[Feature] Allow to disable member recreation (#761)

CHANGELOG.md

@@ -2,6 +2,7 @@
 
 ## [master](https://github.com/arangodb/kube-arangodb/tree/master) (N/A)
 - Fix ArangoMember race with multiple ArangoDeployments within single namespace
+- Allow to define Member Recreation Policy within group
 
 ## [1.2.0](https://github.com/arangodb/kube-arangodb/tree/1.2.0) (2021-07-16)
 - Enable "Operator Internal Metrics Exporter" by default

pkg/apis/deployment/v1/deployment_spec.go

@@ -163,6 +163,26 @@ type DeploymentSpec struct {
 	CommunicationMethod *DeploymentCommunicationMethod `json:"communicationMethod,omitempty"`
 }
 
+// GetAllowMemberRecreation returns member recreation policy based on group and settings
+func (s *DeploymentSpec) GetAllowMemberRecreation(group ServerGroup) bool {
+	if s == nil {
+		return false
+	}
+
+	groupSpec := s.GetServerGroupSpec(group)
+
+	switch group {
+	case ServerGroupDBServers, ServerGroupCoordinators:
+		if v := groupSpec.AllowMemberRecreation; v == nil {
+			return true
+		} else {
+			return *v
+		}
+	default:
+		return false
+	}
+}
+
 // GetRestoreFrom returns the restore from string or empty string if not set
 func (s *DeploymentSpec) GetRestoreFrom() string {
 	return util.StringOrDefault(s.RestoreFrom)

pkg/apis/deployment/v1/server_group_spec.go

@@ -141,6 +141,8 @@ type ServerGroupSpec struct {
 	ShutdownDelay *int `json:"shutdownDelay,omitempty"`
 	// InternalPort define port used in internal communication, can be accessed over localhost via sidecar
 	InternalPort *int `json:"internalPort,omitempty"`
+	// AllowMemberRecreation allows to recreate member. Value is used only for Coordinator and DBServer with default to True, for all other groups set to false.
+	AllowMemberRecreation *bool `json:"allowMemberRecreation,omitempty"`
 }
 
 // ServerGroupSpecSecurityContext contains specification for pod security context

pkg/apis/deployment/v1/zz_generated.deepcopy.go

@@ -1522,6 +1522,11 @@ func (in *ServerGroupSpec) DeepCopyInto(out *ServerGroupSpec) {
 		*out = new(int)
 		**out = **in
 	}
+	if in.AllowMemberRecreation != nil {
+		in, out := &in.AllowMemberRecreation, &out.AllowMemberRecreation
+		*out = new(bool)
+		**out = **in
+	}
 	return
 }
 

pkg/apis/deployment/v2alpha1/arango_member_spec.go

@@ -22,11 +22,15 @@
 
 package v2alpha1
 
-import core "k8s.io/api/core/v1"
+import (
+	core "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/types"
+)
 
 type ArangoMemberSpec struct {
-	Group ServerGroup `json:"group,omitempty"`
-	ID    string      `json:"id,omitempty"`
+	Group         ServerGroup `json:"group,omitempty"`
+	ID            string      `json:"id,omitempty"`
+	DeploymentUID types.UID   `json:"deploymentUID,omitempty"`
 
 	Template         *core.PodTemplate `json:"template,omitempty"`
 	TemplateChecksum string            `json:"templateChecksum,omitempty"`

pkg/apis/deployment/v2alpha1/deployment_spec.go

@@ -163,6 +163,26 @@ type DeploymentSpec struct {
 	CommunicationMethod *DeploymentCommunicationMethod `json:"communicationMethod,omitempty"`
 }
 
+// GetAllowMemberRecreation returns member recreation policy based on group and settings
+func (s *DeploymentSpec) GetAllowMemberRecreation(group ServerGroup) bool {
+	if s == nil {
+		return false
+	}
+
+	groupSpec := s.GetServerGroupSpec(group)
+
+	switch group {
+	case ServerGroupDBServers, ServerGroupCoordinators:
+		if v := groupSpec.AllowMemberRecreation; v == nil {
+			return true
+		} else {
+			return *v
+		}
+	default:
+		return false
+	}
+}
+
 // GetRestoreFrom returns the restore from string or empty string if not set
 func (s *DeploymentSpec) GetRestoreFrom() string {
 	return util.StringOrDefault(s.RestoreFrom)

pkg/apis/deployment/v2alpha1/server_group_spec.go

@@ -141,6 +141,8 @@ type ServerGroupSpec struct {
 	ShutdownDelay *int `json:"shutdownDelay,omitempty"`
 	// InternalPort define port used in internal communication, can be accessed over localhost via sidecar
 	InternalPort *int `json:"internalPort,omitempty"`
+	// AllowMemberRecreation allows to recreate member. Value is used only for Coordinator and DBServer with default to True, for all other groups set to false.
+	AllowMemberRecreation *bool `json:"allowMemberRecreation,omitempty"`
 }
 
 // ServerGroupSpecSecurityContext contains specification for pod security context

pkg/apis/deployment/v2alpha1/zz_generated.deepcopy.go

@@ -1522,6 +1522,11 @@ func (in *ServerGroupSpec) DeepCopyInto(out *ServerGroupSpec) {
 		*out = new(int)
 		**out = **in
 	}
+	if in.AllowMemberRecreation != nil {
+		in, out := &in.AllowMemberRecreation, &out.AllowMemberRecreation
+		*out = new(bool)
+		**out = **in
+	}
 	return
 }
 

pkg/deployment/reconcile/plan_builder.go

@@ -169,13 +169,18 @@ func createPlan(ctx context.Context, log zerolog.Logger, apiObject k8sutil.APIOb
 				plan = append(plan,
 					api.NewAction(api.ActionTypeRecreateMember, group, m.ID))
 			default:
-				memberLog.Msg("Creating member replacement plan because member has failed")
-				plan = append(plan,
-					api.NewAction(api.ActionTypeRemoveMember, group, m.ID),
-					api.NewAction(api.ActionTypeAddMember, group, ""),
-					api.NewAction(api.ActionTypeWaitForMemberUp, group, api.MemberIDPreviousAction),
-				)
-
+				if spec.GetAllowMemberRecreation(group) {
+					memberLog.Msg("Creating member replacement plan because member has failed")
+					plan = append(plan,
+						api.NewAction(api.ActionTypeRemoveMember, group, m.ID),
+						api.NewAction(api.ActionTypeAddMember, group, ""),
+						api.NewAction(api.ActionTypeWaitForMemberUp, group, api.MemberIDPreviousAction),
+					)
+				} else {
+					memberLog.Msg("Restoring old member. Recreation is disabled for group")
+					plan = append(plan,
+						api.NewAction(api.ActionTypeRecreateMember, group, m.ID))
+				}
 			}
 		}
 		return nil