ConditionFalse,
+				Reason:  infrav1.OpenStackErrorReason,
+				Message: "get server \"" + instanceUUID + "\" detail failed: error",
+			},
+		},
+		{
+			name: "instanceStatus is nil but server found with machine name",
+			openStackServer: &infrav1alpha1.OpenStackServer{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: openStackServerName,
+				},
+				Status: infrav1alpha1.OpenStackServerStatus{},
+			},
+			setupMocks: func(r *recorders) {
+				r.compute.ListServers(servers.ListOpts{
+					Name: "^" + openStackServerName + "$",
+				}).Return([]servers.Server{{ID: instanceUUID}}, nil)
+			},
+			wantErr: false,
+			wantServer: &servers.Server{
+				ID: instanceUUID,
+			},
+		},
+		{
+			name: "instanceStatus is nil and server not found and then created",
+			openStackServer: &infrav1alpha1.OpenStackServer{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: openStackServerName,
+				},
+				Status: infrav1alpha1.OpenStackServerStatus{
+					Resolved: &infrav1alpha1.ResolvedServerSpec{
+						ImageID:  imageUUID,
+						FlavorID: flavorUUID,
+						Ports:    defaultResolvedPorts,
+					},
+				},
+			},
+			setupMocks: func(r *recorders) {
+				r.compute.ListServers(servers.ListOpts{
+					Name: "^" + openStackServerName + "$",
+				}).Return([]servers.Server{}, nil)
+				r.compute.CreateServer(gomock.Any(), gomock.Any()).Return(&servers.Server{ID: instanceUUID}, nil)
+			},
+			wantErr: false,
+			wantServer: &servers.Server{
+				ID: instanceUUID,
+			},
+			// It's off but no condition is set because the server creation was kicked off but we
+			// don't know the result yet in this function.
+		},
+		{
+			name: "instanceStatus is nil and server not found and then created with error",
+			openStackServer: &infrav1alpha1.OpenStackServer{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: openStackServerName,
+				},
+				Status: infrav1alpha1.OpenStackServerStatus{
+					Resolved: &infrav1alpha1.ResolvedServerSpec{
+						ImageID:  imageUUID,
+						FlavorID: flavorUUID,
+						Ports:    defaultResolvedPorts,
+					},
+				},
+			},
+			setupMocks: func(r *recorders) {
+				r.compute.ListServers(servers.ListOpts{
+					Name: "^" + openStackServerName + "$",
+				}).Return([]servers.Server{}, nil)
+				r.compute.CreateServer(gomock.Any(), gomock.Any()).Return(nil, fmt.Errorf("error"))
+			},
+			wantErr: true,
+			wantCondition: &clusterv1.Condition{
+				Type:    infrav1.InstanceReadyCondition,
+				Status:  corev1.ConditionFalse,
+				Reason:  infrav1.InstanceCreateFailedReason,
+				Message: "error creating Openstack instance: " + "error",
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			g := NewGomegaWithT(t)
+			log := testr.New(t)
+
+			mockCtrl := gomock.NewController(t)
+			defer mockCtrl.Finish()
+
+			mockScopeFactory := scope.NewMockScopeFactory(mockCtrl, "")
+			scopeWithLogger := scope.NewWithLogger(mockScopeFactory, log)
+
+			computeRecorder := mockScopeFactory.ComputeClient.EXPECT()
+			imageRecorder := mockScopeFactory.ImageClient.EXPECT()
+			networkRecorder := mockScopeFactory.NetworkClient.EXPECT()
+			volumeRecorder := mockScopeFactory.VolumeClient.EXPECT()
+
+			recorders := &recorders{
+				compute: computeRecorder,
+				image:   imageRecorder,
+				network: networkRecorder,
+				volume:  volumeRecorder,
+			}
+
+			if tt.setupMocks != nil {
+				tt.setupMocks(recorders)
+			}
+
+			computeService, err := compute.NewService(scopeWithLogger)
+			g.Expect(err).ToNot(HaveOccurred())
+
+			reconciler := OpenStackServerReconciler{}
+			status, err := reconciler.getOrCreateServer(ctx, log, tt.openStackServer, computeService, []string{portUUID})
+
+			// Check error result
+			if tt.wantErr {
+				g.Expect(err).To(HaveOccurred())
+			} else {
+				g.Expect(err).ToNot(HaveOccurred())
+			}
+
+			// Check instance status
+			if tt.wantServer != nil {
+				g.Expect(status.ID()).To(Equal(tt.wantServer.ID))
+			}
+
+			// Check the condition is set correctly
+			if tt.wantCondition != nil {
+				// print openstackServer conditions
+				for _, condition := range tt.openStackServer.Status.Conditions {
+					t.Logf("Condition: %s, Status: %s, Reason: %s", condition.Type, condition.Status, condition.Reason)
+				}
+				conditionType := conditions.Get(tt.openStackServer, tt.wantCondition.Type)
+				g.Expect(conditionType).ToNot(BeNil())
+				g.Expect(conditionType.Status).To(Equal(tt.wantCondition.Status))
+				g.Expect(conditionType.Reason).To(Equal(tt.wantCondition.Reason))
+				g.Expect(conditionType.Message).To(Equal(tt.wantCondition.Message))
+			}
+		})
+	}
+}
diff --git a/test/e2e/data/e2e_conf.yaml b/test/e2e/data/e2e_conf.yaml
@@ -188,6 +188,7 @@ intervals:
   default/wait-control-plane: ["30m", "10s"]
   default/wait-worker-nodes: ["30m", "10s"]
   default/wait-delete-cluster: ["5m", "10s"]
+  default/wait-delete-machine: ["30m", "10s"]
   default/wait-alt-az: ["20m", "30s"]
   default/wait-machine-upgrade: ["30m", "10s"]
   default/wait-nodes-ready: ["15m", "10s"]
diff --git a/test/e2e/shared/openstack.go b/test/e2e/shared/openstack.go
@@ -841,6 +841,21 @@ func CreateOpenStackNetwork(e2eCtx *E2EContext, name, cidr string) (*networks.Ne
 	return net, nil
 }
 
