neonvm-controller: add spec.cpuScalingMode, change vm-controller to e…

…xplicitly default the field value if it is not set

Signed-off-by: Misha Sakhnov <[email protected]>

neonvm/apis/neonvm/v1/virtualmachine_types.go

@@ -52,6 +52,14 @@ const (
 	//
 	// The value of this annotation is always a JSON-encoded VirtualMachineResources object.
 	VirtualMachineResourcesAnnotation string = "vm.neon.tech/resources"
+
+	// CpuScalingModeQMP is the value of the VirtualMachineSpec.CpuScalingMode field that indicates
+	// that the VM should use QMP to scale CPUs.
+	CpuScalingModeQMP string = "qmpScaling"
+
+	// CpuScalingModeCpuSysfsState is the value of the VirtualMachineSpec.CpuScalingMode field that
+	// indicates that the VM should use the CPU sysfs state interface to scale CPUs.
+	CpuScalingModeCpuSysfsState string = "cpuSysfsStateScaling"
 )
 
 // VirtualMachineUsage provides information about a VM's current usage. This is the type of the
@@ -150,6 +158,11 @@ type VirtualMachineSpec struct {
 	// propagated to the VM.
 	// +optional
 	TargetRevision *RevisionWithTime `json:"targetRevision,omitempty"`
+
+	// Use QMP scaling mode for CPU or online/offline CPU states.
+	// +kubebuilder:validation:Enum=qmpScaling;cpuSysfsStateScaling
+	// +optional
+	CpuScalingMode *string `json:"cpuScalingMode,omitempty"`
 }
 
 func (spec *VirtualMachineSpec) Resources() VirtualMachineResources {
@@ -605,6 +618,7 @@ func (p VmPhase) IsAlive() bool {
 // +kubebuilder:printcolumn:name="Age",type="date",JSONPath=".metadata.creationTimestamp"
 // +kubebuilder:printcolumn:name="Node",type=string,priority=1,JSONPath=`.status.node`
 // +kubebuilder:printcolumn:name="Image",type=string,priority=1,JSONPath=`.spec.guest.rootDisk.image`
+// +kubebuilder:printcolumn:name="CPUScalingMode",type=string,priority=1,JSONPath=`.spec.cpuScalingMode`
 type VirtualMachine struct {
 	metav1.TypeMeta   `json:",inline"`
 	metav1.ObjectMeta `json:"metadata,omitempty"`

neonvm/config/crd/bases/vm.neon.tech_virtualmachines.yaml

@@ -44,6 +44,10 @@ spec:
       name: Image
       priority: 1
       type: string
+    - jsonPath: .spec.cpuScalingMode
+      name: CPUScalingMode
+      priority: 1
+      type: string
     name: v1
     schema:
       openAPIV3Schema:
@@ -831,6 +835,12 @@ spec:
                         type: array
                     type: object
                 type: object
+              cpuScalingMode:
+                description: Use QMP scaling mode for CPU or online/offline CPU states.
+                enum:
+                - qmpScaling
+                - cpuSysfsStateScaling
+                type: string
               disks:
                 description: List of disk that can be mounted by virtual machine.
                 items:

pkg/neonvm/controllers/vm_controller.go

@@ -165,6 +165,16 @@ func (r *VMReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Re
 		return ctrl.Result{}, nil
 	}
 
+	// examine cpuScalingMode and set it to the default value if it is not set
+	if vm.Spec.CpuScalingMode == nil || *vm.Spec.CpuScalingMode == "" {
+		vm.Spec.CpuScalingMode = lo.ToPtr(vmv1.CpuScalingModeQMP)
+		if err := r.tryUpdateVM(ctx, &vm); err != nil {
+			log.Error(err, "Failed to set default CPU scaling mode")
+			return ctrl.Result{}, err
+		}
+		return ctrl.Result{Requeue: true}, nil
+	}
+
 	statusBefore := vm.Status.DeepCopy()
 	if err := r.doReconcile(ctx, &vm); err != nil {
 		r.Recorder.Eventf(&vm, corev1.EventTypeWarning, "Failed",

pkg/neonvm/controllers/vm_controller_unit_test.go

@@ -166,6 +166,11 @@ func TestReconcile(t *testing.T) {
 	}, res)
 	assert.Contains(t, params.getVM().Finalizers, virtualmachineFinalizer)
 
+	// Added default value for the cpuScalingMode
+	res, err = params.r.Reconcile(params.ctx, req)
+	assert.NoError(t, err)
+	assert.Equal(t, true, res.Requeue)
+
 	// Round 2
 	res, err = params.r.Reconcile(params.ctx, req)
 	assert.NoError(t, err)
@@ -184,8 +189,11 @@ func TestReconcile(t *testing.T) {
 	// We now have a pod
 	vm := params.getVM()
 	assert.NotEmpty(t, vm.Status.PodName)
-	// Spec is unchanged
-	assert.Equal(t, vm.Spec, origVM.Spec)
+	// Spec is unchanged except cpuScalingMode
+	var origVMWithCPUScalingMode vmv1.VirtualMachine
+	origVM.DeepCopy().DeepCopyInto(&origVMWithCPUScalingMode)
+	origVMWithCPUScalingMode.Spec.CpuScalingMode = lo.ToPtr(vmv1.CpuScalingModeQMP)
+	assert.Equal(t, vm.Spec, origVMWithCPUScalingMode.Spec)
 
 	// Round 4
 	res, err = params.r.Reconcile(params.ctx, req)
@@ -216,8 +224,13 @@ func TestRunningPod(t *testing.T) {
 	// Round 1
 	params.mockRecorder.On("Event", mock.Anything, "Normal", "Created",
 		mock.Anything)
+	// Requeue because we expect default values to be set
 	res, err := params.r.Reconcile(params.ctx, req)
 	require.NoError(t, err)
+	assert.Equal(t, true, res.Requeue)
+
+	res, err = params.r.Reconcile(params.ctx, req)
+	require.NoError(t, err)
 	assert.Equal(t, false, res.Requeue)
 
 	// We now have a pod

tests/e2e/tmp-default-cpu-scaling-mode/00-assert.yaml

@@ -0,0 +1,13 @@
+apiVersion: kuttl.dev/v1beta1
+kind: TestAssert
+timeout: 90
+---
+apiVersion: vm.neon.tech/v1
+kind: VirtualMachine
+metadata:
+  name: example
+# temporary added this check to validate that defaulting cpuScalingMode works as expected
+# field spec.cpuScalingMode should go to default value `qmp_scaling` if it is not set
+# TODO: delete once the https://github.com/neondatabase/autoscaling/issues/1082 went live
+spec:
+  cpuScalingMode: qmpScaling

tests/e2e/tmp-default-cpu-scaling-mode/00-create-vm.yaml

@@ -0,0 +1,23 @@
+apiVersion: kuttl.dev/v1beta1
+kind: TestStep
+unitTest: false
+---
+apiVersion: vm.neon.tech/v1
+kind: VirtualMachine
+metadata:
+  name: example
+spec:
+  schedulerName: autoscale-scheduler
+  guest:
+    cpus:
+      min: 0.25
+      use: 0.25
+      max: 0.25
+    memorySlotSize: 1Gi
+    memorySlots:
+      min: 1
+      use: 1
+      max: 1
+    rootDisk:
+      image: vm-postgres:15-bullseye
+      size: 1Gi