Critical operation PDB (#2830)

Create the second PDB to cover Pods with a special "critical operation" label set.

This label is going to be assigned to all pg cluster's Pods by the Operator during a PG major version upgrade, by Patroni during a cluster/replica bootstrap. It can also be set manually or by any other automation tool.

pkg/cluster/cluster.go

@@ -59,16 +59,17 @@ type Config struct {
 }
 
 type kubeResources struct {
-	Services            map[PostgresRole]*v1.Service
-	Endpoints           map[PostgresRole]*v1.Endpoints
-	PatroniEndpoints    map[string]*v1.Endpoints
-	PatroniConfigMaps   map[string]*v1.ConfigMap
-	Secrets             map[types.UID]*v1.Secret
-	Statefulset         *appsv1.StatefulSet
-	VolumeClaims        map[types.UID]*v1.PersistentVolumeClaim
-	PodDisruptionBudget *policyv1.PodDisruptionBudget
-	LogicalBackupJob    *batchv1.CronJob
-	Streams             map[string]*zalandov1.FabricEventStream
+	Services                      map[PostgresRole]*v1.Service
+	Endpoints                     map[PostgresRole]*v1.Endpoints
+	PatroniEndpoints              map[string]*v1.Endpoints
+	PatroniConfigMaps             map[string]*v1.ConfigMap
+	Secrets                       map[types.UID]*v1.Secret
+	Statefulset                   *appsv1.StatefulSet
+	VolumeClaims                  map[types.UID]*v1.PersistentVolumeClaim
+	PrimaryPodDisruptionBudget    *policyv1.PodDisruptionBudget
+	CriticalOpPodDisruptionBudget *policyv1.PodDisruptionBudget
+	LogicalBackupJob              *batchv1.CronJob
+	Streams                       map[string]*zalandov1.FabricEventStream
 	//Pods are treated separately
 }
 
@@ -343,14 +344,10 @@ func (c *Cluster) Create() (err error) {
 	c.logger.Infof("secrets have been successfully created")
 	c.eventRecorder.Event(c.GetReference(), v1.EventTypeNormal, "Secrets", "The secrets have been successfully created")
 
-	if c.PodDisruptionBudget != nil {
-		return fmt.Errorf("pod disruption budget already exists in the cluster")
+	if err = c.createPodDisruptionBudgets(); err != nil {
+		return fmt.Errorf("could not create pod disruption budgets: %v", err)
 	}
-	pdb, err := c.createPodDisruptionBudget()
-	if err != nil {
-		return fmt.Errorf("could not create pod disruption budget: %v", err)
-	}
-	c.logger.Infof("pod disruption budget %q has been successfully created", util.NameFromMeta(pdb.ObjectMeta))
+	c.logger.Info("pod disruption budgets have been successfully created")
 
 	if c.Statefulset != nil {
 		return fmt.Errorf("statefulset already exists in the cluster")
@@ -1081,9 +1078,9 @@ func (c *Cluster) Update(oldSpec, newSpec *acidv1.Postgresql) error {
 		}
 	}
 
-	// pod disruption budget
-	if err := c.syncPodDisruptionBudget(true); err != nil {
-		c.logger.Errorf("could not sync pod disruption budget: %v", err)
+	// pod disruption budgets
+	if err := c.syncPodDisruptionBudgets(true); err != nil {
+		c.logger.Errorf("could not sync pod disruption budgets: %v", err)
 		updateFailed = true
 	}
 
@@ -1228,10 +1225,10 @@ func (c *Cluster) Delete() error {
 		c.logger.Info("not deleting secrets because disabled in configuration")
 	}
 
-	if err := c.deletePodDisruptionBudget(); err != nil {
+	if err := c.deletePodDisruptionBudgets(); err != nil {
 		anyErrors = true
-		c.logger.Warningf("could not delete pod disruption budget: %v", err)
-		c.eventRecorder.Eventf(c.GetReference(), v1.EventTypeWarning, "Delete", "could not delete pod disruption budget: %v", err)
+		c.logger.Warningf("could not delete pod disruption budgets: %v", err)
+		c.eventRecorder.Eventf(c.GetReference(), v1.EventTypeWarning, "Delete", "could not delete pod disruption budgets: %v", err)
 	}
 
