Merge pull request 'Make intra-cluster NOTIFY loop-free (TSIG-keyed allow-notify, no pod IPs in restart config)' (#15) from benvin/notify-fix-loopfree into main

Reviewed-on: #15

.woodpecker/build.yaml

@@ -8,3 +8,10 @@ steps:
       repo: git.unkin.net/unkin/bind-operator
       dockerfile: Dockerfile.operator
       dry_run: true
+
+  - name: docker-build-tsig-api
+    image: woodpeckerci/plugin-docker-buildx
+    settings:
+      repo: git.unkin.net/unkin/bind-tsig-api
+      dockerfile: Dockerfile.tsigapi
+      dry_run: true

.woodpecker/docker.yaml

@@ -15,3 +15,16 @@ steps:
       tags:
         - ${CI_COMMIT_TAG}
         - latest
+
+  - name: docker-tsig-api
+    image: woodpeckerci/plugin-docker-buildx
+    settings:
+      registry: git.unkin.net
+      repo: git.unkin.net/unkin/bind-tsig-api
+      dockerfile: Dockerfile.tsigapi
+      username: droneci
+      password:
+        from_secret: DRONECI_PASSWORD
+      tags:
+        - ${CI_COMMIT_TAG}
+        - latest

Dockerfile.tsigapi

@@ -0,0 +1,20 @@
+FROM golang:1.25-alpine AS builder
+
+RUN apk add --no-cache git
+
+WORKDIR /build
+
+COPY go.mod go.sum ./
+RUN go mod download
+
+COPY . .
+
+RUN CGO_ENABLED=0 go build -ldflags="-s -w" -o tsig-api ./cmd/tsigapi
+
+FROM gcr.io/distroless/static-debian12:nonroot
+
+COPY --from=builder /build/tsig-api /usr/local/bin/tsig-api
+
+EXPOSE 8443
+
+ENTRYPOINT ["tsig-api"]

README.md

@@ -42,7 +42,7 @@ script picks one based on the pod ordinal.
 | `BindZone` | A forward/reverse zone (`primary`/`secondary`/`forward`/`stub`), records inline, optional dynamic-update + DNSSEC + catalog membership. |
 | `DNSRecord` | A single record set applied via TSIG `nsupdate` — external-dns as a CRD. |
 | `BindView` | A split-horizon view (`match-clients`, ordering, per-view recursion). |
-| `BindTSIGKey` | A TSIG key; the operator generates material into a Secret (never stored in the CR). |
+| `BindTSIGKey` | A TSIG key; the operator generates material into a Secret (never stored in the CR). `spec.secretTemplate` stamps extra labels/annotations onto that Secret (e.g. reflection hints to mirror it into another namespace). |
 | `BindACL` | A reusable named `address_match_list`. |
 | `BindCatalogZone` | A BIND catalog zone so secondaries auto-provision member zones. |
 | `BindPolicy` | A Response Policy Zone (RPZ) / DNS firewall. |

