GT-134 Add annotation to change architecture (#1107)

jwierzbo · ajanikow · commit effed19d538a · 2022-10-05T07:54:51.000Z
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -3,6 +3,7 @@
 ## [master](https://github.com/arangodb/kube-arangodb/tree/master) (N/A)
 - (Feature) Add action progress
 - (Feature) Ensure consistency during replication cancellation
+- (Feature) Add annotation to change architecture of a member
 
 ## [1.2.19](https://github.com/arangodb/kube-arangodb/tree/1.2.19) (2022-10-05)
 - (Bugfix) Prevent changes when UID is wrong
diff --git a/docs/design/arch_change.md b/docs/design/arch_change.md
@@ -0,0 +1,20 @@
+# Member Architecture change
+
+To change manually architecture of specific member, you can use annotation:
+```bash
+kubectl annotate pod arango-pod deployment.arangodb.com/arch=arm64
+```
+
+It will add to the member status `ArchitectureMismatch` condition, e.g.:
+```yaml
+  - lastTransitionTime: "2022-09-15T07:38:05Z"
+    lastUpdateTime: "2022-09-15T07:38:05Z"
+    reason: Member has a different architecture than the deployment
+    status: "True"
+    type: ArchitectureMismatch
+```
+
+To provide requested arch changes for the member we need rotate it, so additional step is required:
+```bash
+`kubectl annotate pod arango-pod deployment.arangodb.com/rotate=true`
+```
diff --git a/docs/generated/actions.md b/docs/generated/actions.md
@@ -60,6 +60,7 @@
 |            SetCondition            |  10m0s  | Community & Enterprise |                                       (Deprecated) Set deployment condition                                        |
 |           SetConditionV2           |  10m0s  | Community & Enterprise |                                              Set deployment condition                                              |
 |          SetCurrentImage           | 6h0m0s  | Community & Enterprise |                               Update deployment current image after image discovery                                |
+|        SetCurrentMemberArch        |  10m0s  | Community & Enterprise |                                          Set current member architecture                                           |
 |      SetMaintenanceCondition       |  10m0s  | Community & Enterprise |                                       Update ArangoDB maintenance condition                                        |
 |         SetMemberCondition         |  10m0s  | Community & Enterprise |                                         (Deprecated) Set member condition                                          |
 |        SetMemberConditionV2        |  10m0s  | Community & Enterprise |                                                Set member condition                                                |
@@ -141,6 +142,7 @@ spec:
       SetCondition: 10m0s
       SetConditionV2: 10m0s
       SetCurrentImage: 6h0m0s
+      SetCurrentMemberArch: 10m0s
       SetMaintenanceCondition: 10m0s
       SetMemberCondition: 10m0s
       SetMemberConditionV2: 10m0s
diff --git a/internal/actions.yaml b/internal/actions.yaml
@@ -84,6 +84,9 @@ actions:
     enterprise: true
     description: Update certificate in SNI
     timeout: 10m
+  SetCurrentMemberArch:
+    description: Set current member architecture
+    timeout: 10m
   SetCurrentImage:
     description: Update deployment current image after image discovery
     timeout: 6h
diff --git a/pkg/apis/deployment/annotations.go b/pkg/apis/deployment/annotations.go
@@ -23,6 +23,7 @@ package deployment
 const (
 	ArangoDeploymentAnnotationPrefix                = "deployment.arangodb.com"
 	ArangoDeploymentPodMaintenanceAnnotation        = ArangoDeploymentAnnotationPrefix + "/maintenance"
+	ArangoDeploymentPodChangeArchAnnotation         = ArangoDeploymentAnnotationPrefix + "/arch"
 	ArangoDeploymentPodRotateAnnotation             = ArangoDeploymentAnnotationPrefix + "/rotate"
 	ArangoDeploymentPodReplaceAnnotation            = ArangoDeploymentAnnotationPrefix + "/replace"
 	ArangoDeploymentPodDeleteNow                    = ArangoDeploymentAnnotationPrefix + "/delete_now"
diff --git a/pkg/apis/deployment/v1/actions.generated.go b/pkg/apis/deployment/v1/actions.generated.go
@@ -137,6 +137,8 @@ const (
 	ActionSetConditionV2DefaultTimeout time.Duration = ActionsDefaultTimeout
 	// ActionSetCurrentImageDefaultTimeout define default timeout for action ActionSetCurrentImage
 	ActionSetCurrentImageDefaultTimeout time.Duration = 21600 * time.Second // 6h0m0s
+	// ActionSetCurrentMemberArchDefaultTimeout define default timeout for action ActionSetCurrentMemberArch
+	ActionSetCurrentMemberArchDefaultTimeout time.Duration = 600 * time.Second // 10m0s
 	// ActionSetMaintenanceConditionDefaultTimeout define default timeout for action ActionSetMaintenanceCondition
 	ActionSetMaintenanceConditionDefaultTimeout time.Duration = ActionsDefaultTimeout
 	// ActionSetMemberConditionDefaultTimeout define default timeout for action ActionSetMemberCondition
@@ -286,6 +288,8 @@ const (
 	ActionTypeSetConditionV2 ActionType = "SetConditionV2"
 	// ActionTypeSetCurrentImage in scopes Normal. Update deployment current image after image discovery
 	ActionTypeSetCurrentImage ActionType = "SetCurrentImage"
+	// ActionTypeSetCurrentMemberArch in scopes Normal. Set current member architecture
+	ActionTypeSetCurrentMemberArch ActionType = "SetCurrentMemberArch"
 	// ActionTypeSetMaintenanceCondition in scopes Normal. Update ArangoDB maintenance condition
 	ActionTypeSetMaintenanceCondition ActionType = "SetMaintenanceCondition"
 	// ActionTypeSetMemberCondition in scopes High. (Deprecated) Set member condition
@@ -436,6 +440,8 @@ func ActionDefaultTimeout(in ActionType) time.Duration {
 		return ActionSetConditionV2DefaultTimeout
 	case ActionTypeSetCurrentImage:
 		return ActionSetCurrentImageDefaultTimeout
+	case ActionTypeSetCurrentMemberArch:
+		return ActionSetCurrentMemberArchDefaultTimeout
 	case ActionTypeSetMaintenanceCondition:
 		return ActionSetMaintenanceConditionDefaultTimeout
 	case ActionTypeSetMemberCondition:
@@ -589,6 +595,8 @@ func GetActionPriority(in ActionType) ActionPriority {
 		return ActionPriorityHigh
 	case ActionTypeSetCurrentImage:
 		return ActionPriorityNormal
+	case ActionTypeSetCurrentMemberArch:
+		return ActionPriorityNormal
 	case ActionTypeSetMaintenanceCondition:
 		return ActionPriorityNormal
 	case ActionTypeSetMemberCondition:
diff --git a/pkg/apis/deployment/v1/architecture.go b/pkg/apis/deployment/v1/architecture.go
@@ -49,6 +49,16 @@ func (a ArangoDeploymentArchitecture) Validate() error {
 	return nil
 }
 