 	for _, role := range []PostgresRole{Master, Replica} {
@@ -1730,16 +1727,17 @@ func (c *Cluster) GetCurrentProcess() Process {
 // GetStatus provides status of the cluster
 func (c *Cluster) GetStatus() *ClusterStatus {
 	status := &ClusterStatus{
-		Cluster:             c.Name,
-		Namespace:           c.Namespace,
-		Team:                c.Spec.TeamID,
-		Status:              c.Status,
-		Spec:                c.Spec,
-		MasterService:       c.GetServiceMaster(),
-		ReplicaService:      c.GetServiceReplica(),
-		StatefulSet:         c.GetStatefulSet(),
-		PodDisruptionBudget: c.GetPodDisruptionBudget(),
-		CurrentProcess:      c.GetCurrentProcess(),
+		Cluster:                       c.Name,
+		Namespace:                     c.Namespace,
+		Team:                          c.Spec.TeamID,
+		Status:                        c.Status,
+		Spec:                          c.Spec,
+		MasterService:                 c.GetServiceMaster(),
+		ReplicaService:                c.GetServiceReplica(),
+		StatefulSet:                   c.GetStatefulSet(),
+		PrimaryPodDisruptionBudget:    c.GetPrimaryPodDisruptionBudget(),
+		CriticalOpPodDisruptionBudget: c.GetCriticalOpPodDisruptionBudget(),
+		CurrentProcess:                c.GetCurrentProcess(),
 
 		Error: fmt.Errorf("error: %s", c.Error),
 	}

pkg/cluster/k8sres.go

@@ -109,10 +109,15 @@ func (c *Cluster) servicePort(role PostgresRole) int32 {
 	return pgPort
 }
 
-func (c *Cluster) podDisruptionBudgetName() string {
+func (c *Cluster) PrimaryPodDisruptionBudgetName() string {
 	return c.OpConfig.PDBNameFormat.Format("cluster", c.Name)
 }
 
+func (c *Cluster) criticalOpPodDisruptionBudgetName() string {
+	pdbTemplate := config.StringTemplate("postgres-{cluster}-critical-op-pdb")
+	return pdbTemplate.Format("cluster", c.Name)
+}
+
 func makeDefaultResources(config *config.Config) acidv1.Resources {
 
 	defaultRequests := acidv1.ResourceDescription{
@@ -2207,7 +2212,7 @@ func (c *Cluster) generateStandbyEnvironment(description *acidv1.StandbyDescript
 	return result
 }
 
-func (c *Cluster) generatePodDisruptionBudget() *policyv1.PodDisruptionBudget {
+func (c *Cluster) generatePrimaryPodDisruptionBudget() *policyv1.PodDisruptionBudget {
 	minAvailable := intstr.FromInt(1)
 	pdbEnabled := c.OpConfig.EnablePodDisruptionBudget
 	pdbMasterLabelSelector := c.OpConfig.PDBMasterLabelSelector
@@ -2225,7 +2230,36 @@ func (c *Cluster) generatePodDisruptionBudget() *policyv1.PodDisruptionBudget {
 
 	return &policyv1.PodDisruptionBudget{
 		ObjectMeta: metav1.ObjectMeta{
-			Name:            c.podDisruptionBudgetName(),
+			Name:            c.PrimaryPodDisruptionBudgetName(),
+			Namespace:       c.Namespace,
+			Labels:          c.labelsSet(true),
+			Annotations:     c.annotationsSet(nil),
+			OwnerReferences: c.ownerReferences(),
+		},
+		Spec: policyv1.PodDisruptionBudgetSpec{
+			MinAvailable: &minAvailable,
+			Selector: &metav1.LabelSelector{
+				MatchLabels: labels,
+			},
+		},
+	}
+}
+
+func (c *Cluster) generateCriticalOpPodDisruptionBudget() *policyv1.PodDisruptionBudget {
+	minAvailable := intstr.FromInt32(c.Spec.NumberOfInstances)
+	pdbEnabled := c.OpConfig.EnablePodDisruptionBudget
+
+	// if PodDisruptionBudget is disabled or if there are no DB pods, set the budget to 0.
+	if (pdbEnabled != nil && !(*pdbEnabled)) || c.Spec.NumberOfInstances <= 0 {
+		minAvailable = intstr.FromInt(0)
+	}
+
+	labels := c.labelsSet(false)
+	labels["critical-operation"] = "true"
+
+	return &policyv1.PodDisruptionBudget{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:            c.criticalOpPodDisruptionBudgetName(),
 			Namespace:       c.Namespace,
 			Labels:          c.labelsSet(true),
 			Annotations:     c.annotationsSet(nil),

pkg/cluster/k8sres_test.go

@@ -2349,22 +2349,34 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 		}
 	}
 