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/bindtsigapi_types.go

@@ -0,0 +1,104 @@
+package v1alpha1
+
+import (
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+// BindTSIGAPISpec configures the companion TSIG API that the operator deploys.
+// The API exposes an HTTP contract (used by vault-plugin-secrets-bind-tsig) for
+// creating, rotating and deleting TSIG keys; it does so by managing BindTSIGKey
+// custom resources, which the operator then reconciles into key material.
+type BindTSIGAPISpec struct {
+	// Image is the companion API container image.
+	// +kubebuilder:default="git.unkin.net/unkin/bind-tsig-api:latest"
+	// +optional
+	Image string `json:"image,omitempty"`
+
+	// ImagePullPolicy for the API container.
+	// +optional
+	ImagePullPolicy corev1.PullPolicy `json:"imagePullPolicy,omitempty"`
+
+	// Replicas of the API. Defaults to 1.
+	// +kubebuilder:default=1
+	// +optional
+	Replicas int32 `json:"replicas,omitempty"`
+
+	// Port the API listens on. Defaults to 8443.
+	// +kubebuilder:default=8443
+	// +optional
+	Port int32 `json:"port,omitempty"`
+
+	// TargetNamespace is where the API creates BindTSIGKey resources. Defaults
+	// to the API's own namespace.
+	// +optional
+	TargetNamespace string `json:"targetNamespace,omitempty"`
+
+	// TokenSecretName holds the master access token clients present to the API.
+	// The operator generates a token if the Secret does not exist, so a
+	// VaultStaticSecret may pre-seed it instead. Defaults to "<name>-token".
+	// +optional
+	TokenSecretName string `json:"tokenSecretName,omitempty"`
+
+	// Env are extra environment variables rendered into the API ConfigMap.
+	// +optional
+	Env map[string]string `json:"env,omitempty"`
+
+	// Service controls how the API is exposed (defaults to ClusterIP).
+	// +optional
+	Service ClusterServiceSpec `json:"service,omitempty"`
+
+	// Resources for the API container.
+	// +optional
+	Resources corev1.ResourceRequirements `json:"resources,omitempty"`
+}
+
+// BindTSIGAPIStatus reports observed API state.
+type BindTSIGAPIStatus struct {
+	// +optional
+	Phase string `json:"phase,omitempty"`
+	// +optional
+	ReadyReplicas int32 `json:"readyReplicas,omitempty"`
+	// Endpoint is the in-cluster URL clients (Vault) use to reach the API.
+	// +optional
+	Endpoint string `json:"endpoint,omitempty"`
+	// TokenSecret is the Secret holding the master access token.
+	// +optional
+	TokenSecret string `json:"tokenSecret,omitempty"`
+	// +optional
+	ObservedGeneration int64 `json:"observedGeneration,omitempty"`
+	// +optional
+	// +listType=map
+	// +listMapKey=type
+	Conditions []metav1.Condition `json:"conditions,omitempty"`
+}
+
+// +kubebuilder:object:root=true
+// +kubebuilder:subresource:status
+// +kubebuilder:resource:shortName=btapi
+// +kubebuilder:printcolumn:name="Endpoint",type=string,JSONPath=`.status.endpoint`
+// +kubebuilder:printcolumn:name="Ready",type=integer,JSONPath=`.status.readyReplicas`
+// +kubebuilder:printcolumn:name="Phase",type=string,JSONPath=`.status.phase`
+
+// BindTSIGAPI deploys the companion TSIG API. Creating one makes the operator
+// reconcile a Deployment, Service, ConfigMap, token Secret and RBAC for it.
+type BindTSIGAPI struct {
+	metav1.TypeMeta   `json:",inline"`
+	metav1.ObjectMeta `json:"metadata,omitempty"`
+
+	Spec   BindTSIGAPISpec   `json:"spec,omitempty"`
+	Status BindTSIGAPIStatus `json:"status,omitempty"`
+}
+
+// +kubebuilder:object:root=true
+
+// BindTSIGAPIList contains a list of BindTSIGAPI.
+type BindTSIGAPIList struct {
+	metav1.TypeMeta `json:",inline"`
+	metav1.ListMeta `json:"metadata,omitempty"`
+	Items           []BindTSIGAPI `json:"items"`
+}
+
+func init() {
+	SchemeBuilder.Register(&BindTSIGAPI{}, &BindTSIGAPIList{})
+}

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
@@ -35,6 +41,24 @@ type BindTSIGKeySpec struct {
 	// `secret` key and the operator will not generate new material.
 	// +optional
 	ImportExisting bool `json:"importExisting,omitempty"`
+
+	// SecretTemplate customizes metadata written onto the managed key Secret.
+	// Useful, for example, to let secret-reflection tooling mirror the key into
+	// another namespace. Operator-managed labels are always preserved.
+	// +optional
+	SecretTemplate *SecretMetadata `json:"secretTemplate,omitempty"`
+}
+
+// SecretMetadata carries extra labels and annotations to stamp onto a
+// Secret managed by the operator.
+type SecretMetadata struct {
+	// Annotations to set on the Secret.
+	// +optional
+	Annotations map[string]string `json:"annotations,omitempty"`
+
+	// Labels to set on the Secret.
+	// +optional
+	Labels map[string]string `json:"labels,omitempty"`
 }
 
 // BindTSIGKeyStatus reports observed TSIG key state.

