Support toggling provisioning checks

Signed-off-by: ekarlso <endre.karlson@gmail.com>
ionos-cloud · Sep 30, 2024 · 5299af1 · 5299af1
1 parent 321d629
commit 5299af1
Show file tree

Hide file tree

Showing 11 changed files with 226 additions and 9 deletions.
diff --git a/api/v1alpha1/proxmoxmachine_types.go b/api/v1alpha1/proxmoxmachine_types.go
@@ -50,6 +50,15 @@ const (
 	IPV6Format = "v6"
 )
 
+type ProxmoxMachineChecks struct {
+	// Skip checking CloudInit which can be very useful for specific Operating Systems like TalOS
+	// +optional
+	SkipCloudInitStatus *bool `json:"skipCloudInitStatus,omitempty"`
+	// Skip checking QEMU Agent readiness which can be very useful for specific Operating Systems like TalOS
+	// +optional
+	SkipQemuGuestAgent *bool `json:"skipQemuGuestAgent,omitempty"`
+}
+
 // ProxmoxMachineSpec defines the desired state of a ProxmoxMachine.
 type ProxmoxMachineSpec struct {
 	VirtualMachineCloneSpec `json:",inline"`
@@ -90,6 +99,8 @@ type ProxmoxMachineSpec struct {
 	// Network is the network configuration for this machine's VM.
 	// +optional
 	Network *NetworkSpec `json:"network,omitempty"`
+
+	Checks *ProxmoxMachineChecks `json:"checks,omitempty"`
 }
 
 // Storage is the physical storage on the node.

diff --git a/api/v1alpha1/zz_generated.deepcopy.go b/api/v1alpha1/zz_generated.deepcopy.go
diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxclusters.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxclusters.yaml
@@ -72,6 +72,15 @@ spec:
                       description: ProxmoxMachineSpec defines the desired state of
                         a ProxmoxMachine.
                       properties:
+                        checks:
+                          properties:
+                            skipCloudInitStatus:
+                              description: Skip checking CloudInit which can be very
+                                useful for specific Operating Systems like TalOS
+                              type: boolean
+                            skipQemuGuestAgent:
+                              type: boolean
+                          type: object
                         description:
                           description: Description for the new VM.
                           type: string

diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxclustertemplates.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxclustertemplates.yaml
@@ -90,6 +90,16 @@ spec:
                               description: ProxmoxMachineSpec defines the desired
                                 state of a ProxmoxMachine.
                               properties:
+                                checks:
+                                  properties:
+                                    skipCloudInitStatus:
+                                      description: Skip checking CloudInit which can
+                                        be very useful for specific Operating Systems
+                                        like TalOS
+                                      type: boolean
+                                    skipQemuGuestAgent:
+                                      type: boolean
+                                  type: object
                                 description:
                                   description: Description for the new VM.
                                   type: string

diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxmachines.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxmachines.yaml
@@ -65,6 +65,15 @@ spec:
           spec:
             description: ProxmoxMachineSpec defines the desired state of a ProxmoxMachine.
             properties:
+              checks:
+                properties:
+                  skipCloudInitStatus:
+                    description: Skip checking CloudInit which can be very useful
+                      for specific Operating Systems like TalOS
+                    type: boolean
+                  skipQemuGuestAgent:
+                    type: boolean
+                type: object
               description:
                 description: Description for the new VM.
                 type: string

diff --git a/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxmachinetemplates.yaml b/config/crd/bases/infrastructure.cluster.x-k8s.io_proxmoxmachinetemplates.yaml
@@ -72,6 +72,15 @@ spec:
                     description: ProxmoxMachineSpec defines the desired state of a
                       ProxmoxMachine.
                     properties:
+                      checks:
+                        properties:
+                          skipCloudInitStatus:
+                            description: Skip checking CloudInit which can be very
+                              useful for specific Operating Systems like TalOS
+                            type: boolean
+                          skipQemuGuestAgent:
+                            type: boolean
+                        type: object
                       description:
                         description: Description for the new VM.
                         type: string

diff --git a/internal/service/vmservice/vm.go b/internal/service/vmservice/vm.go
@@ -108,22 +108,46 @@ func ReconcileVM(ctx context.Context, scope *scope.MachineScope) (infrav1alpha1.
 	return vm, nil
 }
 
+func skipQemuGuestCheck(mc *scope.MachineScope) bool {
+	if mc.ProxmoxMachine.Spec.Checks != nil {
+		return ptr.Deref(mc.ProxmoxMachine.Spec.Checks.SkipQemuGuestAgent, false)
+	}
+
+	return false
+}
+
+func skipCloudInitCheck(mc *scope.MachineScope) bool {
+	if mc.ProxmoxMachine.Spec.Checks != nil {
+		return ptr.Deref(mc.ProxmoxMachine.Spec.Checks.SkipCloudInitStatus, false)
+	}
+
+	return false
+}
+
 func checkCloudInitStatus(ctx context.Context, machineScope *scope.MachineScope) (requeue bool, err error) {
 	if !machineScope.VirtualMachine.IsRunning() {
 		// skip if the vm is not running.
 		return true, nil
 	}
 