-	testLabelsAndSelectors := func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error {
-		masterLabelSelectorDisabled := cluster.OpConfig.PDBMasterLabelSelector != nil && !*cluster.OpConfig.PDBMasterLabelSelector
-		if podDisruptionBudget.ObjectMeta.Namespace != "myapp" {
-			return fmt.Errorf("Object Namespace incorrect.")
-		}
-		if !reflect.DeepEqual(podDisruptionBudget.Labels, map[string]string{"team": "myapp", "cluster-name": "myapp-database"}) {
-			return fmt.Errorf("Labels incorrect.")
-		}
-		if !masterLabelSelectorDisabled &&
-			!reflect.DeepEqual(podDisruptionBudget.Spec.Selector, &metav1.LabelSelector{
-				MatchLabels: map[string]string{"spilo-role": "master", "cluster-name": "myapp-database"}}) {
+	testLabelsAndSelectors := func(isPrimary bool) func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error {
+		return func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error {
+			masterLabelSelectorDisabled := cluster.OpConfig.PDBMasterLabelSelector != nil && !*cluster.OpConfig.PDBMasterLabelSelector
+			if podDisruptionBudget.ObjectMeta.Namespace != "myapp" {
+				return fmt.Errorf("Object Namespace incorrect.")
+			}
+			expectedLabels := map[string]string{"team": "myapp", "cluster-name": "myapp-database"}
+			if !reflect.DeepEqual(podDisruptionBudget.Labels, expectedLabels) {
+				return fmt.Errorf("Labels incorrect, got %#v, expected %#v", podDisruptionBudget.Labels, expectedLabels)
+			}
+			if !masterLabelSelectorDisabled {
+				if isPrimary {
+					expectedLabels := &metav1.LabelSelector{
+						MatchLabels: map[string]string{"spilo-role": "master", "cluster-name": "myapp-database"}}
+					if !reflect.DeepEqual(podDisruptionBudget.Spec.Selector, expectedLabels) {
+						return fmt.Errorf("MatchLabels incorrect, got %#v, expected %#v", podDisruptionBudget.Spec.Selector, expectedLabels)
+					}
+				} else {
+					expectedLabels := &metav1.LabelSelector{
+						MatchLabels: map[string]string{"cluster-name": "myapp-database", "critical-operation": "true"}}
+					if !reflect.DeepEqual(podDisruptionBudget.Spec.Selector, expectedLabels) {
+						return fmt.Errorf("MatchLabels incorrect, got %#v, expected %#v", podDisruptionBudget.Spec.Selector, expectedLabels)
+					}
+				}
+			}
 
-			return fmt.Errorf("MatchLabels incorrect.")
+			return nil
 		}
-
-		return nil
 	}
 
 	testPodDisruptionBudgetOwnerReference := func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error {
@@ -2400,7 +2412,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 				testPodDisruptionBudgetOwnerReference,
 				hasName("postgres-myapp-database-pdb"),
 				hasMinAvailable(1),
-				testLabelsAndSelectors,
+				testLabelsAndSelectors(true),
 			},
 		},
 		{
@@ -2417,7 +2429,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 				testPodDisruptionBudgetOwnerReference,
 				hasName("postgres-myapp-database-pdb"),
 				hasMinAvailable(0),
-				testLabelsAndSelectors,
+				testLabelsAndSelectors(true),
 			},
 		},
 		{
@@ -2434,7 +2446,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 				testPodDisruptionBudgetOwnerReference,
 				hasName("postgres-myapp-database-pdb"),
 				hasMinAvailable(0),
-				testLabelsAndSelectors,
+				testLabelsAndSelectors(true),
 			},
 		},
 		{
@@ -2451,7 +2463,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 				testPodDisruptionBudgetOwnerReference,
 				hasName("postgres-myapp-database-databass-budget"),
 				hasMinAvailable(1),
-				testLabelsAndSelectors,
+				testLabelsAndSelectors(true),
 			},
 		},
 		{
@@ -2468,7 +2480,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 				testPodDisruptionBudgetOwnerReference,
 				hasName("postgres-myapp-database-pdb"),
 				hasMinAvailable(1),
-				testLabelsAndSelectors,
+				testLabelsAndSelectors(true),
 			},
 		},
 		{
@@ -2485,13 +2497,99 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 				testPodDisruptionBudgetOwnerReference,
 				hasName("postgres-myapp-database-pdb"),
 				hasMinAvailable(1),
-				testLabelsAndSelectors,
+				testLabelsAndSelectors(true),
 			},
 		},
 	}
 