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))
@@ -576,12 +581,117 @@ func (in *BindPolicyStatus) DeepCopy() *BindPolicyStatus {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *BindTSIGAPI) DeepCopyInto(out *BindTSIGAPI) {
+	*out = *in
+	out.TypeMeta = in.TypeMeta
+	in.ObjectMeta.DeepCopyInto(&out.ObjectMeta)
+	in.Spec.DeepCopyInto(&out.Spec)
+	in.Status.DeepCopyInto(&out.Status)
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new BindTSIGAPI.
+func (in *BindTSIGAPI) DeepCopy() *BindTSIGAPI {
+	if in == nil {
+		return nil
+	}
+	out := new(BindTSIGAPI)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyObject is an autogenerated deepcopy function, copying the receiver, creating a new runtime.Object.
+func (in *BindTSIGAPI) DeepCopyObject() runtime.Object {
+	if c := in.DeepCopy(); c != nil {
+		return c
+	}
+	return nil
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *BindTSIGAPIList) DeepCopyInto(out *BindTSIGAPIList) {
+	*out = *in
+	out.TypeMeta = in.TypeMeta
+	in.ListMeta.DeepCopyInto(&out.ListMeta)
+	if in.Items != nil {
+		in, out := &in.Items, &out.Items
+		*out = make([]BindTSIGAPI, len(*in))
+		for i := range *in {
+			(*in)[i].DeepCopyInto(&(*out)[i])
+		}
+	}
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new BindTSIGAPIList.
+func (in *BindTSIGAPIList) DeepCopy() *BindTSIGAPIList {
+	if in == nil {
+		return nil
+	}
+	out := new(BindTSIGAPIList)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyObject is an autogenerated deepcopy function, copying the receiver, creating a new runtime.Object.
+func (in *BindTSIGAPIList) DeepCopyObject() runtime.Object {
+	if c := in.DeepCopy(); c != nil {
+		return c
+	}
+	return nil
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *BindTSIGAPISpec) DeepCopyInto(out *BindTSIGAPISpec) {
+	*out = *in
+	if in.Env != nil {
+		in, out := &in.Env, &out.Env
+		*out = make(map[string]string, len(*in))
+		for key, val := range *in {
+			(*out)[key] = val
+		}
+	}
+	in.Service.DeepCopyInto(&out.Service)
+	in.Resources.DeepCopyInto(&out.Resources)
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new BindTSIGAPISpec.
+func (in *BindTSIGAPISpec) DeepCopy() *BindTSIGAPISpec {
+	if in == nil {
+		return nil
+	}
+	out := new(BindTSIGAPISpec)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *BindTSIGAPIStatus) DeepCopyInto(out *BindTSIGAPIStatus) {
+	*out = *in
+	if in.Conditions != nil {
+		in, out := &in.Conditions, &out.Conditions
+		*out = make([]v1.Condition, len(*in))
+		for i := range *in {
+			(*in)[i].DeepCopyInto(&(*out)[i])
+		}
+	}
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new BindTSIGAPIStatus.
+func (in *BindTSIGAPIStatus) DeepCopy() *BindTSIGAPIStatus {
+	if in == nil {
+		return nil
+	}
+	out := new(BindTSIGAPIStatus)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *BindTSIGKey) DeepCopyInto(out *BindTSIGKey) {
 	*out = *in
 	out.TypeMeta = in.TypeMeta
 	in.ObjectMeta.DeepCopyInto(&out.ObjectMeta)
-	out.Spec = in.Spec
+	in.Spec.DeepCopyInto(&out.Spec)
 	in.Status.DeepCopyInto(&out.Status)
 }
 
