SCF-830: Added state management for cluster hibernation

Author: adshin21

manifests/postgresql.crd.yaml

@@ -3246,6 +3246,13 @@ spec:
                   - name
                   type: object
                 type: array
+              lifecycle:
+                description: LifecycleSpec describes the lifecycle state of a Postgres
+                  cluster.
+                properties:
+                  phase:
+                    type: string
+                type: object
               logicalBackupRetention:
                 type: string
               logicalBackupSchedule:
@@ -4197,6 +4204,9 @@ spec:
             properties:
               PostgresClusterStatus:
                 type: string
+              previousNumberOfInstances:
+                format: int32
+                type: integer
             required:
             - PostgresClusterStatus
             type: object

pkg/apis/acid.zalan.do/v1/const.go

@@ -9,6 +9,8 @@ const (
 	ClusterStatusSyncFailed   = "SyncFailed"
 	ClusterStatusAddFailed    = "CreateFailed"
 	ClusterStatusRunning      = "Running"
+	ClusterStatusStopping     = "Stopping"
+	ClusterStatusStopped      = "Stopped"
 	ClusterStatusInvalid      = "Invalid"
 )
 

pkg/apis/acid.zalan.do/v1/postgresql.crd.yaml

@@ -3246,6 +3246,13 @@ spec:
                   - name
                   type: object
                 type: array
+              lifecycle:
+                description: LifecycleSpec describes the lifecycle state of a Postgres
+                  cluster.
+                properties:
+                  phase:
+                    type: string
+                type: object
               logicalBackupRetention:
                 type: string
               logicalBackupSchedule:
@@ -4197,6 +4204,9 @@ spec:
             properties:
               PostgresClusterStatus:
                 type: string
+              previousNumberOfInstances:
+                format: int32
+                type: integer
             required:
             - PostgresClusterStatus
             type: object

pkg/apis/acid.zalan.do/v1/postgresql_type.go

@@ -115,6 +115,7 @@ type PostgresSpec struct {
 	TLS                       *TLSDescription    `json:"tls,omitempty"`
 	AdditionalVolumes         []AdditionalVolume `json:"additionalVolumes,omitempty"`
 	Streams                   []Stream           `json:"streams,omitempty"`
+	Lifecycle                 *LifecycleSpec     `json:"lifecycle,omitempty"`
 	Env                       []v1.EnvVar        `json:"env,omitempty"`
 
 	// deprecated
@@ -257,6 +258,11 @@ type StandbyDescription struct {
 	StandbyPrimarySlotName string `json:"standby_primary_slot_name,omitempty"`
 }
 
+// LifecycleSpec describes the lifecycle state of a Postgres cluster.
+type LifecycleSpec struct {
+	Phase string `json:"phase,omitempty"`
+}
+
 // TLSDescription specs TLS properties
 type TLSDescription struct {
 	// +required
@@ -302,7 +308,8 @@ type UserFlags []string
 
 // PostgresStatus contains status of the PostgreSQL cluster (running, creation failed etc.)
 type PostgresStatus struct {
-	PostgresClusterStatus string `json:"PostgresClusterStatus"`
+	PostgresClusterStatus     string `json:"PostgresClusterStatus"`
+	PreviousNumberOfInstances int32  `json:"previousNumberOfInstances,omitempty"`
 }
 
 // ConnectionPooler Options for connection pooler

pkg/apis/acid.zalan.do/v1/util.go

@@ -101,6 +101,16 @@ func (postgresStatus PostgresStatus) Creating() bool {
 	return postgresStatus.PostgresClusterStatus == ClusterStatusCreating
 }
 