+func (a ArangoDeploymentArchitecture) IsArchAllowed(arch ArangoDeploymentArchitectureType) bool {
+	for id := range a {
+		if a[id] == arch {
+			return true
+		}
+	}
+
+	return false
+}
+
 type ArangoDeploymentArchitectureType string
 
 const (
@@ -93,7 +103,22 @@ func (a ArangoDeploymentArchitectureType) AsNodeSelectorRequirement() core.NodeS
 	}
 }
 
-func GetArchsFromNodeSelector(selectors []core.NodeSelectorTerm) map[ArangoDeploymentArchitectureType]bool {
+func (a ArangoDeploymentArchitectureType) IsArchMismatch(deploymentArch ArangoDeploymentArchitecture, memberArch *ArangoDeploymentArchitectureType) bool {
+	if a.Validate() != nil {
+		return false
+	}
+
+	if deploymentArch.IsArchAllowed(a) {
+		if memberArch == nil {
+			return true
+		} else if a != *memberArch {
+			return true
+		}
+	}
+	return false
+}
+
+func GetAllArchFromNodeSelector(selectors []core.NodeSelectorTerm) map[ArangoDeploymentArchitectureType]bool {
 	result := make(map[ArangoDeploymentArchitectureType]bool)
 	for _, selector := range selectors {
 		if selector.MatchExpressions != nil {
diff --git a/pkg/apis/deployment/v1/conditions.go b/pkg/apis/deployment/v1/conditions.go
@@ -74,6 +74,8 @@ const (
 	ConditionTypeScaleDownCandidate ConditionType = "ScaleDownCandidate"
 	// ConditionTypeUpgradeFailed indicates that upgrade failed
 	ConditionTypeUpgradeFailed ConditionType = "UpgradeFailed"
+	// ConditionTypeArchitectureMismatch indicates that the member has a different architecture than the deployment.
+	ConditionTypeArchitectureMismatch ConditionType = "ArchitectureMismatch"
 
