Merge pull request 'Add a primary (write) Service routing to pod-0' (#7) from benvin/primary-write-service into main

Reviewed-on: #7

api/v1alpha1/bindcluster_types.go

@@ -33,6 +33,13 @@ type ClusterServiceSpec struct {
 	// +optional
 	LoadBalancerIP string `json:"loadBalancerIP,omitempty"`
 
+	// ExternalTrafficPolicy for a LoadBalancer/NodePort Service. Local preserves
+	// client source IPs (required for source-IP ACLs on the DNS servers) but
+	// only routes to nodes running a pod. Defaults to Cluster.
+	// +kubebuilder:validation:Enum=Cluster;Local
+	// +optional
+	ExternalTrafficPolicy corev1.ServiceExternalTrafficPolicy `json:"externalTrafficPolicy,omitempty"`
+
 	// Annotations added to the client-facing Service (e.g. PureLB/MetalLB hints).
 	// +optional
 	Annotations map[string]string `json:"annotations,omitempty"`
@@ -98,10 +105,18 @@ type BindClusterSpec struct {
 	// +optional
 	Resources corev1.ResourceRequirements `json:"resources,omitempty"`
 
-	// Service controls how the cluster is exposed.
+	// Service controls how the cluster is exposed for reads (all pods).
 	// +optional
 	Service ClusterServiceSpec `json:"service,omitempty"`
 
+	// PrimaryService, when set, creates an additional Service routing only to the
+	// primary pod (ordinal 0) — the write endpoint for RFC2136/nsupdate, since
+	// secondaries reject updates. Reads still use Service (all pods). Use
+	// ClusterIP for in-cluster writers (e.g. external-dns) or LoadBalancer for
+	// external writers.
+	// +optional
+	PrimaryService *ClusterServiceSpec `json:"primaryService,omitempty"`
+
 	// NodeSelector for the BIND pods.
 	// +optional
 	NodeSelector map[string]string `json:"nodeSelector,omitempty"`

api/v1alpha1/bindtsigkey_types.go

@@ -16,6 +16,12 @@ const (
 // BindTSIGKeySpec defines a TSIG key. If no existing key material is imported,
 // the operator generates a random key and stores it in a Secret.
 type BindTSIGKeySpec struct {
+	// ClusterRef names the BindCluster this key is included in. When empty the
+	// key is shared with every cluster in the namespace (useful when multiple
+	// clusters share one namespace).
+	// +optional
+	ClusterRef string `json:"clusterRef,omitempty"`
+
 	// Algorithm is the HMAC algorithm. Defaults to hmac-sha256.
 	// +kubebuilder:default="hmac-sha256"
 	// +optional

api/v1alpha1/zz_generated.deepcopy.go

@@ -301,6 +301,11 @@ func (in *BindClusterSpec) DeepCopyInto(out *BindClusterSpec) {
 	}
 	in.Resources.DeepCopyInto(&out.Resources)
 	in.Service.DeepCopyInto(&out.Service)
+	if in.PrimaryService != nil {
+		in, out := &in.PrimaryService, &out.PrimaryService
+		*out = new(ClusterServiceSpec)
+		(*in).DeepCopyInto(*out)
+	}
 	if in.NodeSelector != nil {
 		in, out := &in.NodeSelector, &out.NodeSelector
 		*out = make(map[string]string, len(*in))