 	for _, tt := range tests {
-		result := tt.spec.generatePodDisruptionBudget()
+		result := tt.spec.generatePrimaryPodDisruptionBudget()
+		for _, check := range tt.check {
+			err := check(tt.spec, result)
+			if err != nil {
+				t.Errorf("%s [%s]: PodDisruptionBudget spec is incorrect, %+v",
+					testName, tt.scenario, err)
+			}
+		}
+	}
+
+	testCriticalOp := []struct {
+		scenario string
+		spec     *Cluster
+		check    []func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error
+	}{
+		{
+			scenario: "With multiple instances",
+			spec: New(
+				Config{OpConfig: config.Config{Resources: config.Resources{ClusterNameLabel: "cluster-name", PodRoleLabel: "spilo-role"}, PDBNameFormat: "postgres-{cluster}-pdb"}},
+				k8sutil.KubernetesClient{},
+				acidv1.Postgresql{
+					ObjectMeta: metav1.ObjectMeta{Name: "myapp-database", Namespace: "myapp"},
+					Spec:       acidv1.PostgresSpec{TeamID: "myapp", NumberOfInstances: 3}},
+				logger,
+				eventRecorder),
+			check: []func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error{
+				testPodDisruptionBudgetOwnerReference,
+				hasName("postgres-myapp-database-critical-op-pdb"),
+				hasMinAvailable(3),
+				testLabelsAndSelectors(false),
+			},
+		},
+		{
+			scenario: "With zero instances",
+			spec: New(
+				Config{OpConfig: config.Config{Resources: config.Resources{ClusterNameLabel: "cluster-name", PodRoleLabel: "spilo-role"}, PDBNameFormat: "postgres-{cluster}-pdb"}},
+				k8sutil.KubernetesClient{},
+				acidv1.Postgresql{
+					ObjectMeta: metav1.ObjectMeta{Name: "myapp-database", Namespace: "myapp"},
+					Spec:       acidv1.PostgresSpec{TeamID: "myapp", NumberOfInstances: 0}},
+				logger,
+				eventRecorder),
+			check: []func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error{
+				testPodDisruptionBudgetOwnerReference,
+				hasName("postgres-myapp-database-critical-op-pdb"),
+				hasMinAvailable(0),
+				testLabelsAndSelectors(false),
+			},
+		},
+		{
+			scenario: "With PodDisruptionBudget disabled",
+			spec: New(
+				Config{OpConfig: config.Config{Resources: config.Resources{ClusterNameLabel: "cluster-name", PodRoleLabel: "spilo-role"}, PDBNameFormat: "postgres-{cluster}-pdb", EnablePodDisruptionBudget: util.False()}},
+				k8sutil.KubernetesClient{},
+				acidv1.Postgresql{
+					ObjectMeta: metav1.ObjectMeta{Name: "myapp-database", Namespace: "myapp"},
+					Spec:       acidv1.PostgresSpec{TeamID: "myapp", NumberOfInstances: 3}},
+				logger,
+				eventRecorder),
+			check: []func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error{
+				testPodDisruptionBudgetOwnerReference,
+				hasName("postgres-myapp-database-critical-op-pdb"),
+				hasMinAvailable(0),
+				testLabelsAndSelectors(false),
+			},
+		},
+		{
+			scenario: "With OwnerReference enabled",
+			spec: New(
+				Config{OpConfig: config.Config{Resources: config.Resources{ClusterNameLabel: "cluster-name", PodRoleLabel: "spilo-role", EnableOwnerReferences: util.True()}, PDBNameFormat: "postgres-{cluster}-pdb", EnablePodDisruptionBudget: util.True()}},
+				k8sutil.KubernetesClient{},
+				acidv1.Postgresql{
+					ObjectMeta: metav1.ObjectMeta{Name: "myapp-database", Namespace: "myapp"},
+					Spec:       acidv1.PostgresSpec{TeamID: "myapp", NumberOfInstances: 3}},
+				logger,
+				eventRecorder),
+			check: []func(cluster *Cluster, podDisruptionBudget *policyv1.PodDisruptionBudget) error{
+				testPodDisruptionBudgetOwnerReference,
+				hasName("postgres-myapp-database-critical-op-pdb"),
+				hasMinAvailable(3),
+				testLabelsAndSelectors(false),
+			},
+		},
+	}
+
+	for _, tt := range testCriticalOp {
+		result := tt.spec.generateCriticalOpPodDisruptionBudget()
 		for _, check := range tt.check {
 			err := check(tt.spec, result)
 			if err != nil {

pkg/cluster/resources.go

@@ -23,8 +23,13 @@ const (
 )
 
 func (c *Cluster) listResources() error {
-	if c.PodDisruptionBudget != nil {
-		c.logger.Infof("found pod disruption budget: %q (uid: %q)", util.NameFromMeta(c.PodDisruptionBudget.ObjectMeta), c.PodDisruptionBudget.UID)
+	if c.PrimaryPodDisruptionBudget != nil {
+		c.logger.Infof("found primary pod disruption budget: %q (uid: %q)", util.NameFromMeta(c.PrimaryPodDisruptionBudget.ObjectMeta), c.PrimaryPodDisruptionBudget.UID)
+	}
+
+	if c.CriticalOpPodDisruptionBudget != nil {
+		c.logger.Infof("found pod disruption budget for critical operations: %q (uid: %q)", util.NameFromMeta(c.CriticalOpPodDisruptionBudget.ObjectMeta), c.CriticalOpPodDisruptionBudget.UID)
+
 	}
 
 	if c.Statefulset != nil {
@@ -417,59 +422,128 @@ func (c *Cluster) generateEndpointSubsets(role PostgresRole) []v1.EndpointSubset
 	return result
 }
 