@@ -638,6 +748,11 @@ func (in *BindTSIGKeyList) DeepCopyObject() runtime.Object {
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *BindTSIGKeySpec) DeepCopyInto(out *BindTSIGKeySpec) {
 	*out = *in
+	if in.SecretTemplate != nil {
+		in, out := &in.SecretTemplate, &out.SecretTemplate
+		*out = new(SecretMetadata)
+		(*in).DeepCopyInto(*out)
+	}
 }
 
 // DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new BindTSIGKeySpec.
@@ -1098,3 +1213,32 @@ func (in *Record) DeepCopy() *Record {
 	in.DeepCopyInto(out)
 	return out
 }
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *SecretMetadata) DeepCopyInto(out *SecretMetadata) {
+	*out = *in
+	if in.Annotations != nil {
+		in, out := &in.Annotations, &out.Annotations
+		*out = make(map[string]string, len(*in))
+		for key, val := range *in {
+			(*out)[key] = val
+		}
+	}
+	if in.Labels != nil {
+		in, out := &in.Labels, &out.Labels
+		*out = make(map[string]string, len(*in))
+		for key, val := range *in {
+			(*out)[key] = val
+		}
+	}
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new SecretMetadata.
+func (in *SecretMetadata) DeepCopy() *SecretMetadata {
+	if in == nil {
+		return nil
+	}
+	out := new(SecretMetadata)
+	in.DeepCopyInto(out)
+	return out
+}

cmd/tsigapi/main.go

@@ -0,0 +1,72 @@
+package main
+
+import (
+	"net/http"
+	"os"
+	"time"
+
+	"k8s.io/apimachinery/pkg/runtime"
+	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
+	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/log/zap"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+	"git.unkin.net/unkin/bind-operator/internal/tsigapi"
+)
+
+var scheme = runtime.NewScheme()
+
+func init() {
+	utilruntime.Must(clientgoscheme.AddToScheme(scheme))
+	utilruntime.Must(bindv1alpha1.AddToScheme(scheme))
+}
+
+func main() {
+	ctrl.SetLogger(zap.New(zap.UseDevMode(false)))
+	logger := ctrl.Log.WithName("tsig-api")
+
+	addr := envOr("LISTEN_ADDR", ":8443")
+	namespace := envOr("TARGET_NAMESPACE", currentNamespace())
+	token := os.Getenv("API_TOKEN")
+
+	c, err := client.New(ctrl.GetConfigOrDie(), client.Options{Scheme: scheme})
+	if err != nil {
+		logger.Error(err, "unable to build kubernetes client")
+		os.Exit(1)
+	}
+
+	srv := &tsigapi.Server{
+		Client:      c,
+		Namespace:   namespace,
+		Token:       token,
+		Log:         logger,
+		WaitTimeout: 15 * time.Second,
+	}
+
+	httpSrv := &http.Server{
+		Addr:              addr,
+		Handler:           srv.Handler(),
+		ReadHeaderTimeout: 10 * time.Second,
+	}
+	logger.Info("starting tsig api", "addr", addr, "namespace", namespace)
+	if err := httpSrv.ListenAndServe(); err != nil {
+		logger.Error(err, "tsig api exited")
+		os.Exit(1)
+	}
+}
+
+func envOr(key, fallback string) string {
+	if v := os.Getenv(key); v != "" {
+		return v
+	}
+	return fallback
+}
+
+func currentNamespace() string {
+	if b, err := os.ReadFile("/var/run/secrets/kubernetes.io/serviceaccount/namespace"); err == nil {
+		return string(b)
+	}
+	return "bind-system"
+}