+// Stopping status of cluster
+func (postgresStatus PostgresStatus) Stopping() bool {
+	return postgresStatus.PostgresClusterStatus == ClusterStatusStopping
+}
+
+// Stopped status of cluster
+func (postgresStatus PostgresStatus) Stopped() bool {
+	return postgresStatus.PostgresClusterStatus == ClusterStatusStopped
+}
+
 func (postgresStatus PostgresStatus) String() string {
 	return postgresStatus.PostgresClusterStatus
 }

pkg/apis/acid.zalan.do/v1/zz_generated.deepcopy.go

@@ -1008,13 +1008,94 @@ func (c *Cluster) Update(oldSpec, newSpec *acidv1.Postgresql) error {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
+	// Block all spec changes when cluster is stopped or stopping
+	if c.Status.Stopped() || c.Status.Stopping() {
+		lifecyclePhase := ""
+		if newSpec.Spec.Lifecycle != nil {
+			lifecyclePhase = newSpec.Spec.Lifecycle.Phase
+		}
+		// During Stopping: block ALL spec changes (no cancellation allowed)
+		// During Stopped: only block if keeping lifecycle.phase="stopped"
+		if c.Status.Stopping() {
+			return fmt.Errorf("cannot update cluster while it is stopping. Wait for it to fully stop first")
+		}
+		if lifecyclePhase == "stopped" {
+			return fmt.Errorf("cannot update cluster while stopped. Remove lifecycle.phase to wake up the cluster")
+		}
+	}
+
 	newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusUpdating
 
 	newSpec, err := c.KubeClient.SetPostgresCRDStatus(c.clusterName(), newSpec)
 	if err != nil {
 		return fmt.Errorf("could not set cluster status to updating: %w", err)
 	}
 
+	// Check if user is initiating hibernate (Running -> Stopping)
+	if c.Status.Running() && newSpec.Spec.Lifecycle != nil && newSpec.Spec.Lifecycle.Phase == "stopped" {
+		c.logger.Infof("[lifecycle] initiating hibernate for cluster %s: current numberOfInstances=%d", c.Name, c.Spec.NumberOfInstances)
+
+		// Store previousNumberOfInstances BEFORE setting numberOfInstances to 0
+		newSpec.Status.PreviousNumberOfInstances = c.Spec.NumberOfInstances
+		newSpec.Spec.NumberOfInstances = 0
+		newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusStopping
+
+		c.logger.Infof("[lifecycle] hibernate initiated: setting numberOfInstances=0, previousNumberOfInstances=%d", newSpec.Status.PreviousNumberOfInstances)
+
+		// Update spec first (Update only updates spec when CR has status subresource)
+		pgUpdated, err := c.KubeClient.UpdatePostgresCR(c.clusterName(), newSpec)
+		if err != nil {
+			return fmt.Errorf("could not update spec during hibernate: %w", err)
+		}
+		c.logger.Infof("[lifecycle] hibernate: spec updated successfully")
+
+		// Update status separately - we need to preserve the status values we set
+		// because UpdatePostgresCR returns object with status zeroed (subresource behavior)
+		pgUpdated.Status.PreviousNumberOfInstances = newSpec.Status.PreviousNumberOfInstances
+		pgUpdated.Status.PostgresClusterStatus = newSpec.Status.PostgresClusterStatus
+
+		pgUpdated, err = c.KubeClient.SetPostgresCRDStatus(c.clusterName(), pgUpdated)
+		if err != nil {
+			return fmt.Errorf("could not update status during hibernate: %w", err)
+		}
+		c.logger.Infof("[lifecycle] hibernate: status updated successfully, previousNumberOfInstances=%d", pgUpdated.Status.PreviousNumberOfInstances)
+
+		c.setSpec(pgUpdated)
+		return nil
+	}
+
+	// Check if user is waking up from stopped state (Stopped -> Running)
+	// This is when user clears lifecycle.phase to wake up the cluster
+	if c.Status.Stopped() && (newSpec.Spec.Lifecycle == nil || newSpec.Spec.Lifecycle.Phase != "stopped") {
+		if newSpec.Status.PreviousNumberOfInstances > 0 {
+			c.logger.Infof("[lifecycle] waking up cluster %s: restoring numberOfInstances=%d", c.Name, newSpec.Status.PreviousNumberOfInstances)
+
+			// Restore numberOfInstances from previousNumberOfInstances
+			newSpec.Spec.NumberOfInstances = newSpec.Status.PreviousNumberOfInstances
+			newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusUpdating
+
+			// Update spec first
+			pgUpdated, err := c.KubeClient.UpdatePostgresCR(c.clusterName(), newSpec)
+			if err != nil {
+				return fmt.Errorf("could not update spec during wake-up: %w", err)
+			}
+			c.logger.Infof("[lifecycle] wake-up: spec updated successfully")
+
+			// Update status separately, and clear previousNumberOfInstances after restore
+			pgUpdated.Status.PreviousNumberOfInstances = 0 // Clear after successful restore
+			pgUpdated.Status.PostgresClusterStatus = newSpec.Status.PostgresClusterStatus
+
+			pgUpdated, err = c.KubeClient.SetPostgresCRDStatus(c.clusterName(), pgUpdated)
+			if err != nil {
+				return fmt.Errorf("could not update status during wake-up: %w", err)
+			}
+			c.logger.Infof("[lifecycle] wake-up: status updated successfully, previousNumberOfInstances cleared")
+
+			c.setSpec(pgUpdated)
+			return nil
+		}
+	}
+
 	if !c.isInMaintenanceWindow(newSpec.Spec.MaintenanceWindows) {
 		// do not apply any major version related changes yet
 		newSpec.Spec.PostgresqlParam.PgVersion = oldSpec.Spec.PostgresqlParam.PgVersion

