[Bugfix] Fix AKS Volume resize mode (#688)

Author: ajanikow

CHANGELOG.md

@@ -1,6 +1,7 @@
 # Change Log
 
 ## [master](https://github.com/arangodb/kube-arangodb/tree/master) (N/A)
+- Fix AKS Volume Resize mode
 
 ## [1.1.4](https://github.com/arangodb/kube-arangodb/tree/1.1.4) (2021-02-15)
 - Add support for spec.ClusterDomain to be able to use FQDN in ArangoDB cluster communication

pkg/apis/deployment/v1/member_phase.go

@@ -42,6 +42,8 @@ const (
 	MemberPhaseShuttingDown MemberPhase = "ShuttingDown"
 	// MemberPhaseRotating indicates that a member is being rotated
 	MemberPhaseRotating MemberPhase = "Rotating"
+	// MemberPhaseRotateStart indicates that a member is being rotated but wont get up outside of plan
+	MemberPhaseRotateStart MemberPhase = "RotateStart"
 	// MemberPhaseUpgrading indicates that a member is in the process of upgrading its database data format
 	MemberPhaseUpgrading MemberPhase = "Upgrading"
 )

pkg/apis/deployment/v2alpha1/member_phase.go

@@ -42,6 +42,8 @@ const (
 	MemberPhaseShuttingDown MemberPhase = "ShuttingDown"
 	// MemberPhaseRotating indicates that a member is being rotated
 	MemberPhaseRotating MemberPhase = "Rotating"
+	// MemberPhaseRotateStart indicates that a member is being rotated but wont get up outside of plan
+	MemberPhaseRotateStart MemberPhase = "RotateStart"
 	// MemberPhaseUpgrading indicates that a member is in the process of upgrading its database data format
 	MemberPhaseUpgrading MemberPhase = "Upgrading"
 )

pkg/deployment/reconcile/action_rotate_start_member.go

@@ -92,7 +92,7 @@ func (a *actionRotateStartMember) Start(ctx context.Context) (bool, error) {
 		}
 	}
 	// Update status
-	m.Phase = api.MemberPhaseRotating
+	m.Phase = api.MemberPhaseRotateStart
 
 	if err := a.actionCtx.UpdateMember(m); err != nil {
 		return false, errors.WithStack(err)

pkg/deployment/reconcile/plan_builder_storage.go

@@ -141,13 +141,23 @@ func pvcResizePlan(log zerolog.Logger, group api.ServerGroup, groupSpec api.Serv
 			api.NewAction(api.ActionTypePVCResize, group, memberID),
 		}
 	case api.PVCResizeModeRotate:
-		return api.Plan{
+		var plan api.Plan
+
+		if group == api.ServerGroupDBServers {
+			plan = append(plan,
+				api.NewAction(api.ActionTypeCleanOutMember, group, memberID),
+			)
+		}
+
+		plan = append(plan,
 			api.NewAction(api.ActionTypeRotateStartMember, group, memberID),
 			api.NewAction(api.ActionTypePVCResize, group, memberID),
 			api.NewAction(api.ActionTypePVCResized, group, memberID),
 			api.NewAction(api.ActionTypeRotateStopMember, group, memberID),
 			api.NewAction(api.ActionTypeWaitForMemberUp, group, memberID),
-		}
+		)
+
+		return plan
 	default:
 		log.Error().Str("server-group", group.AsRole()).Str("mode", mode.String()).
 			Msg("Requested mode is not supported")

pkg/deployment/reconcile/timeouts.go

@@ -33,7 +33,7 @@ const (
 	renewTLSCACertificateTimeout     = time.Minute * 30
 	operationTLSCACertificateTimeout = time.Minute * 30
 	rotateMemberTimeout              = time.Minute * 15
-	pvcResizeTimeout                 = time.Minute * 15
+	pvcResizeTimeout                 = time.Minute * 30
 	pvcResizedTimeout                = time.Minute * 15
 	backupRestoreTimeout             = time.Minute * 15
 	shutdownMemberTimeout            = time.Minute * 30

pkg/deployment/resilience/member_failure.go

@@ -56,7 +56,7 @@ func (r *Resilience) CheckMemberFailure() error {
 			switch m.Phase {
 			case api.MemberPhaseNone:
 				continue
-			case api.MemberPhaseUpgrading, api.MemberPhaseRotating, api.MemberPhaseCleanOut:
+			case api.MemberPhaseUpgrading, api.MemberPhaseRotating, api.MemberPhaseCleanOut, api.MemberPhaseRotateStart:
 				if len(status.Plan) == 0 {
 					log.Error().Msgf("No plan but member is in phase %s - marking as failed", m.Phase)
 					m.Phase = api.MemberPhaseFailed

pkg/deployment/resources/pod_inspector.go

@@ -248,7 +248,7 @@ func (r *Resources) InspectPods(ctx context.Context, cachedStatus inspector.Insp
 					case api.MemberPhaseNone:
 						// Do nothing
 						log.Debug().Str("pod-name", podName).Msg("PodPhase is None, waiting for the pod to be recreated")
-					case api.MemberPhaseShuttingDown, api.MemberPhaseUpgrading, api.MemberPhaseFailed:
+					case api.MemberPhaseShuttingDown, api.MemberPhaseUpgrading, api.MemberPhaseFailed, api.MemberPhaseRotateStart:
 						// Shutdown was intended, so not need to do anything here.
 						// Just mark terminated
 						wasTerminated := m.Conditions.IsTrue(api.ConditionTypeTerminated)