-	if running, err := machineScope.InfraCluster.ProxmoxClient.CloudInitStatus(ctx, machineScope.VirtualMachine); err != nil || running {
-		if running {
-			return true, nil
+	if !skipQemuGuestCheck(machineScope) {
+		if err := machineScope.InfraCluster.ProxmoxClient.QemuAgentStatus(ctx, machineScope.VirtualMachine); err != nil {
+			return true, errors.Wrap(err, "error waiting for agent")
 		}
-		if errors.Is(goproxmox.ErrCloudInitFailed, err) {
-			conditions.MarkFalse(machineScope.ProxmoxMachine, infrav1alpha1.VMProvisionedCondition, infrav1alpha1.VMProvisionFailedReason, clusterv1.ConditionSeverityError, err.Error())
-			machineScope.SetFailureMessage(err)
-			machineScope.SetFailureReason(capierrors.MachineStatusError("BootstrapFailed"))
+	}
+
+	if !skipCloudInitCheck(machineScope) {
+		if running, err := machineScope.InfraCluster.ProxmoxClient.CloudInitStatus(ctx, machineScope.VirtualMachine); err != nil || running {
+			if running {
+				return true, nil
+			}
+			if errors.Is(goproxmox.ErrCloudInitFailed, err) {
+				conditions.MarkFalse(machineScope.ProxmoxMachine, infrav1alpha1.VMProvisionedCondition, infrav1alpha1.VMProvisionFailedReason, clusterv1.ConditionSeverityError, err.Error())
+				machineScope.SetFailureMessage(err)
+				machineScope.SetFailureReason(capierrors.MachineStatusError("BootstrapFailed"))
+			}
+			return false, err
 		}
-		return false, err
 	}
 
 	return false, nil

diff --git a/internal/service/vmservice/vm_test.go b/internal/service/vmservice/vm_test.go
@@ -42,6 +42,67 @@ func TestReconcileVM_EverythingReady(t *testing.T) {
 
 	proxmoxClient.EXPECT().GetVM(context.Background(), "node1", int64(123)).Return(vm, nil).Once()
 	proxmoxClient.EXPECT().CloudInitStatus(context.Background(), vm).Return(false, nil).Once()
+	proxmoxClient.EXPECT().QemuAgentStatus(context.Background(), vm).Return(nil).Once()
+
+	result, err := ReconcileVM(context.Background(), machineScope)
+	require.NoError(t, err)
+	require.Equal(t, infrav1alpha1.VirtualMachineStateReady, result.State)
+	require.Equal(t, "10.10.10.10", machineScope.ProxmoxMachine.Status.Addresses[1].Address)
+}
+
+func TestReconcileVM_QemuAgentCheckDisabled(t *testing.T) {
+	machineScope, proxmoxClient, _ := setupReconcilerTest(t)
+	vm := newRunningVM()
+	machineScope.SetVirtualMachineID(int64(vm.VMID))
+	machineScope.ProxmoxMachine.Status.IPAddresses = map[string]infrav1alpha1.IPAddress{infrav1alpha1.DefaultNetworkDevice: {IPV4: "10.10.10.10"}}
+	machineScope.ProxmoxMachine.Status.BootstrapDataProvided = ptr.To(true)
+	machineScope.ProxmoxMachine.Status.Ready = true
+	machineScope.ProxmoxMachine.Spec.Checks = &infrav1alpha1.ProxmoxMachineChecks{
+		SkipQemuGuestAgent: ptr.To(true),
+	}
+
+	proxmoxClient.EXPECT().GetVM(context.Background(), "node1", int64(123)).Return(vm, nil).Once()
+	proxmoxClient.EXPECT().CloudInitStatus(context.Background(), vm).Return(false, nil).Once()
+
+	result, err := ReconcileVM(context.Background(), machineScope)
+	require.NoError(t, err)
+	require.Equal(t, infrav1alpha1.VirtualMachineStateReady, result.State)
+	require.Equal(t, "10.10.10.10", machineScope.ProxmoxMachine.Status.Addresses[1].Address)
+}
+
+func TestReconcileVM_CloudInitCheckDisabled(t *testing.T) {
+	machineScope, proxmoxClient, _ := setupReconcilerTest(t)
+	vm := newRunningVM()
+	machineScope.SetVirtualMachineID(int64(vm.VMID))
+	machineScope.ProxmoxMachine.Status.IPAddresses = map[string]infrav1alpha1.IPAddress{infrav1alpha1.DefaultNetworkDevice: {IPV4: "10.10.10.10"}}
+	machineScope.ProxmoxMachine.Status.BootstrapDataProvided = ptr.To(true)
+	machineScope.ProxmoxMachine.Status.Ready = true
+	machineScope.ProxmoxMachine.Spec.Checks = &infrav1alpha1.ProxmoxMachineChecks{
+		SkipCloudInitStatus: ptr.To(true),
+	}
+
+	proxmoxClient.EXPECT().GetVM(context.Background(), "node1", int64(123)).Return(vm, nil).Once()
+	proxmoxClient.EXPECT().QemuAgentStatus(context.Background(), vm).Return(nil)
+
+	result, err := ReconcileVM(context.Background(), machineScope)
+	require.NoError(t, err)
+	require.Equal(t, infrav1alpha1.VirtualMachineStateReady, result.State)
+	require.Equal(t, "10.10.10.10", machineScope.ProxmoxMachine.Status.Addresses[1].Address)
+}
+
+func TestReconcileVM_InitCheckDisabled(t *testing.T) {
+	machineScope, proxmoxClient, _ := setupReconcilerTest(t)
+	vm := newRunningVM()
+	machineScope.SetVirtualMachineID(int64(vm.VMID))
+	machineScope.ProxmoxMachine.Status.IPAddresses = map[string]infrav1alpha1.IPAddress{infrav1alpha1.DefaultNetworkDevice: {IPV4: "10.10.10.10"}}
+	machineScope.ProxmoxMachine.Status.BootstrapDataProvided = ptr.To(true)
+	machineScope.ProxmoxMachine.Status.Ready = true
+	machineScope.ProxmoxMachine.Spec.Checks = &infrav1alpha1.ProxmoxMachineChecks{
+		SkipCloudInitStatus: ptr.To(true),
+		SkipQemuGuestAgent:  ptr.To(true),
+	}
+
+	proxmoxClient.EXPECT().GetVM(context.Background(), "node1", int64(123)).Return(vm, nil).Once()
 