pkg/cluster/cluster.go

@@ -0,0 +1,78 @@
+package cluster
+
+import (
+	acidv1 "github.com/zalando/postgres-operator/pkg/apis/acid.zalan.do/v1"
+)
+
+// manageHibernateState manages cluster hibernate/wake-up state transitions.
+// Returns true if sync should continue, false if it should return early.
+//
+// This function handles the following state transitions:
+//   - Running -> Stopping: When user sets lifecycle.phase = "stopped"
+//   - Stopping -> Stopped: When StatefulSet replicas reach 0
+//   - Stopped -> Updating: When user clears lifecycle.phase (wake-up)
+//   - Updating -> Running: Normal sync continues, defer sets final status
+func (c *Cluster) manageHibernateState(oldSpec acidv1.Postgresql, newSpec *acidv1.Postgresql) bool {
+
+	// FIX B: Detect wake-up by comparing oldSpec status vs newSpec status
+	// When Update() is called, it sets status=Updating before Sync() runs.
+	// So we need to check if oldSpec.Status was Stopped and newSpec is Updating
+	// with lifecycle cleared to properly detect wake-up.
+	isWakingUp := oldSpec.Status.Stopped() &&
+		newSpec.Status.PostgresClusterStatus == acidv1.ClusterStatusUpdating &&
+		(newSpec.Spec.Lifecycle == nil || newSpec.Spec.Lifecycle.Phase != "stopped")
+
+	// FIX C: Additional wake-up detection with simpler condition
+	// If lifecycle was cleared and we have previousNumberOfInstances and numberOfInstances is 0
+	isWakingUpSimple := newSpec.Spec.Lifecycle == nil || newSpec.Spec.Lifecycle.Phase != "stopped"
+	hasPreviousInstances := newSpec.Status.PreviousNumberOfInstances > 0
+	needsRestore := newSpec.Spec.NumberOfInstances == 0
+
+	isWakingUp = isWakingUp || (isWakingUpSimple && hasPreviousInstances && needsRestore)
+
+	// === INITIATE HIBERNATE: Running -> Stopping ===
+	// Only initiate if not already stopping or stopped, and lifecycle.phase = "stopped"
+	if newSpec.Spec.Lifecycle != nil &&
+		newSpec.Spec.Lifecycle.Phase == "stopped" &&
+		!newSpec.Status.Stopping() &&
+		!newSpec.Status.Stopped() {
+
+		newSpec.Status.PreviousNumberOfInstances = newSpec.Spec.NumberOfInstances
+		newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusStopping
+		newSpec.Spec.NumberOfInstances = 0
+		c.logger.Infof("[lifecycle] cluster is going to hibernate, stored previous number of instances: %d",
+			newSpec.Status.PreviousNumberOfInstances)
+		return true
+	}
+
+	// === STOPPING -> STOPPED: Check actual StatefulSet replicas ===
+	// Only transition to Stopped when StatefulSet replicas have actually reached 0
+	if newSpec.Status.Stopping() {
+		if c.Statefulset != nil && *c.Statefulset.Spec.Replicas == 0 {
+			newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusStopped
+			c.logger.Infof("[lifecycle] cluster has stopped, all pods are terminated")
+		}
+		return true
+	}
+
+	// === WAKE-UP: Stopped/Updating -> Running ===
+	// Restore numberOfInstances from previousNumberOfInstances when waking up
+	if newSpec.Status.Stopped() || isWakingUp {
+		// Check if lifecycle.phase was cleared (user wants to wake up)
+		if isWakingUp || newSpec.Spec.Lifecycle == nil || newSpec.Spec.Lifecycle.Phase != "stopped" {
+			if newSpec.Status.PreviousNumberOfInstances > 0 {
+				newSpec.Spec.NumberOfInstances = newSpec.Status.PreviousNumberOfInstances
+				c.logger.Infof("[lifecycle] cluster is waking up, restoring number of instances: %d",
+					newSpec.Status.PreviousNumberOfInstances)
+			} else {
+				c.logger.Warningf("[lifecycle] cluster is waking up but previousNumberOfInstances is 0, cannot restore")
+			}
+			newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusUpdating
+			return true
+		}
+		// Still stopped and lifecycle.phase = "stopped", skip further sync
+		return false
+	}
+
+	return true
+}