-func (c *Cluster) createPodDisruptionBudget() (*policyv1.PodDisruptionBudget, error) {
-	podDisruptionBudgetSpec := c.generatePodDisruptionBudget()
+func (c *Cluster) createPrimaryPodDisruptionBudget() error {
+	c.logger.Debug("creating primary pod disruption budget")
+	if c.PrimaryPodDisruptionBudget != nil {
+		c.logger.Warning("primary pod disruption budget already exists in the cluster")
+		return nil
+	}
+
+	podDisruptionBudgetSpec := c.generatePrimaryPodDisruptionBudget()
 	podDisruptionBudget, err := c.KubeClient.
 		PodDisruptionBudgets(podDisruptionBudgetSpec.Namespace).
 		Create(context.TODO(), podDisruptionBudgetSpec, metav1.CreateOptions{})
 
 	if err != nil {
-		return nil, err
+		return err
 	}
-	c.PodDisruptionBudget = podDisruptionBudget
+	c.logger.Infof("primary pod disruption budget %q has been successfully created", util.NameFromMeta(podDisruptionBudget.ObjectMeta))
+	c.PrimaryPodDisruptionBudget = podDisruptionBudget
 
-	return podDisruptionBudget, nil
+	return nil
 }
 
-func (c *Cluster) updatePodDisruptionBudget(pdb *policyv1.PodDisruptionBudget) error {
-	if c.PodDisruptionBudget == nil {
-		return fmt.Errorf("there is no pod disruption budget in the cluster")
+func (c *Cluster) createCriticalOpPodDisruptionBudget() error {
+	c.logger.Debug("creating pod disruption budget for critical operations")
+	if c.CriticalOpPodDisruptionBudget != nil {
+		c.logger.Warning("pod disruption budget for critical operations already exists in the cluster")
+		return nil
 	}
 
-	if err := c.deletePodDisruptionBudget(); err != nil {
-		return fmt.Errorf("could not delete pod disruption budget: %v", err)
+	podDisruptionBudgetSpec := c.generateCriticalOpPodDisruptionBudget()
+	podDisruptionBudget, err := c.KubeClient.
+		PodDisruptionBudgets(podDisruptionBudgetSpec.Namespace).
+		Create(context.TODO(), podDisruptionBudgetSpec, metav1.CreateOptions{})
+
+	if err != nil {
+		return err
+	}
+	c.logger.Infof("pod disruption budget for critical operations %q has been successfully created", util.NameFromMeta(podDisruptionBudget.ObjectMeta))
+	c.CriticalOpPodDisruptionBudget = podDisruptionBudget
+
+	return nil
+}
+
+func (c *Cluster) createPodDisruptionBudgets() error {
+	errors := make([]string, 0)
+
+	err := c.createPrimaryPodDisruptionBudget()
+	if err != nil {
+		errors = append(errors, fmt.Sprintf("could not create primary pod disruption budget: %v", err))
+	}
+
+	err = c.createCriticalOpPodDisruptionBudget()
+	if err != nil {
+		errors = append(errors, fmt.Sprintf("could not create pod disruption budget for critical operations: %v", err))
+	}
+
+	if len(errors) > 0 {
+		return fmt.Errorf("%v", strings.Join(errors, `', '`))
+	}
+	return nil
+}
+
+func (c *Cluster) updatePrimaryPodDisruptionBudget(pdb *policyv1.PodDisruptionBudget) error {
+	c.logger.Debug("updating primary pod disruption budget")
+	if c.PrimaryPodDisruptionBudget == nil {
+		return fmt.Errorf("there is no primary pod disruption budget in the cluster")
+	}
+
+	if err := c.deletePrimaryPodDisruptionBudget(); err != nil {
+		return fmt.Errorf("could not delete primary pod disruption budget: %v", err)
 	}
 
 	newPdb, err := c.KubeClient.
 		PodDisruptionBudgets(pdb.Namespace).
 		Create(context.TODO(), pdb, metav1.CreateOptions{})
 	if err != nil {
-		return fmt.Errorf("could not create pod disruption budget: %v", err)
+		return fmt.Errorf("could not create primary pod disruption budget: %v", err)
 	}
-	c.PodDisruptionBudget = newPdb
+	c.PrimaryPodDisruptionBudget = newPdb
 