+func DeleteOpenStackServer(ctx context.Context, e2eCtx *E2EContext, id string) error {
+	providerClient, clientOpts, _, err := GetTenantProviderClient(e2eCtx)
+	if err != nil {
+		_, _ = fmt.Fprintf(GinkgoWriter, "error creating provider client: %s\n", err)
+		return err
+	}
+
+	computeClient, err := openstack.NewComputeV2(providerClient, gophercloud.EndpointOpts{Region: clientOpts.RegionName})
+	if err != nil {
+		return fmt.Errorf("error creating compute client: %s", err)
+	}
+
+	return servers.Delete(ctx, computeClient, id).ExtractErr()
+}
+
 func DeleteOpenStackNetwork(ctx context.Context, e2eCtx *E2EContext, id string) error {
 	providerClient, clientOpts, _, err := GetTenantProviderClient(e2eCtx)
 	if err != nil {
diff --git a/test/e2e/suites/e2e/e2e_test.go b/test/e2e/suites/e2e/e2e_test.go
@@ -55,7 +55,10 @@ import (
 	"sigs.k8s.io/cluster-api/test/framework/clusterctl"
 	crclient "sigs.k8s.io/controller-runtime/pkg/client"
 
+	"github.com/k-orc/openstack-resource-controller/pkg/utils/ssa"
+
 	infrav1 "sigs.k8s.io/cluster-api-provider-openstack/api/v1beta1"
+	"sigs.k8s.io/cluster-api-provider-openstack/pkg/generated/applyconfiguration/api/v1beta1"
 	"sigs.k8s.io/cluster-api-provider-openstack/test/e2e/shared"
 )
 
@@ -1006,6 +1009,56 @@ var _ = Describe("e2e tests [PR-Blocking]", func() {
 				Expect(additionalVolume.AvailabilityZone).To(Equal(failureDomain))
 				Expect(additionalVolume.VolumeType).To(Equal(volumeTypeAlt))
 			}
+
+			// This last block tests a scenario where an external agent deletes a server (e.g. a user via Horizon).
+			// We want to ensure that the OpenStackMachine conditions are updated to reflect the server deletion.
+			// Context: https://github.com/kubernetes-sigs/cluster-api-provider-openstack/issues/2474
+			shared.Logf("Deleting a server")
+			serverToDelete := allServers[controlPlaneMachines[0].Spec.InfrastructureRef.Name]
+			err = shared.DeleteOpenStackServer(ctx, e2eCtx, serverToDelete.ID)
+			Expect(err).NotTo(HaveOccurred())
+
+			shared.Logf("Waiting for the OpenStackMachine to have a condition that the server has been unexpectedly deleted")
+			retries := 0
+			Eventually(func() (clusterv1.Condition, error) {
+				k8sClient := e2eCtx.Environment.BootstrapClusterProxy.GetClient()
+
+				openStackMachine := &infrav1.OpenStackMachine{}
+				err := k8sClient.Get(ctx, crclient.ObjectKey{Name: controlPlaneMachines[0].Name, Namespace: controlPlaneMachines[0].Namespace}, openStackMachine)
+				if err != nil {
+					return clusterv1.Condition{}, err
+				}
+				for _, condition := range openStackMachine.Status.Conditions {
+					if condition.Type == infrav1.InstanceReadyCondition {
+						return condition, nil
+					}
+				}
+
+				// Make some non-functional change to the object which will
+				// cause CAPO to reconcile it, otherwise we won't notice the
+				// server is gone until the configured controller-runtime
+				// resync.
+				retries++
+				applyConfig := v1beta1.OpenStackMachine(openStackMachine.Name, openStackMachine.Namespace).
+					WithAnnotations(map[string]string{
+						"e2e-test-retries": fmt.Sprintf("%d", retries),
+					})
+				err = k8sClient.Patch(ctx, openStackMachine, ssa.ApplyConfigPatch(applyConfig), crclient.ForceOwnership, crclient.FieldOwner("capo-e2e"))
+				if err != nil {
+					return clusterv1.Condition{}, err
+				}
+
+				return clusterv1.Condition{}, errors.New("condition InstanceReadyCondition not found")
+			}, time.Minute*3, time.Second*10).Should(MatchFields(
+				IgnoreExtras,
+				Fields{
+					"Type":     Equal(infrav1.InstanceReadyCondition),
+					"Status":   Equal(corev1.ConditionFalse),
+					"Reason":   Equal(infrav1.InstanceDeletedReason),
+					"Message":  Equal(infrav1.ServerUnexpectedDeletedMessage),
+					"Severity": Equal(clusterv1.ConditionSeverityError),
+				},
+			), "OpenStackMachine should be marked not ready with InstanceDeletedReason")
 		})
 	})
 })

Original file line number	Diff line number	Diff line change
`@@ -44,6 +44,9 @@ const (`
`44`	`44`	`OpenStackErrorReason = "OpenStackError"`
`45`	`45`	`// DependencyFailedReason indicates that a dependent object failed.`
`46`	`46`	`DependencyFailedReason = "DependencyFailed"`
	`47`	`+`
	`48`	`+ // ServerUnexpectedDeletedMessage is the message used when the server is unexpectedly deleted via an external agent.`
	`49`	`+ ServerUnexpectedDeletedMessage = "The server was unexpectedly deleted"`
`47`	`50`	`)`
`48`	`51`
`49`	`52`	`const (`