Make Patroni TTL, LoopWait and RetryTimeout configurable (#385)

* Make Patroni TTL, LoopWait and RetryTimeout configurable, modify default values

* Update tests

* hardcode patroni loopwait, calculate patroni ttl

Author: eberlep

api/v1/postgres_types.go

@@ -63,6 +63,9 @@ const (
 
 	teamIDPrefix = "pg"
 
+	DefaultPatroniParamValueLoopWait     uint32 = 10
+	DefaultPatroniParamValueRetryTimeout uint32 = 60
+
 	defaultPostgresParamValueTCPKeepAlivesIdle      = "200"
 	defaultPostgresParamValueTCPKeepAlivesInterval  = "30"
 	defaultPostgresParamValueLogFileMode            = "0600"
@@ -482,7 +485,7 @@ func (p *Postgres) ToPeripheralResourceLookupKey() types.NamespacedName {
 	}
 }
 
-func (p *Postgres) ToUnstructuredZalandoPostgresql(z *zalando.Postgresql, c *corev1.ConfigMap, sc string, pgParamBlockList map[string]bool, rbs *BackupConfig, srcDB *Postgres) (*unstructured.Unstructured, error) {
+func (p *Postgres) ToUnstructuredZalandoPostgresql(z *zalando.Postgresql, c *corev1.ConfigMap, sc string, pgParamBlockList map[string]bool, rbs *BackupConfig, srcDB *Postgres, patroniTTL, patroniLoopWait, patroniRetryTimeout uint32) (*unstructured.Unstructured, error) {
 	if z == nil {
 		z = &zalando.Postgresql{}
 	}
@@ -515,8 +518,9 @@ func (p *Postgres) ToUnstructuredZalandoPostgresql(z *zalando.Postgresql, c *cor
 	z.Spec.Volume.Size = p.Spec.Size.StorageSize
 	z.Spec.Volume.StorageClass = sc
 
-	// TODO make configurable?
-	z.Spec.Patroni.TTL = 130
+	z.Spec.Patroni.TTL = patroniTTL
+	z.Spec.Patroni.LoopWait = patroniLoopWait
+	z.Spec.Patroni.RetryTimeout = patroniRetryTimeout
 	z.Spec.Patroni.SynchronousMode = true
 	z.Spec.Patroni.SynchronousModeStrict = false
 

api/v1/postgres_types_test.go

@@ -349,7 +349,7 @@ func TestPostgresRestoreTimestamp_ToUnstructuredZalandoPostgresql(t *testing.T)
 			p := &Postgres{
 				Spec: tt.spec,
 			}
-			got, _ := p.ToUnstructuredZalandoPostgresql(nil, tt.c, tt.sc, tt.pgParamBlockList, tt.rbs, tt.srcDB)
+			got, _ := p.ToUnstructuredZalandoPostgresql(nil, tt.c, tt.sc, tt.pgParamBlockList, tt.rbs, tt.srcDB, 130, 10, 60)
 
 			jsonZ, err := runtime.DefaultUnstructuredConverter.ToUnstructured(got)
 			if err != nil {

controllers/postgres_controller.go

@@ -71,6 +71,9 @@ type PostgresReconciler struct {
 	SidecarsConfigMapName       string
 	EnableNetPol                bool
 	EtcdHost                    string
+	PatroniTTL                  uint32
+	PatroniLoopWait             uint32
+	PatroniRetryTimeout         uint32
 }
 
 // Reconcile is the entry point for postgres reconciliation.
@@ -247,7 +250,7 @@ func (r *PostgresReconciler) Reconcile(ctx context.Context, req ctrl.Request) (c
 		return ctrl.Result{}, fmt.Errorf("error while creating sidecars servicemonitor %v: %w", namespace, err)
 	}
 
-	if err := r.createOrUpdateZalandoPostgresql(ctx, instance, log, globalSidecarsCM); err != nil {
+	if err := r.createOrUpdateZalandoPostgresql(ctx, instance, log, globalSidecarsCM, r.PatroniTTL, r.PatroniLoopWait, r.PatroniRetryTimeout); err != nil {
 		r.recorder.Eventf(instance, "Warning", "Error", "failed to create Zalando resource: %v", err)
 		return ctrl.Result{}, fmt.Errorf("failed to create or update zalando postgresql: %w", err)
 	}
@@ -288,7 +291,7 @@ func (r *PostgresReconciler) SetupWithManager(mgr ctrl.Manager) error {
 		Complete(r)
 }
 