 	return nil
 }
 
-func (c *Cluster) deletePodDisruptionBudget() error {
-	c.logger.Debug("deleting pod disruption budget")
-	if c.PodDisruptionBudget == nil {
-		c.logger.Debug("there is no pod disruption budget in the cluster")
+func (c *Cluster) updateCriticalOpPodDisruptionBudget(pdb *policyv1.PodDisruptionBudget) error {
+	c.logger.Debug("updating pod disruption budget for critical operations")
+	if c.CriticalOpPodDisruptionBudget == nil {
+		return fmt.Errorf("there is no pod disruption budget for critical operations in the cluster")
+	}
+
+	if err := c.deleteCriticalOpPodDisruptionBudget(); err != nil {
+		return fmt.Errorf("could not delete pod disruption budget for critical operations: %v", err)
+	}
+
+	newPdb, err := c.KubeClient.
+		PodDisruptionBudgets(pdb.Namespace).
+		Create(context.TODO(), pdb, metav1.CreateOptions{})
+	if err != nil {
+		return fmt.Errorf("could not create pod disruption budget for critical operations: %v", err)
+	}
+	c.CriticalOpPodDisruptionBudget = newPdb
+
+	return nil
+}
+
+func (c *Cluster) deletePrimaryPodDisruptionBudget() error {
+	c.logger.Debug("deleting primary pod disruption budget")
+	if c.PrimaryPodDisruptionBudget == nil {
+		c.logger.Debug("there is no primary pod disruption budget in the cluster")
 		return nil
 	}
 
-	pdbName := util.NameFromMeta(c.PodDisruptionBudget.ObjectMeta)
+	pdbName := util.NameFromMeta(c.PrimaryPodDisruptionBudget.ObjectMeta)
 	err := c.KubeClient.
-		PodDisruptionBudgets(c.PodDisruptionBudget.Namespace).
-		Delete(context.TODO(), c.PodDisruptionBudget.Name, c.deleteOptions)
+		PodDisruptionBudgets(c.PrimaryPodDisruptionBudget.Namespace).
+		Delete(context.TODO(), c.PrimaryPodDisruptionBudget.Name, c.deleteOptions)
 	if k8sutil.ResourceNotFound(err) {
-		c.logger.Debugf("PodDisruptionBudget %q has already been deleted", util.NameFromMeta(c.PodDisruptionBudget.ObjectMeta))
+		c.logger.Debugf("PodDisruptionBudget %q has already been deleted", util.NameFromMeta(c.PrimaryPodDisruptionBudget.ObjectMeta))
 	} else if err != nil {
-		return fmt.Errorf("could not delete PodDisruptionBudget: %v", err)
+		return fmt.Errorf("could not delete primary pod disruption budget: %v", err)
 	}
 
-	c.logger.Infof("pod disruption budget %q has been deleted", util.NameFromMeta(c.PodDisruptionBudget.ObjectMeta))
-	c.PodDisruptionBudget = nil
+	c.logger.Infof("pod disruption budget %q has been deleted", util.NameFromMeta(c.PrimaryPodDisruptionBudget.ObjectMeta))
+	c.PrimaryPodDisruptionBudget = nil
 
 	err = retryutil.Retry(c.OpConfig.ResourceCheckInterval, c.OpConfig.ResourceCheckTimeout,
 		func() (bool, error) {
@@ -483,12 +557,67 @@ func (c *Cluster) deletePodDisruptionBudget() error {
 			return false, err2
 		})
 	if err != nil {
-		return fmt.Errorf("could not delete pod disruption budget: %v", err)
+		return fmt.Errorf("could not delete primary pod disruption budget: %v", err)
 	}
 
 	return nil
 }
 
+func (c *Cluster) deleteCriticalOpPodDisruptionBudget() error {
+	c.logger.Debug("deleting pod disruption budget for critical operations")
+	if c.CriticalOpPodDisruptionBudget == nil {
+		c.logger.Debug("there is no pod disruption budget for critical operations in the cluster")
+		return nil
+	}
+
+	pdbName := util.NameFromMeta(c.CriticalOpPodDisruptionBudget.ObjectMeta)
+	err := c.KubeClient.
+		PodDisruptionBudgets(c.CriticalOpPodDisruptionBudget.Namespace).
+		Delete(context.TODO(), c.CriticalOpPodDisruptionBudget.Name, c.deleteOptions)
+	if k8sutil.ResourceNotFound(err) {
+		c.logger.Debugf("PodDisruptionBudget %q has already been deleted", util.NameFromMeta(c.CriticalOpPodDisruptionBudget.ObjectMeta))
+	} else if err != nil {
+		return fmt.Errorf("could not delete pod disruption budget for critical operations: %v", err)
+	}
+
+	c.logger.Infof("pod disruption budget %q has been deleted", util.NameFromMeta(c.CriticalOpPodDisruptionBudget.ObjectMeta))
+	c.CriticalOpPodDisruptionBudget = nil
+
+	err = retryutil.Retry(c.OpConfig.ResourceCheckInterval, c.OpConfig.ResourceCheckTimeout,
+		func() (bool, error) {
+			_, err2 := c.KubeClient.PodDisruptionBudgets(pdbName.Namespace).Get(context.TODO(), pdbName.Name, metav1.GetOptions{})
+			if err2 == nil {
+				return false, nil
+			}
+			if k8sutil.ResourceNotFound(err2) {
+				return true, nil
+			}
+			return false, err2
+		})
+	if err != nil {
+		return fmt.Errorf("could not delete pod disruption budget for critical operations: %v", err)
+	}
+
+	return nil
+}
+
+func (c *Cluster) deletePodDisruptionBudgets() error {
+	errors := make([]string, 0)
+
+	if err := c.deletePrimaryPodDisruptionBudget(); err != nil {
+		errors = append(errors, fmt.Sprintf("%v", err))
+	}
+
+	if err := c.deleteCriticalOpPodDisruptionBudget(); err != nil {
+		errors = append(errors, fmt.Sprintf("%v", err))
+	}
+
+	if len(errors) > 0 {
+		return fmt.Errorf("%v", strings.Join(errors, `', '`))
+	}
+	return nil
+}
+
 func (c *Cluster) deleteEndpoint(role PostgresRole) error {
 	c.setProcessName("deleting endpoint")
 	c.logger.Debugf("deleting %s endpoint", role)
@@ -705,7 +834,12 @@ func (c *Cluster) GetStatefulSet() *appsv1.StatefulSet {
 	return c.Statefulset
 }
 
