Allow setting a cluster offline

api/v1beta2/mysqlcluster_types.go

@@ -122,6 +122,10 @@ type MySQLClusterSpec struct {
 	// During container init moco-agent will set mysql admin interface is bound to localhost. The moco-agent will also
 	// communicate with mysqld over localhost when acting as a sidecar.
 	AgentUseLocalhost bool `json:"agentUseLocalhost,omitempty"`
+
+	// Offline sets the cluster offline, releasing compute resources. Data is not removed.
+	// +optional
+	Offline bool `json:"offline,omitempty"`
 }
 
 func (s MySQLClusterSpec) validateCreate() (admission.Warnings, field.ErrorList) {

backup/backup.go

@@ -102,6 +102,10 @@ func (bm *BackupManager) Backup(ctx context.Context) error {
 		return fmt.Errorf("failed to get pod list: %w", err)
 	}
 
+	if bm.cluster.Spec.Offline {
+		return fmt.Errorf("cluster is configured to be offline %s/%s", bm.cluster.Namespace, bm.cluster.Name)
+	}
+
 	if len(pods.Items) != int(bm.cluster.Spec.Replicas) {
 		return fmt.Errorf("too few Pods for %s/%s", bm.cluster.Namespace, bm.cluster.Name)
 	}

charts/moco/templates/generated/crds/moco_crds.yaml

@@ -2151,6 +2151,9 @@ spec:
                   description: 'MySQLConfigMapName is a `ConfigMap` name of MySQL '
                   nullable: true
                   type: string
+                offline:
+                  description: Offline sets the cluster offline, releasing comput
+                  type: boolean
                 podTemplate:
                   description: PodTemplate is a `Pod` template for MySQL server c
                   properties:

clustering/process.go

@@ -187,6 +187,8 @@ func (p *managerProcess) do(ctx context.Context) (bool, error) {
 
 	logFromContext(ctx).Info("cluster state is " + ss.State.String())
 	switch ss.State {
+	case StateOffline:
+		return false, nil
 	case StateCloning:
 		if p.isCloning(ctx, ss) {
 			return false, nil
@@ -281,6 +283,7 @@ func (p *managerProcess) updateStatus(ctx context.Context, ss *StatusSet) error
 		case StateFailed:
 		case StateLost:
 		case StateIncomplete:
+		case StateOffline:
 		}
 
 		meta.SetStatusCondition(&cluster.Status.Conditions, updateCond(mocov1beta2.ConditionInitialized, initialized))
@@ -309,7 +312,7 @@ func (p *managerProcess) updateStatus(ctx context.Context, ss *StatusSet) error
 
 		var syncedReplicas int
 		for _, pod := range ss.Pods {
-			if isPodReady(pod) {
+			if pod != nil && isPodReady(pod) {
 				syncedReplicas++
 			}
 		}

clustering/status.go

@@ -50,6 +50,7 @@ const (
 	StateDegraded
 	StateFailed
 	StateLost
+	StateOffline
 )
 
 // String returns a unique string for each ClusterState.
@@ -71,6 +72,8 @@ func (s ClusterState) String() string {
 		return "Failed"
 	case StateLost:
 		return "Lost"
+	case StateOffline:
+		return "Offline"
 	}
 
 	panic(int(s))
@@ -107,6 +110,8 @@ func (ss *StatusSet) Close() {
 // It may also set `ss.NeedSwitch` and `ss.Candidate` for switchover.
 func (ss *StatusSet) DecideState() {
 	switch {
+	case isOffline(ss):
+		ss.State = StateOffline
 	case isCloning(ss):
 		ss.State = StateCloning
 	case isRestoring(ss):
@@ -160,7 +165,7 @@ func (p *managerProcess) GatherStatus(ctx context.Context) (*StatusSet, error) {
 		return nil, fmt.Errorf("failed to list Pods: %w", err)
 	}
 
-	if int(cluster.Spec.Replicas) != len(pods.Items) {
+	if !cluster.Spec.Offline && int(cluster.Spec.Replicas) != len(pods.Items) {
 		return nil, fmt.Errorf("too few pods; only %d pods exist", len(pods.Items))
 	}
 	ss.Pods = make([]*corev1.Pod, cluster.Spec.Replicas)
@@ -547,6 +552,10 @@ func isLost(ss *StatusSet) bool {
 	return okReplicas <= (int(ss.Cluster.Spec.Replicas) / 2)
 }
 