-func (r *PostgresReconciler) createOrUpdateZalandoPostgresql(ctx context.Context, instance *pg.Postgres, log logr.Logger, sidecarsCM *corev1.ConfigMap) error {
+func (r *PostgresReconciler) createOrUpdateZalandoPostgresql(ctx context.Context, instance *pg.Postgres, log logr.Logger, sidecarsCM *corev1.ConfigMap, patroniTTL, patroniLoopWait, patroniRetryTimout uint32) error {
 	var restoreBackupConfig *pg.BackupConfig
 	var restoreSouceInstance *pg.Postgres
 	if instance.Spec.PostgresRestore != nil {
@@ -323,7 +326,7 @@ func (r *PostgresReconciler) createOrUpdateZalandoPostgresql(ctx context.Context
 			return fmt.Errorf("failed to fetch zalando postgresql: %w", err)
 		}
 
-		u, err := instance.ToUnstructuredZalandoPostgresql(nil, sidecarsCM, r.StorageClass, r.PgParamBlockList, restoreBackupConfig, restoreSouceInstance)
+		u, err := instance.ToUnstructuredZalandoPostgresql(nil, sidecarsCM, r.StorageClass, r.PgParamBlockList, restoreBackupConfig, restoreSouceInstance, patroniTTL, patroniLoopWait, patroniRetryTimout)
 		if err != nil {
 			return fmt.Errorf("failed to convert to unstructured zalando postgresql: %w", err)
 		}
@@ -339,7 +342,7 @@ func (r *PostgresReconciler) createOrUpdateZalandoPostgresql(ctx context.Context
 	// Update zalando postgresql
 	mergeFrom := client.MergeFrom(rawZ.DeepCopy())
 
-	u, err := instance.ToUnstructuredZalandoPostgresql(rawZ, sidecarsCM, r.StorageClass, r.PgParamBlockList, restoreBackupConfig, restoreSouceInstance)
+	u, err := instance.ToUnstructuredZalandoPostgresql(rawZ, sidecarsCM, r.StorageClass, r.PgParamBlockList, restoreBackupConfig, restoreSouceInstance, patroniTTL, patroniLoopWait, patroniRetryTimout)
 	if err != nil {
 		return fmt.Errorf("failed to convert to unstructured zalando postgresql: %w", err)
 	}

main.go

@@ -57,6 +57,7 @@ const (
 	sidecarsCMNameFlg              = "sidecars-configmap-name"
 	enableNetPolFlg                = "enable-netpol"
 	enablePodAntiaffinityFlg       = "enable-pod-antiaffinity"
+	patroniRetryTimeoutFlg         = "patroni-retry-timeout"
 )
 
 var (
@@ -78,6 +79,7 @@ func main() {
 	var metricsAddrCtrlMgr, metricsAddrSvcMgr, partitionID, tenant, ctrlClusterKubeconfig, pspName, lbIP, storageClass, postgresImage, etcdHost, operatorImage, majorVersionUpgradeMode, postgresletNamespace, sidecarsCMName string
 	var enableLeaderElection, enableCRDValidation, enableNetPol, enablePodAntiaffinity bool
 	var portRangeStart, portRangeSize int
+	var patroniTTL, patroniLoopWait, patroniRetryTimeout uint32
 	var pgParamBlockList map[string]bool
 	var standbyClusterSourceRanges []string
 
@@ -159,6 +161,16 @@ func main() {
 	viper.SetDefault(enablePodAntiaffinityFlg, false)
 	enablePodAntiaffinity = viper.GetBool(enablePodAntiaffinityFlg)
 
+	// hard coded value
+	patroniLoopWait = databasev1.DefaultPatroniParamValueLoopWait
+
+	// user defined value
+	viper.SetDefault(patroniRetryTimeoutFlg, databasev1.DefaultPatroniParamValueRetryTimeout)
+	patroniRetryTimeout = viper.GetUint32(patroniRetryTimeoutFlg)
+
+	// derived value
+	patroniTTL = (2 * patroniRetryTimeout) + patroniLoopWait
+
 	ctrl.SetLogger(zap.New(zap.UseDevMode(true)))
 
 	ctrl.Log.Info("flag",
@@ -184,6 +196,7 @@ func main() {
 		sidecarsCMNameFlg, sidecarsCMName,
 		enableNetPolFlg, enableNetPol,
 		enablePodAntiaffinityFlg, enablePodAntiaffinity,
+		patroniRetryTimeoutFlg, patroniRetryTimeout,
 	)
 
 	svcClusterConf := ctrl.GetConfigOrDie()
@@ -254,6 +267,9 @@ func main() {
 		SidecarsConfigMapName:       sidecarsCMName,
 		EnableNetPol:                enableNetPol,
 		EtcdHost:                    etcdHost,
+		PatroniTTL:                  patroniTTL,
+		PatroniLoopWait:             patroniLoopWait,
+		PatroniRetryTimeout:         patroniRetryTimeout,
 	}).SetupWithManager(ctrlPlaneClusterMgr); err != nil {
 		setupLog.Error(err, "unable to create controller", "controller", "Postgres")
 		os.Exit(1)