-// GetPodDisruptionBudget returns cluster's kubernetes PodDisruptionBudget
-func (c *Cluster) GetPodDisruptionBudget() *policyv1.PodDisruptionBudget {
-	return c.PodDisruptionBudget
+// GetPrimaryPodDisruptionBudget returns cluster's primary kubernetes PodDisruptionBudget
+func (c *Cluster) GetPrimaryPodDisruptionBudget() *policyv1.PodDisruptionBudget {
+	return c.PrimaryPodDisruptionBudget
+}
+
+// GetCriticalOpPodDisruptionBudget returns cluster's kubernetes PodDisruptionBudget for critical operations
+func (c *Cluster) GetCriticalOpPodDisruptionBudget() *policyv1.PodDisruptionBudget {
+	return c.CriticalOpPodDisruptionBudget
 }

pkg/cluster/sync.go

@@ -117,8 +117,8 @@ func (c *Cluster) Sync(newSpec *acidv1.Postgresql) error {
 	}
 
 	c.logger.Debug("syncing pod disruption budgets")
-	if err = c.syncPodDisruptionBudget(false); err != nil {
-		err = fmt.Errorf("could not sync pod disruption budget: %v", err)
+	if err = c.syncPodDisruptionBudgets(false); err != nil {
+		err = fmt.Errorf("could not sync pod disruption budgets: %v", err)
 		return err
 	}
 
@@ -452,22 +452,22 @@ func (c *Cluster) syncEndpoint(role PostgresRole) error {
 	return nil
 }
 
-func (c *Cluster) syncPodDisruptionBudget(isUpdate bool) error {
+func (c *Cluster) syncPrimaryPodDisruptionBudget(isUpdate bool) error {
 	var (
 		pdb *policyv1.PodDisruptionBudget
 		err error
 	)
-	if pdb, err = c.KubeClient.PodDisruptionBudgets(c.Namespace).Get(context.TODO(), c.podDisruptionBudgetName(), metav1.GetOptions{}); err == nil {
-		c.PodDisruptionBudget = pdb
-		newPDB := c.generatePodDisruptionBudget()
+	if pdb, err = c.KubeClient.PodDisruptionBudgets(c.Namespace).Get(context.TODO(), c.PrimaryPodDisruptionBudgetName(), metav1.GetOptions{}); err == nil {
+		c.PrimaryPodDisruptionBudget = pdb
+		newPDB := c.generatePrimaryPodDisruptionBudget()
 		match, reason := c.comparePodDisruptionBudget(pdb, newPDB)
 		if !match {
 			c.logPDBChanges(pdb, newPDB, isUpdate, reason)
-			if err = c.updatePodDisruptionBudget(newPDB); err != nil {
+			if err = c.updatePrimaryPodDisruptionBudget(newPDB); err != nil {
 				return err
 			}
 		} else {
-			c.PodDisruptionBudget = pdb
+			c.PrimaryPodDisruptionBudget = pdb
 		}
 		return nil
 
@@ -476,21 +476,74 @@ func (c *Cluster) syncPodDisruptionBudget(isUpdate bool) error {
 		return fmt.Errorf("could not get pod disruption budget: %v", err)
 	}
 	// no existing pod disruption budget, create new one
-	c.logger.Infof("could not find the cluster's pod disruption budget")
+	c.logger.Infof("could not find the primary pod disruption budget")
 