 	// ConditionTypeMemberMaintenanceMode indicates that Maintenance is enabled on particular member
 	ConditionTypeMemberMaintenanceMode ConditionType = "MemberMaintenanceMode"
diff --git a/pkg/apis/deployment/v1/plan.go b/pkg/apis/deployment/v1/plan.go
@@ -97,6 +97,8 @@ type Action struct {
 	Locals PlanLocals `json:"locals,omitempty"`
 	// ID reference of the task involved in this action (if any)
 	TaskID types.UID `json:"taskID,omitempty"`
+	// Architecture of the member involved in this action (if any)
+	Architecture ArangoDeploymentArchitectureType `json:"arch,omitempty"`
 	// Progress describes what is a status of the current action.
 	Progress string `json:"progress,omitempty"`
 }
@@ -115,6 +117,7 @@ func (a Action) Equal(other Action) bool {
 		equality.Semantic.DeepEqual(a.Params, other.Params) &&
 		a.Locals.Equal(other.Locals) &&
 		a.TaskID == other.TaskID &&
+		a.Architecture == other.Architecture &&
 		a.Progress == other.Progress
 }
 
@@ -194,6 +197,12 @@ func (a Action) SetImage(image string) Action {
 	return a
 }
 
+// SetArch sets the Architecture field to the given value and returns the modified
+func (a Action) SetArch(arch ArangoDeploymentArchitectureType) Action {
+	a.Architecture = arch
+	return a
+}
+
 // IsStarted returns true if the action has been started already.
 func (a Action) IsStarted() bool {
 	return !a.StartTime.IsZero()
diff --git a/pkg/apis/deployment/v2alpha1/actions.generated.go b/pkg/apis/deployment/v2alpha1/actions.generated.go
@@ -137,6 +137,8 @@ const (
 	ActionSetConditionV2DefaultTimeout time.Duration = ActionsDefaultTimeout
 	// ActionSetCurrentImageDefaultTimeout define default timeout for action ActionSetCurrentImage
 	ActionSetCurrentImageDefaultTimeout time.Duration = 21600 * time.Second // 6h0m0s
+	// ActionSetCurrentMemberArchDefaultTimeout define default timeout for action ActionSetCurrentMemberArch
+	ActionSetCurrentMemberArchDefaultTimeout time.Duration = 600 * time.Second // 10m0s
 	// ActionSetMaintenanceConditionDefaultTimeout define default timeout for action ActionSetMaintenanceCondition
 	ActionSetMaintenanceConditionDefaultTimeout time.Duration = ActionsDefaultTimeout
 	// ActionSetMemberConditionDefaultTimeout define default timeout for action ActionSetMemberCondition
@@ -286,6 +288,8 @@ const (
 	ActionTypeSetConditionV2 ActionType = "SetConditionV2"
 	// ActionTypeSetCurrentImage in scopes Normal. Update deployment current image after image discovery
 	ActionTypeSetCurrentImage ActionType = "SetCurrentImage"
+	// ActionTypeSetCurrentMemberArch in scopes Normal. Set current member architecture
+	ActionTypeSetCurrentMemberArch ActionType = "SetCurrentMemberArch"
 	// ActionTypeSetMaintenanceCondition in scopes Normal. Update ArangoDB maintenance condition
 	ActionTypeSetMaintenanceCondition ActionType = "SetMaintenanceCondition"
 	// ActionTypeSetMemberCondition in scopes High. (Deprecated) Set member condition
@@ -436,6 +440,8 @@ func ActionDefaultTimeout(in ActionType) time.Duration {
 		return ActionSetConditionV2DefaultTimeout
 	case ActionTypeSetCurrentImage:
 		return ActionSetCurrentImageDefaultTimeout
+	case ActionTypeSetCurrentMemberArch:
+		return ActionSetCurrentMemberArchDefaultTimeout
 	case ActionTypeSetMaintenanceCondition:
 		return ActionSetMaintenanceConditionDefaultTimeout
 	case ActionTypeSetMemberCondition:
@@ -589,6 +595,8 @@ func GetActionPriority(in ActionType) ActionPriority {
 		return ActionPriorityHigh
 	case ActionTypeSetCurrentImage:
 		return ActionPriorityNormal
+	case ActionTypeSetCurrentMemberArch:
+		return ActionPriorityNormal
 	case ActionTypeSetMaintenanceCondition:
 		return ActionPriorityNormal
 	case ActionTypeSetMemberCondition:
diff --git a/pkg/apis/deployment/v2alpha1/architecture.go b/pkg/apis/deployment/v2alpha1/architecture.go
@@ -49,6 +49,16 @@ func (a ArangoDeploymentArchitecture) Validate() error {
 	return nil
 }
 