+func isOffline(ss *StatusSet) bool {
+	return ss.Cluster.Spec.Offline
+}
+
 func needSwitch(pod *corev1.Pod) bool {
 	if pod.DeletionTimestamp != nil {
 		return true

cmd/kubectl-moco/cmd/switchover.go

@@ -30,6 +30,10 @@ func switchover(ctx context.Context, name string) error {
 		return err
 	}
 
+	if cluster.Spec.Offline {
+		return errors.New("offline cluster is not able to switch")
+	}
+
 	if cluster.Spec.Replicas == 1 {
 		return errors.New("single-instance cluster is not able to switch")
 	}

config/crd/bases/moco.cybozu.com_mysqlclusters.yaml

@@ -82,6 +82,9 @@ spec:
                 description: 'MySQLConfigMapName is a `ConfigMap` name of MySQL '
                 nullable: true
                 type: string
+              offline:
+                description: Offline sets the cluster offline, releasing comput
+                type: boolean
               podTemplate:
                 description: PodTemplate is a `Pod` template for MySQL server c
                 properties:

config/crd/tests/apiextensions.k8s.io_v1_customresourcedefinition_mysqlclusters.moco.cybozu.com.yaml

@@ -82,6 +82,9 @@ spec:
                 description: 'MySQLConfigMapName is a `ConfigMap` name of MySQL '
                 nullable: true
                 type: string
+              offline:
+                description: Offline sets the cluster offline, releasing comput
+                type: boolean
               podTemplate:
                 description: PodTemplate is a `Pod` template for MySQL server c
                 properties:

controllers/mysqlcluster_controller.go

@@ -705,10 +705,14 @@ func (r *MySQLClusterReconciler) reconcileV1StatefulSet(ctx context.Context, req
 		return fmt.Errorf("failed to get StatefulSet %s/%s: %w", cluster.Namespace, cluster.PrefixedName(), err)
 	}
 
+	replicas := cluster.Spec.Replicas
+	if cluster.Spec.Offline {
+		replicas = 0
+	}
 	sts := appsv1ac.StatefulSet(cluster.PrefixedName(), cluster.Namespace).
 		WithLabels(labelSet(cluster, false)).
 		WithSpec(appsv1ac.StatefulSetSpec().
-			WithReplicas(cluster.Spec.Replicas).
+			WithReplicas(replicas).
 			WithSelector(metav1ac.LabelSelector().
 				WithMatchLabels(labelSet(cluster, false))).
 			WithPodManagementPolicy(appsv1.ParallelPodManagement).
@@ -972,7 +976,7 @@ func (r *MySQLClusterReconciler) reconcileV1PDB(ctx context.Context, req ctrl.Re
 	pdb.Namespace = cluster.Namespace
 	pdb.Name = cluster.PrefixedName()
 
-	if cluster.Spec.Replicas < 3 {
+	if cluster.Spec.Offline || cluster.Spec.Replicas < 3 {
 		err := r.Delete(ctx, pdb)
 		if err == nil {
 			log.Info("removed pod disruption budget")

docs/crd_mysqlcluster_v1beta2.md

@@ -84,6 +84,7 @@ MySQLClusterSpec defines the desired state of MySQLCluster
 | restore | Restore is the specification to perform Point-in-Time-Recovery from existing cluster. If this field is not null, MOCO restores the data as specified and create a new cluster with the data.  This field is not editable. | *[RestoreSpec](#restorespec) | false |
 | disableSlowQueryLogContainer | DisableSlowQueryLogContainer controls whether to add a sidecar container named \"slow-log\" to output slow logs as the containers output. If set to true, the sidecar container is not added. The default is false. | bool | false |
 | agentUseLocalhost | AgentUseLocalhost configures the mysqld interface to bind and be accessed over localhost instead of pod name. During container init moco-agent will set mysql admin interface is bound to localhost. The moco-agent will also communicate with mysqld over localhost when acting as a sidecar. | bool | false |
+| offline | Offline sets the cluster offline, releasing compute resources. Data is not removed. | bool | false |
 
 [Back to Custom Resources](#custom-resources)