config/crd/bases/bind.unkin.net_bindclusters.yaml

@@ -1013,6 +1013,41 @@ spec:
                   type: string
                 description: NodeSelector for the BIND pods.
                 type: object
+              primaryService:
+                description: |-
+                  PrimaryService, when set, creates an additional Service routing only to the
+                  primary pod (ordinal 0) — the write endpoint for RFC2136/nsupdate, since
+                  secondaries reject updates. Reads still use Service (all pods). Use
+                  ClusterIP for in-cluster writers (e.g. external-dns) or LoadBalancer for
+                  external writers.
+                properties:
+                  annotations:
+                    additionalProperties:
+                      type: string
+                    description: Annotations added to the client-facing Service (e.g.
+                      PureLB/MetalLB hints).
+                    type: object
+                  externalTrafficPolicy:
+                    description: |-
+                      ExternalTrafficPolicy for a LoadBalancer/NodePort Service. Local preserves
+                      client source IPs (required for source-IP ACLs on the DNS servers) but
+                      only routes to nodes running a pod. Defaults to Cluster.
+                    enum:
+                    - Cluster
+                    - Local
+                    type: string
+                  loadBalancerIP:
+                    description: LoadBalancerIP requests a specific address when Type
+                      is LoadBalancer.
+                    type: string
+                  type:
+                    description: Type of the client-facing Service. Defaults to ClusterIP.
+                    enum:
+                    - ClusterIP
+                    - LoadBalancer
+                    - NodePort
+                    type: string
+                type: object
               recursion:
                 description: |-
                   Recursion overrides the default per-mode recursion setting. When nil,
@@ -1086,7 +1121,8 @@ spec:
                     type: object
                 type: object
               service:
-                description: Service controls how the cluster is exposed.
+                description: Service controls how the cluster is exposed for reads
+                  (all pods).
                 properties:
                   annotations:
                     additionalProperties:
@@ -1094,6 +1130,15 @@ spec:
                     description: Annotations added to the client-facing Service (e.g.
                       PureLB/MetalLB hints).
                     type: object
+                  externalTrafficPolicy:
+                    description: |-
+                      ExternalTrafficPolicy for a LoadBalancer/NodePort Service. Local preserves
+                      client source IPs (required for source-IP ACLs on the DNS servers) but
+                      only routes to nodes running a pod. Defaults to Cluster.
+                    enum:
+                    - Cluster
+                    - Local
+                    type: string
                   loadBalancerIP:
                     description: LoadBalancerIP requests a specific address when Type
                       is LoadBalancer.

config/crd/bases/bind.unkin.net_bindtsigkeys.yaml

@@ -66,6 +66,12 @@ spec:
                 - hmac-sha1
                 - hmac-md5
                 type: string
+              clusterRef:
+                description: |-
+                  ClusterRef names the BindCluster this key is included in. When empty the
+                  key is shared with every cluster in the namespace (useful when multiple
+                  clusters share one namespace).
+                type: string
               importExisting:
                 description: |-
                   ImportExisting, when true, means the referenced Secret already contains a

config/crd/install.yaml

@@ -1318,6 +1318,41 @@ spec:
                   type: string
                 description: NodeSelector for the BIND pods.
                 type: object
+              primaryService:
+                description: |-
+                  PrimaryService, when set, creates an additional Service routing only to the
+                  primary pod (ordinal 0) — the write endpoint for RFC2136/nsupdate, since
+                  secondaries reject updates. Reads still use Service (all pods). Use
+                  ClusterIP for in-cluster writers (e.g. external-dns) or LoadBalancer for
+                  external writers.
+                properties:
+                  annotations:
+                    additionalProperties:
+                      type: string
+                    description: Annotations added to the client-facing Service (e.g.
+                      PureLB/MetalLB hints).
+                    type: object
+                  externalTrafficPolicy:
+                    description: |-
+                      ExternalTrafficPolicy for a LoadBalancer/NodePort Service. Local preserves
+                      client source IPs (required for source-IP ACLs on the DNS servers) but
+                      only routes to nodes running a pod. Defaults to Cluster.
+                    enum:
+                    - Cluster
+                    - Local
+                    type: string
+                  loadBalancerIP:
+                    description: LoadBalancerIP requests a specific address when Type
+                      is LoadBalancer.
+                    type: string
+                  type:
+                    description: Type of the client-facing Service. Defaults to ClusterIP.
+                    enum:
+                    - ClusterIP
+                    - LoadBalancer
+                    - NodePort
+                    type: string
+                type: object
               recursion:
                 description: |-
                   Recursion overrides the default per-mode recursion setting. When nil,
@@ -1391,7 +1426,8 @@ spec:
                     type: object
                 type: object
               service:
-                description: Service controls how the cluster is exposed.
+                description: Service controls how the cluster is exposed for reads
+                  (all pods).
                 properties:
                   annotations:
                     additionalProperties:
@@ -1399,6 +1435,15 @@ spec:
                     description: Annotations added to the client-facing Service (e.g.
                       PureLB/MetalLB hints).
                     type: object
+                  externalTrafficPolicy:
+                    description: |-
+                      ExternalTrafficPolicy for a LoadBalancer/NodePort Service. Local preserves
+                      client source IPs (required for source-IP ACLs on the DNS servers) but
+                      only routes to nodes running a pod. Defaults to Cluster.
+                    enum:
+                    - Cluster
+                    - Local
+                    type: string
                   loadBalancerIP:
                     description: LoadBalancerIP requests a specific address when Type
                       is LoadBalancer.
@@ -2043,6 +2088,12 @@ spec:
                 - hmac-sha1
                 - hmac-md5
                 type: string
+              clusterRef:
+                description: |-
+                  ClusterRef names the BindCluster this key is included in. When empty the
+                  key is shared with every cluster in the namespace (useful when multiple
+                  clusters share one namespace).
+                type: string
               importExisting:
                 description: |-
                   ImportExisting, when true, means the referenced Secret already contains a