+func (a ArangoDeploymentArchitecture) IsArchAllowed(arch ArangoDeploymentArchitectureType) bool {
+	for id := range a {
+		if a[id] == arch {
+			return true
+		}
+	}
+
+	return false
+}
+
 type ArangoDeploymentArchitectureType string
 
 const (
@@ -93,7 +103,22 @@ func (a ArangoDeploymentArchitectureType) AsNodeSelectorRequirement() core.NodeS
 	}
 }
 
-func GetArchsFromNodeSelector(selectors []core.NodeSelectorTerm) map[ArangoDeploymentArchitectureType]bool {
+func (a ArangoDeploymentArchitectureType) IsArchMismatch(deploymentArch ArangoDeploymentArchitecture, memberArch *ArangoDeploymentArchitectureType) bool {
+	if a.Validate() != nil {
+		return false
+	}
+
+	if deploymentArch.IsArchAllowed(a) {
+		if memberArch == nil {
+			return true
+		} else if a != *memberArch {
+			return true
+		}
+	}
+	return false
+}
+
+func GetAllArchFromNodeSelector(selectors []core.NodeSelectorTerm) map[ArangoDeploymentArchitectureType]bool {
 	result := make(map[ArangoDeploymentArchitectureType]bool)
 	for _, selector := range selectors {
 		if selector.MatchExpressions != nil {
diff --git a/pkg/apis/deployment/v2alpha1/conditions.go b/pkg/apis/deployment/v2alpha1/conditions.go
@@ -74,6 +74,8 @@ const (
 	ConditionTypeScaleDownCandidate ConditionType = "ScaleDownCandidate"
 	// ConditionTypeUpgradeFailed indicates that upgrade failed
 	ConditionTypeUpgradeFailed ConditionType = "UpgradeFailed"
+	// ConditionTypeArchitectureMismatch indicates that the member has a different architecture than the deployment.
+	ConditionTypeArchitectureMismatch ConditionType = "ArchitectureMismatch"
 
 	// ConditionTypeMemberMaintenanceMode indicates that Maintenance is enabled on particular member
 	ConditionTypeMemberMaintenanceMode ConditionType = "MemberMaintenanceMode"
diff --git a/pkg/apis/deployment/v2alpha1/plan.go b/pkg/apis/deployment/v2alpha1/plan.go
@@ -97,6 +97,8 @@ type Action struct {
 	Locals PlanLocals `json:"locals,omitempty"`
 	// ID reference of the task involved in this action (if any)
 	TaskID types.UID `json:"taskID,omitempty"`
+	// Architecture of the member involved in this action (if any)
+	Architecture ArangoDeploymentArchitectureType `json:"arch,omitempty"`
 	// Progress describes what is a status of the current action.
 	Progress string `json:"progress,omitempty"`
 }
@@ -115,6 +117,7 @@ func (a Action) Equal(other Action) bool {
 		equality.Semantic.DeepEqual(a.Params, other.Params) &&
 		a.Locals.Equal(other.Locals) &&
 		a.TaskID == other.TaskID &&
+		a.Architecture == other.Architecture &&
 		a.Progress == other.Progress
 }
 
@@ -194,6 +197,12 @@ func (a Action) SetImage(image string) Action {
 	return a
 }
 
+// SetArch sets the Architecture field to the given value and returns the modified
+func (a Action) SetArch(arch ArangoDeploymentArchitectureType) Action {
+	a.Architecture = arch
+	return a
+}
+
 // IsStarted returns true if the action has been started already.
 func (a Action) IsStarted() bool {
 	return !a.StartTime.IsZero()
diff --git a/pkg/deployment/member/phase_updates.go b/pkg/deployment/member/phase_updates.go
@@ -101,6 +101,7 @@ func removeMemberConditionsMapFunc(m *api.MemberStatus) {
 	m.Conditions.Remove(api.ConditionTypeTopologyAware)
 	m.Conditions.Remove(api.MemberReplacementRequired)
 	m.Conditions.Remove(api.ConditionTypePVCResizePending)
+	m.Conditions.Remove(api.ConditionTypeArchitectureMismatch)
 
 	m.RemoveTerminationsBefore(time.Now().Add(-1 * recentTerminationsKeepPeriod))
 
diff --git a/pkg/deployment/pod/affinity.go b/pkg/deployment/pod/affinity.go
@@ -60,6 +60,27 @@ func AppendArchSelector(a *core.NodeAffinity, arch api.ArangoDeploymentArchitect
 	a.RequiredDuringSchedulingIgnoredDuringExecution.NodeSelectorTerms = append(a.RequiredDuringSchedulingIgnoredDuringExecution.NodeSelectorTerms, arch.AsNodeSelectorRequirement())
 }
 
