Broaden Svc Selector (#395)

* Broaden selector in service

* Force single instance for standby

* Revert "Force single instance for standby"

This reverts commit 6a8c229.

* Select the first pod for standby dbs, ignore spilo-role

* Remove incorrect comment

* Make the use of the existing selector via a new feature flag

* Cleanup pod-name from selector

* Use const instead of string literal

* Properly clean up the selector in database switch

* reduce, reuse, recycle

* move new var into own line

Author: eberlep

api/v1/postgres_types.go

@@ -68,6 +68,7 @@ const (
 	SpiloRoleLabelName               = "spilo-role"
 	SpiloRoleLabelValueMaster        = "master"
 	SpiloRoleLabelValueStandbyLeader = "standby_leader"
+	StatefulsetPodNameLabelName      = "statefulset.kubernetes.io/pod-name"
 
 	teamIDPrefix = "pg"
 
@@ -321,7 +322,7 @@ func (p *Postgres) ToKey() *types.NamespacedName {
 	}
 }
 
-func (p *Postgres) ToSvcLB(lbIP string, lbPort int32, enableStandbyLeaderSelector bool) *corev1.Service {
+func (p *Postgres) ToSvcLB(lbIP string, lbPort int32, enableStandbyLeaderSelector bool, enableLegacyStandbySelector bool) *corev1.Service {
 	lb := &corev1.Service{}
 	lb.Spec.Type = "LoadBalancer"
 
@@ -342,16 +343,24 @@ func (p *Postgres) ToSvcLB(lbIP string, lbPort int32, enableStandbyLeaderSelecto
 	port.TargetPort = intstr.FromInt(5432)
 	lb.Spec.Ports = []corev1.ServicePort{port}
 
-	spiloRole := SpiloRoleLabelValueMaster
-	if enableStandbyLeaderSelector && !p.IsReplicationPrimary() {
-		spiloRole = SpiloRoleLabelValueStandbyLeader
-	}
 	lb.Spec.Selector = map[string]string{
 		ApplicationLabelName: ApplicationLabelValue,
 		"cluster-name":       p.ToPeripheralResourceName(),
-		SpiloRoleLabelName:   spiloRole,
 		"team":               p.generateTeamID(),
 	}
+	if p.IsReplicationPrimary() {
+		lb.Spec.Selector[SpiloRoleLabelName] = SpiloRoleLabelValueMaster
+	} else {
+		if enableStandbyLeaderSelector {
+			// Only set this value when we are NOT a primary and the StandbyLeaderSelector is enabled.
+			lb.Spec.Selector[SpiloRoleLabelName] = SpiloRoleLabelValueStandbyLeader
+		} else if enableLegacyStandbySelector {
+			lb.Spec.Selector[SpiloRoleLabelName] = SpiloRoleLabelValueMaster
+		} else {
+			// select the first pod in the statefulset
+			lb.Spec.Selector[StatefulsetPodNameLabelName] = p.ToPeripheralResourceName() + "-0"
+		}
+	}
 
 	if len(lbIP) > 0 {
 		// if no ip is set, a new loadbalancer will be created automatically

main.go

@@ -61,6 +61,7 @@ const (
 	patroniRetryTimeoutFlg         = "patroni-retry-timeout"
 	enableStandbyLeaderSelectorFlg = "enable-standby-leader-selector"
 	ControlPlaneNamespaceFlg       = "control-plane-namespace"
+	enableLegacyStandbySelectorFlg = "enable-legacy-standby-selector"
 	deployEtcdFlg                  = "deploy-etcd"
 	etcdImageFlg                   = "etcd-image"
 	etcdBackupSidecarImageFlg      = "etcd-backup-sidecar-image"
@@ -85,6 +86,7 @@ func init() {
 }
 
 func main() {
+
 	var (
 		metricsAddrCtrlMgr      string
 		metricsAddrSvcMgr       string
@@ -112,6 +114,7 @@ func main() {
 		enableNetPol                bool
 		enablePodAntiaffinity       bool
 		enableStandbyLeaderSelector bool
+		enableLegacyStandbySelector bool
 		deployEtcd                  bool
 
 		portRangeStart int
@@ -220,6 +223,9 @@ func main() {
 	viper.SetDefault(ControlPlaneNamespaceFlg, "metal-extension-postgres")
 	controlPlaneNamespace = viper.GetString(ControlPlaneNamespaceFlg)
 
+	viper.SetDefault(enableLegacyStandbySelectorFlg, false)
+	enableLegacyStandbySelector = viper.GetBool(enableLegacyStandbySelectorFlg)
+
 	viper.SetDefault(deployEtcdFlg, false)
 	deployEtcd = viper.GetBool(deployEtcdFlg)
 
@@ -262,6 +268,7 @@ func main() {
 		patroniRetryTimeoutFlg, patroniRetryTimeout,
 		enableStandbyLeaderSelectorFlg, enableStandbyLeaderSelector,
 		ControlPlaneNamespaceFlg, controlPlaneNamespace,
+		enableLegacyStandbySelectorFlg, enableLegacyStandbySelector,
 		deployEtcdFlg, deployEtcd,
 		etcdImageFlg, etcdImage,
 		etcdBackupSidecarImageFlg, etcdBackupSidecarImage,
@@ -348,6 +355,7 @@ func main() {
 		PortRangeStart:              int32(portRangeStart),
 		PortRangeSize:               int32(portRangeSize),
 		EnableStandbyLeaderSelector: enableStandbyLeaderSelector,
+		EnableLegacyStandbySelector: enableLegacyStandbySelector,
 	}
 	if err = (&controllers.PostgresReconciler{
 		CtrlClient:                  ctrlPlaneClusterMgr.GetClient(),

pkg/lbmanager/lbmanager.go

@@ -16,6 +16,7 @@ type Options struct {
 	PortRangeStart              int32
 	PortRangeSize               int32
 	EnableStandbyLeaderSelector bool
+	EnableLegacyStandbySelector bool
 }
 
 // LBManager Responsible for the creation and deletion of externally accessible Services to access the Postgresql clusters managed by the Postgreslet.
@@ -59,18 +60,14 @@ func (m *LBManager) CreateSvcLBIfNone(ctx context.Context, in *api.Postgres) err
 			lbIPToUse = ""
 		}
 
-		if err := m.Create(ctx, in.ToSvcLB(lbIPToUse, nextFreePort, m.options.EnableStandbyLeaderSelector)); err != nil {
+		if err := m.Create(ctx, in.ToSvcLB(lbIPToUse, nextFreePort, m.options.EnableStandbyLeaderSelector, m.options.EnableLegacyStandbySelector)); err != nil {
 			return fmt.Errorf("failed to create Service of type LoadBalancer: %w", err)
 		}
 		return nil
 	}
 
 	// update the selector, and only the selector (we do NOT want the change the ip or port here!!!)
-	if m.options.EnableStandbyLeaderSelector && !in.IsReplicationPrimary() {
-		svc.Spec.Selector[api.SpiloRoleLabelName] = api.SpiloRoleLabelValueStandbyLeader
-	} else {
-		svc.Spec.Selector[api.SpiloRoleLabelName] = api.SpiloRoleLabelValueMaster
-	}
+	svc.Spec.Selector = in.ToSvcLB("", 0, m.options.EnableStandbyLeaderSelector, m.options.EnableLegacyStandbySelector).Spec.Selector
 
 	if err := m.Update(ctx, svc); err != nil {
 		return fmt.Errorf("failed to update Service of type LoadBalancer: %w", err)