internal/bind/render.go

@@ -16,6 +16,10 @@ type RenderInput struct {
 	Policies       []bindv1alpha1.BindPolicy
 	DNSSECPolicies []bindv1alpha1.BindDNSSECPolicy
 	Catalog        *bindv1alpha1.BindCatalogZone
+	// Forwards are type:forward BindZones. They are pure configuration (no
+	// replicated data), so they are rendered into named.conf on every pod
+	// rather than added dynamically to the primary.
+	Forwards []bindv1alpha1.BindZone
 	// PrimaryAddress is the in-cluster address secondaries transfer from.
 	PrimaryAddress string
 }
@@ -90,6 +94,29 @@ func render(in RenderInput, isPrimary bool) string {
 		b.WriteString(renderCatalogZoneDecl(in, isPrimary, ""))
 	}
 
+	// Top-level forward zones (BIND only allows top-level zones when no views
+	// are defined; in-view forward zones are rendered inside renderView).
+	if len(in.Views) == 0 {
+		for _, z := range in.Forwards {
+			if z.Spec.ViewRef == "" {
+				b.WriteString(renderForwardZone(z, ""))
+			}
+		}
+	}
+
+	return b.String()
+}
+
+// renderForwardZone renders a type:forward zone clause.
+func renderForwardZone(z bindv1alpha1.BindZone, indent string) string {
+	var b strings.Builder
+	b.WriteString(fmt.Sprintf("%szone \"%s\" {\n", indent, z.Spec.ZoneName))
+	b.WriteString(indent + "    type forward;\n")
+	b.WriteString(indent + "    forward only;\n")
+	if len(z.Spec.Forwarders) > 0 {
+		b.WriteString(fmt.Sprintf("%s    forwarders { %s };\n", indent, terminate(z.Spec.Forwarders)))
+	}
+	b.WriteString(indent + "};\n")
 	return b.String()
 }
 
@@ -122,6 +149,12 @@ func renderView(v bindv1alpha1.BindView, in RenderInput, isPrimary bool) string
 	if in.Catalog != nil {
 		b.WriteString(renderCatalogZoneDecl(in, isPrimary, "    "))
 	}
+	// Forward zones bound to this view.
+	for _, z := range in.Forwards {
+		if z.Spec.ViewRef == v.Name {
+			b.WriteString(renderForwardZone(z, "    "))
+		}
+	}
 	b.WriteString("};\n\n")
 	return b.String()
 }

internal/bind/render_test.go

@@ -80,6 +80,29 @@ func TestRenderCatalogUsesPrimaryIP(t *testing.T) {
 	}
 }
 
+func TestRenderForwardZoneInView(t *testing.T) {
+	rec := true
+	in := RenderInput{
+		Cluster: newCluster(bindv1alpha1.ModeResolver),
+		Views: []bindv1alpha1.BindView{{
+			ObjectMeta: metav1.ObjectMeta{Name: "openforwarder"},
+			Spec:       bindv1alpha1.BindViewSpec{ClusterRef: "auth", MatchClients: []string{"acl-main"}, Recursion: &rec},
+		}},
+		Forwards: []bindv1alpha1.BindZone{{
+			Spec: bindv1alpha1.BindZoneSpec{ClusterRef: "auth", ZoneName: "unkin.net", Type: bindv1alpha1.ZoneForward, ViewRef: "openforwarder", Forwarders: []string{"198.18.19.15"}},
+		}},
+	}
+	primary, secondary := RenderNamedConf(in)
+	for _, out := range []string{primary, secondary} {
+		if !strings.Contains(out, `view "openforwarder"`) {
+			t.Fatalf("view missing:\n%s", out)
+		}
+		if !strings.Contains(out, `zone "unkin.net" {`) || !strings.Contains(out, "type forward;") || !strings.Contains(out, "forwarders { 198.18.19.15; }") {
+			t.Fatalf("forward zone not rendered inside view (must be on all pods):\n%s", out)
+		}
+	}
+}
+
 func TestRenderACL(t *testing.T) {
 	in := RenderInput{
 		Cluster: newCluster(bindv1alpha1.ModeAuthoritative),