+func GetArchFromAffinity(a *core.Affinity) api.ArangoDeploymentArchitectureType {
+	if a != nil && a.NodeAffinity != nil && a.NodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution != nil {
+		for _, nst := range a.NodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution.NodeSelectorTerms {
+			for _, req := range nst.MatchExpressions {
+				if req.Key == shared.NodeArchAffinityLabel || req.Key == shared.NodeArchAffinityLabelBeta {
+					for _, arch := range req.Values {
+						return api.ArangoDeploymentArchitectureType(arch)
+					}
+				}
+			}
+		}
+	}
+	return ""
+}
+
+func SetArchInAffinity(a *core.Affinity, arch api.ArangoDeploymentArchitectureType) {
+	if a != nil && a.NodeAffinity != nil && a.NodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution != nil {
+		a.NodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution.NodeSelectorTerms = []core.NodeSelectorTerm{arch.AsNodeSelectorRequirement()}
+	}
+}
+
 func AppendAffinityWithRole(p interfaces.PodCreator, a *core.PodAffinity, role string) {
 	labelSelector := &meta.LabelSelector{
 		MatchLabels: k8sutil.LabelsForDeployment(p.GetName(), role),
diff --git a/pkg/deployment/reconcile/action.register.generated.go b/pkg/deployment/reconcile/action.register.generated.go
@@ -195,6 +195,9 @@ var (
 	_ Action        = &actionSetCurrentImage{}
 	_ actionFactory = newSetCurrentImageAction
 
+	_ Action        = &actionSetCurrentMemberArch{}
+	_ actionFactory = newSetCurrentMemberArchAction
+
 	_ Action        = &actionSetMaintenanceCondition{}
 	_ actionFactory = newSetMaintenanceConditionAction
 
@@ -924,6 +927,18 @@ func init() {
 		registerAction(action, function)
 	}
 
+	// SetCurrentMemberArch
+	{
+		// Get Action defition
+		function := newSetCurrentMemberArchAction
+		action := api.ActionTypeSetCurrentMemberArch
+
+		// Wrap action main function
+
+		// Register action
+		registerAction(action, function)
+	}
+
 	// SetMaintenanceCondition
 	{
 		// Get Action defition
diff --git a/pkg/deployment/reconcile/action.register.generated_test.go b/pkg/deployment/reconcile/action.register.generated_test.go
@@ -256,6 +256,10 @@ func Test_Actions(t *testing.T) {
 		ActionsExistence(t, api.ActionTypeSetCurrentImage)
 	})
 
+	t.Run("SetCurrentMemberArch", func(t *testing.T) {
+		ActionsExistence(t, api.ActionTypeSetCurrentMemberArch)
+	})
+
 	t.Run("SetMaintenanceCondition", func(t *testing.T) {
 		ActionsExistence(t, api.ActionTypeSetMaintenanceCondition)
 	})
diff --git a/pkg/deployment/reconcile/action_set_current_member_arch.go b/pkg/deployment/reconcile/action_set_current_member_arch.go
diff --git a/pkg/deployment/reconcile/condition_member_recreation.go b/pkg/deployment/reconcile/condition_member_recreation.go
diff --git a/pkg/deployment/reconcile/plan_builder_high.go b/pkg/deployment/reconcile/plan_builder_high.go
diff --git a/pkg/deployment/reconcile/plan_builder_rotate_arch.go b/pkg/deployment/reconcile/plan_builder_rotate_arch.go
diff --git a/pkg/deployment/resources/pod_inspector.go b/pkg/deployment/resources/pod_inspector.go
diff --git a/pkg/deployment/rotation/arangod.go b/pkg/deployment/rotation/arangod.go
diff --git a/pkg/deployment/rotation/arangod_test.go b/pkg/deployment/rotation/arangod_test.go
diff --git a/pkg/deployment/rotation/check.go b/pkg/deployment/rotation/check.go