-	if pdb, err = c.createPodDisruptionBudget(); err != nil {
+	if err = c.createPrimaryPodDisruptionBudget(); err != nil {
 		if !k8sutil.ResourceAlreadyExists(err) {
-			return fmt.Errorf("could not create pod disruption budget: %v", err)
+			return fmt.Errorf("could not create primary pod disruption budget: %v", err)
 		}
 		c.logger.Infof("pod disruption budget %q already exists", util.NameFromMeta(pdb.ObjectMeta))
-		if pdb, err = c.KubeClient.PodDisruptionBudgets(c.Namespace).Get(context.TODO(), c.podDisruptionBudgetName(), metav1.GetOptions{}); err != nil {
+		if pdb, err = c.KubeClient.PodDisruptionBudgets(c.Namespace).Get(context.TODO(), c.PrimaryPodDisruptionBudgetName(), metav1.GetOptions{}); err != nil {
 			return fmt.Errorf("could not fetch existing %q pod disruption budget", util.NameFromMeta(pdb.ObjectMeta))
 		}
 	}
 
-	c.logger.Infof("created missing pod disruption budget %q", util.NameFromMeta(pdb.ObjectMeta))
-	c.PodDisruptionBudget = pdb
+	return nil
+}
 
+func (c *Cluster) syncCriticalOpPodDisruptionBudget(isUpdate bool) error {
+	var (
+		pdb *policyv1.PodDisruptionBudget
+		err error
+	)
+	if pdb, err = c.KubeClient.PodDisruptionBudgets(c.Namespace).Get(context.TODO(), c.criticalOpPodDisruptionBudgetName(), metav1.GetOptions{}); err == nil {
+		c.CriticalOpPodDisruptionBudget = pdb
+		newPDB := c.generateCriticalOpPodDisruptionBudget()
+		match, reason := c.comparePodDisruptionBudget(pdb, newPDB)
+		if !match {
+			c.logPDBChanges(pdb, newPDB, isUpdate, reason)
+			if err = c.updateCriticalOpPodDisruptionBudget(newPDB); err != nil {
+				return err
+			}
+		} else {
+			c.CriticalOpPodDisruptionBudget = pdb
+		}
+		return nil
+
+	}
+	if !k8sutil.ResourceNotFound(err) {
+		return fmt.Errorf("could not get pod disruption budget: %v", err)
+	}
+	// no existing pod disruption budget, create new one
+	c.logger.Infof("could not find pod disruption budget for critical operations")
+
+	if err = c.createCriticalOpPodDisruptionBudget(); err != nil {
+		if !k8sutil.ResourceAlreadyExists(err) {
+			return fmt.Errorf("could not create pod disruption budget for critical operations: %v", err)
+		}
+		c.logger.Infof("pod disruption budget %q already exists", util.NameFromMeta(pdb.ObjectMeta))
+		if pdb, err = c.KubeClient.PodDisruptionBudgets(c.Namespace).Get(context.TODO(), c.criticalOpPodDisruptionBudgetName(), metav1.GetOptions{}); err != nil {
+			return fmt.Errorf("could not fetch existing %q pod disruption budget", util.NameFromMeta(pdb.ObjectMeta))
+		}
+	}
+
+	return nil
+}
+
+func (c *Cluster) syncPodDisruptionBudgets(isUpdate bool) error {
+	errors := make([]string, 0)
+
+	if err := c.syncPrimaryPodDisruptionBudget(isUpdate); err != nil {
+		errors = append(errors, fmt.Sprintf("%v", err))
+	}
+
+	if err := c.syncCriticalOpPodDisruptionBudget(isUpdate); err != nil {
+		errors = append(errors, fmt.Sprintf("%v", err))
+	}
+
+	if len(errors) > 0 {
+		return fmt.Errorf("%v", strings.Join(errors, `', '`))
+	}
 	return nil
 }
 

pkg/cluster/types.go

@@ -58,15 +58,16 @@ type WorkerStatus struct {
 
 // ClusterStatus describes status of the cluster
 type ClusterStatus struct {
-	Team                string
-	Cluster             string
-	Namespace           string
-	MasterService       *v1.Service
-	ReplicaService      *v1.Service
-	MasterEndpoint      *v1.Endpoints
-	ReplicaEndpoint     *v1.Endpoints
-	StatefulSet         *appsv1.StatefulSet
-	PodDisruptionBudget *policyv1.PodDisruptionBudget
+	Team                          string
+	Cluster                       string
+	Namespace                     string
+	MasterService                 *v1.Service
+	ReplicaService                *v1.Service
+	MasterEndpoint                *v1.Endpoints
+	ReplicaEndpoint               *v1.Endpoints
+	StatefulSet                   *appsv1.StatefulSet
+	PrimaryPodDisruptionBudget    *policyv1.PodDisruptionBudget
+	CriticalOpPodDisruptionBudget *policyv1.PodDisruptionBudget
 
 	CurrentProcess Process
 	Worker         uint32

pkg/cluster/util_test.go

@@ -329,7 +329,7 @@ func newInheritedAnnotationsCluster(client k8sutil.KubernetesClient) (*Cluster,
 	if err != nil {
 		return nil, err
 	}
-	_, err = cluster.createPodDisruptionBudget()
+	err = cluster.createPodDisruptionBudgets()
 	if err != nil {
 		return nil, err
 	}