internal/controller/bindcluster_controller.go

@@ -128,7 +128,15 @@ func (r *BindClusterReconciler) reconcileKeysSecret(ctx context.Context, c *bind
 	if err := r.List(ctx, &keys, client.InNamespace(c.Namespace)); err != nil {
 		return err
 	}
-	items := append([]bindv1alpha1.BindTSIGKey(nil), keys.Items...)
+	// Include keys scoped to this cluster (spec.clusterRef == name) and shared
+	// keys (empty clusterRef). This keeps keys from leaking across clusters that
+	// share a namespace.
+	var items []bindv1alpha1.BindTSIGKey
+	for _, k := range keys.Items {
+		if k.Spec.ClusterRef == "" || k.Spec.ClusterRef == c.Name {
+			items = append(items, k)
+		}
+	}
 	sort.Slice(items, func(i, j int) bool { return items[i].Name < items[j].Name })
 
 	var b strings.Builder
@@ -198,6 +206,17 @@ func (r *BindClusterReconciler) reconcileConfigMap(ctx context.Context, c *bindv
 		}
 	}
 
+	// Forward zones are configuration (no data), so they are rendered into
+	// named.conf on every pod rather than added dynamically to the primary.
+	var zones bindv1alpha1.BindZoneList
+	if err := r.List(ctx, &zones, client.InNamespace(c.Namespace)); err == nil {
+		for _, z := range zones.Items {
+			if z.Spec.ClusterRef == c.Name && z.Spec.Type == bindv1alpha1.ZoneForward {
+				in.Forwards = append(in.Forwards, z)
+			}
+		}
+	}
+
 	var catalogs bindv1alpha1.BindCatalogZoneList
 	if err := r.List(ctx, &catalogs, client.InNamespace(c.Namespace)); err == nil {
 		for i := range catalogs.Items {
@@ -254,7 +273,56 @@ func (r *BindClusterReconciler) reconcileServices(ctx context.Context, c *bindv1
 			LoadBalancerIP: c.Spec.Service.LoadBalancerIP,
 		},
 	}
-	return r.upsertService(ctx, c, client)
+	// externalTrafficPolicy is only valid for LoadBalancer/NodePort Services.
+	if svcType == corev1.ServiceTypeLoadBalancer || svcType == corev1.ServiceTypeNodePort {
+		client.Spec.ExternalTrafficPolicy = c.Spec.Service.ExternalTrafficPolicy
+	}
+	if err := r.upsertService(ctx, c, client); err != nil {
+		return err
+	}
+
+	// Primary (write) Service: routes only to pod-0. Created when configured,
+	// deleted when removed.
+	return r.reconcilePrimaryService(ctx, c, dnsPorts)
+}
+
+func (r *BindClusterReconciler) reconcilePrimaryService(ctx context.Context, c *bindv1alpha1.BindCluster, dnsPorts []corev1.ServicePort) error {
+	name := primaryServiceName(c.Name)
+	if c.Spec.PrimaryService == nil {
+		var existing corev1.Service
+		err := r.Get(ctx, types.NamespacedName{Namespace: c.Namespace, Name: name}, &existing)
+		if apierrors.IsNotFound(err) {
+			return nil
+		}
+		if err != nil {
+			return err
+		}
+		return client.IgnoreNotFound(r.Delete(ctx, &existing))
+	}
+
+	ps := c.Spec.PrimaryService
+	psType := ps.Type
+	if psType == "" {
+		psType = corev1.ServiceTypeClusterIP
+	}
+	svc := &corev1.Service{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:        name,
+			Namespace:   c.Namespace,
+			Labels:      commonLabels(c.Name),
+			Annotations: ps.Annotations,
+		},
+		Spec: corev1.ServiceSpec{
+			Type:           psType,
+			Selector:       primaryPodSelector(c.Name),
+			Ports:          dnsPorts,
+			LoadBalancerIP: ps.LoadBalancerIP,
+		},
+	}
+	if psType == corev1.ServiceTypeLoadBalancer || psType == corev1.ServiceTypeNodePort {
+		svc.Spec.ExternalTrafficPolicy = ps.ExternalTrafficPolicy
+	}
+	return r.upsertService(ctx, c, svc)
 }
 