 	result, err := ReconcileVM(context.Background(), machineScope)
 	require.NoError(t, err)
@@ -323,6 +384,7 @@ func TestReconcileVM_CloudInitFailed(t *testing.T) {
 
 	proxmoxClient.EXPECT().GetVM(context.Background(), "node1", int64(123)).Return(vm, nil).Once()
 	proxmoxClient.EXPECT().CloudInitStatus(context.Background(), vm).Return(false, goproxmox.ErrCloudInitFailed).Once()
+	proxmoxClient.EXPECT().QemuAgentStatus(context.Background(), vm).Return(nil).Once()
 
 	_, err := ReconcileVM(context.Background(), machineScope)
 	require.Error(t, err, "unknown error")
@@ -340,6 +402,7 @@ func TestReconcileVM_CloudInitRunning(t *testing.T) {
 
 	proxmoxClient.EXPECT().GetVM(context.Background(), "node1", int64(123)).Return(vm, nil).Once()
 	proxmoxClient.EXPECT().CloudInitStatus(context.Background(), vm).Return(true, nil).Once()
+	proxmoxClient.EXPECT().QemuAgentStatus(context.Background(), vm).Return(nil).Once()
 
 	result, err := ReconcileVM(context.Background(), machineScope)
 	require.NoError(t, err)

diff --git a/pkg/proxmox/client.go b/pkg/proxmox/client.go
@@ -48,6 +48,7 @@ type Client interface {
 	TagVM(ctx context.Context, vm *proxmox.VirtualMachine, tag string) (*proxmox.Task, error)
 
 	UnmountCloudInitISO(ctx context.Context, vm *proxmox.VirtualMachine, device string) error
-
 	CloudInitStatus(ctx context.Context, vm *proxmox.VirtualMachine) (bool, error)
+
+	QemuAgentStatus(ctx context.Context, vm *proxmox.VirtualMachine) error
 }
diff --git a/pkg/proxmox/goproxmox/api_client.go b/pkg/proxmox/goproxmox/api_client.go
@@ -300,3 +300,11 @@ func (c *APIClient) CloudInitStatus(ctx context.Context, vm *proxmox.VirtualMach
 
 	return false, nil
 }
+
+func (c *APIClient) QemuAgentStatus(ctx context.Context, vm *proxmox.VirtualMachine) error {
+	if err := vm.WaitForAgent(ctx, 5); err != nil {
+		return errors.Wrap(err, "error waiting for agent")
+	}
+
+	return nil
+}
diff --git a/pkg/proxmox/proxmoxtest/mock_client.go b/pkg/proxmox/proxmoxtest/mock_client.go