pkg/cluster/lifecycle.go

@@ -47,7 +47,7 @@ func (c *Cluster) Sync(newSpec *acidv1.Postgresql) error {
 		if err != nil {
 			c.logger.Warningf("error while syncing cluster state: %v", err)
 			newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusSyncFailed
-		} else if !c.Status.Running() {
+		} else if !c.Status.Running() && !c.Status.Stopping() && !c.Status.Stopped() {
 			newSpec.Status.PostgresClusterStatus = acidv1.ClusterStatusRunning
 		}
 
@@ -65,6 +65,11 @@ func (c *Cluster) Sync(newSpec *acidv1.Postgresql) error {
 		c.logger.Debugf("could not sync finalizers: %v", err)
 	}
 
+	// Handle lifecycle hibernate/wake-up state transitions
+	if !c.manageHibernateState(oldSpec, newSpec) {
+		return nil
+	}
+
 	if err = c.initUsers(); err != nil {
 		err = fmt.Errorf("could not init users: %v", err)
 		return err

pkg/cluster/sync.go

@@ -200,6 +200,16 @@ func (client *KubernetesClient) SetPostgresCRDStatus(clusterName spec.Namespaced
 	return pg, nil
 }
 
+// UpdatePostgresCR of Postgres cluster (updates full resource including spec)
+func (client *KubernetesClient) UpdatePostgresCR(clusterName spec.NamespacedName, pg *apiacidv1.Postgresql) (*apiacidv1.Postgresql, error) {
+	pg, err := client.PostgresqlsGetter.Postgresqls(clusterName.Namespace).Update(context.TODO(), pg, metav1.UpdateOptions{})
+	if err != nil {
+		return pg, fmt.Errorf("could not update PostgresCR: %v", err)
+	}
+
+	return pg, nil
+}
+
 // SetFinalizer of Postgres cluster
 func (client *KubernetesClient) SetFinalizer(clusterName spec.NamespacedName, pg *apiacidv1.Postgresql, finalizers []string) (*apiacidv1.Postgresql, error) {
 	var (