 func (r *BindClusterReconciler) reconcileStatefulSet(ctx context.Context, c *bindv1alpha1.BindCluster) (*appsv1.StatefulSet, error) {
@@ -408,6 +476,15 @@ func (r *BindClusterReconciler) SetupWithManager(mgr ctrl.Manager) error {
 		Watches(&bindv1alpha1.BindDNSSECPolicy{}, handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, o client.Object) []reconcile.Request {
 			return mapToCluster(o.(*bindv1alpha1.BindDNSSECPolicy).Spec.ClusterRef, o.GetNamespace())
 		})).
+		Watches(&bindv1alpha1.BindZone{}, handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, o client.Object) []reconcile.Request {
+			// Only forward zones affect named.conf; primary/secondary zones are
+			// managed dynamically by the BindZone controller.
+			z := o.(*bindv1alpha1.BindZone)
+			if z.Spec.Type != bindv1alpha1.ZoneForward {
+				return nil
+			}
+			return mapToCluster(z.Spec.ClusterRef, o.GetNamespace())
+		})).
 		Watches(&bindv1alpha1.BindCatalogZone{}, handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, o client.Object) []reconcile.Request {
 			return mapToCluster(o.(*bindv1alpha1.BindCatalogZone).Spec.ClusterRef, o.GetNamespace())
 		})).

internal/controller/bindzone_controller.go

@@ -35,6 +35,19 @@ func (r *BindZoneReconciler) Reconcile(ctx context.Context, req ctrl.Request) (c
 		return ctrl.Result{}, client.IgnoreNotFound(err)
 	}
 
+	// Forward zones are pure configuration rendered into named.conf by the
+	// BindCluster controller (on every pod), not added dynamically to the
+	// primary. Nothing to do here beyond reporting readiness.
+	if zone.Spec.Type == bindv1alpha1.ZoneForward {
+		zone.Status.Phase = "Ready"
+		zone.Status.ObservedGeneration = zone.Generation
+		setReady(&zone.Status.Conditions, zone.Generation, true, "Configured", "forward zone rendered into named.conf")
+		if err := r.Status().Update(ctx, &zone); err != nil {
+			return ctrl.Result{}, err
+		}
+		return ctrl.Result{}, nil
+	}
+
 	cluster, err := getCluster(ctx, r.Client, zone.Namespace, zone.Spec.ClusterRef)
 	if err != nil {
 		return r.setPhase(ctx, &zone, "Error", "ClusterMissing", err.Error())

internal/controller/helpers.go

@@ -29,11 +29,20 @@ const (
 )
 
 func headlessServiceName(cluster string) string { return cluster + "-headless" }
-func clientServiceName(cluster string) string   { return cluster }
-func primaryPodName(cluster string) string      { return cluster + "-0" }
-func configMapName(cluster string) string       { return cluster + "-config" }
-func keysSecretName(cluster string) string      { return cluster + "-keys" }
-func rndcSecretName(cluster string) string      { return cluster + "-rndc" }
+func primaryServiceName(cluster string) string  { return cluster + "-primary" }
+
+// primaryPodSelector selects only the primary pod (ordinal 0) via the stable
+// StatefulSet pod-name label, for the write Service.
+func primaryPodSelector(cluster string) map[string]string {
+	s := commonLabels(cluster)
+	s["statefulset.kubernetes.io/pod-name"] = primaryPodName(cluster)
+	return s
+}
+func clientServiceName(cluster string) string { return cluster }
+func primaryPodName(cluster string) string    { return cluster + "-0" }
+func configMapName(cluster string) string     { return cluster + "-config" }
+func keysSecretName(cluster string) string    { return cluster + "-keys" }
+func rndcSecretName(cluster string) string    { return cluster + "-rndc" }
 
 // primaryAddress is the in-cluster DNS name of the primary pod (ordinal 0).
 func primaryAddress(cluster, namespace string) string {

internal/controller/util.go

@@ -57,6 +57,7 @@ func (r *BindClusterReconciler) upsertService(ctx context.Context, c *bindv1alph
 	existing.Spec.Selector = desired.Spec.Selector
 	existing.Spec.Type = desired.Spec.Type
 	existing.Spec.LoadBalancerIP = desired.Spec.LoadBalancerIP
+	existing.Spec.ExternalTrafficPolicy = desired.Spec.ExternalTrafficPolicy
 	if desired.Annotations != nil {
 		if existing.Annotations == nil {
 			existing.Annotations = map[string]string{}