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_bindtsigapis.yaml

@@ -0,0 +1,267 @@
+---
+apiVersion: apiextensions.k8s.io/v1
+kind: CustomResourceDefinition
+metadata:
+  annotations:
+    controller-gen.kubebuilder.io/version: v0.17.3
+  name: bindtsigapis.bind.unkin.net
+spec:
+  group: bind.unkin.net
+  names:
+    kind: BindTSIGAPI
+    listKind: BindTSIGAPIList
+    plural: bindtsigapis
+    shortNames:
+    - btapi
+    singular: bindtsigapi
+  scope: Namespaced
+  versions:
+  - additionalPrinterColumns:
+    - jsonPath: .status.endpoint
+      name: Endpoint
+      type: string
+    - jsonPath: .status.readyReplicas
+      name: Ready
+      type: integer
+    - jsonPath: .status.phase
+      name: Phase
+      type: string
+    name: v1alpha1
+    schema:
+      openAPIV3Schema:
+        description: |-
+          BindTSIGAPI deploys the companion TSIG API. Creating one makes the operator
+          reconcile a Deployment, Service, ConfigMap, token Secret and RBAC for it.
+        properties:
+          apiVersion:
+            description: |-
+              APIVersion defines the versioned schema of this representation of an object.
+              Servers should convert recognized schemas to the latest internal value, and
+              may reject unrecognized values.
+              More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources
+            type: string
+          kind:
+            description: |-
+              Kind is a string value representing the REST resource this object represents.
+              Servers may infer this from the endpoint the client submits requests to.
+              Cannot be updated.
+              In CamelCase.
+              More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds
+            type: string
+          metadata:
+            type: object
+          spec:
+            description: |-
+              BindTSIGAPISpec configures the companion TSIG API that the operator deploys.
+              The API exposes an HTTP contract (used by vault-plugin-secrets-bind-tsig) for
+              creating, rotating and deleting TSIG keys; it does so by managing BindTSIGKey
+              custom resources, which the operator then reconciles into key material.
+            properties:
+              env:
+                additionalProperties:
+                  type: string
+                description: Env are extra environment variables rendered into the
+                  API ConfigMap.
+                type: object
+              image:
+                default: git.unkin.net/unkin/bind-tsig-api:latest
+                description: Image is the companion API container image.
+                type: string
+              imagePullPolicy:
+                description: ImagePullPolicy for the API container.
+                type: string
+              port:
+                default: 8443
+                description: Port the API listens on. Defaults to 8443.
+                format: int32
+                type: integer
+              replicas:
+                default: 1
+                description: Replicas of the API. Defaults to 1.
+                format: int32
+                type: integer
+              resources:
+                description: Resources for the API container.
+                properties:
+                  claims:
+                    description: |-
+                      Claims lists the names of resources, defined in spec.resourceClaims,
+                      that are used by this container.
+
+                      This field depends on the
+                      DynamicResourceAllocation feature gate.
+
+                      This field is immutable. It can only be set for containers.
+                    items:
+                      description: ResourceClaim references one entry in PodSpec.ResourceClaims.
+                      properties:
+                        name:
+                          description: |-
+                            Name must match the name of one entry in pod.spec.resourceClaims of
+                            the Pod where this field is used. It makes that resource available
+                            inside a container.
+                          type: string
+                        request:
+                          description: |-
+                            Request is the name chosen for a request in the referenced claim.
+                            If empty, everything from the claim is made available, otherwise
+                            only the result of this request.
+                          type: string
+                      required:
+                      - name
+                      type: object
+                    type: array
+                    x-kubernetes-list-map-keys:
+                    - name
+                    x-kubernetes-list-type: map
+                  limits:
+                    additionalProperties:
+                      anyOf:
+                      - type: integer
+                      - type: string
+                      pattern: ^(\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))))?$
+                      x-kubernetes-int-or-string: true
+                    description: |-
+                      Limits describes the maximum amount of compute resources allowed.
+                      More info: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/
+                    type: object
+                  requests:
+                    additionalProperties:
+                      anyOf:
+                      - type: integer
+                      - type: string
+                      pattern: ^(\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))))?$
+                      x-kubernetes-int-or-string: true
+                    description: |-
+                      Requests describes the minimum amount of compute resources required.
+                      If Requests is omitted for a container, it defaults to Limits if that is explicitly specified,
+                      otherwise to an implementation-defined value. Requests cannot exceed Limits.
+                      More info: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/
+                    type: object
+                type: object
+              service:
+                description: Service controls how the API is exposed (defaults to
+                  ClusterIP).
+                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
+              targetNamespace:
+                description: |-
+                  TargetNamespace is where the API creates BindTSIGKey resources. Defaults
+                  to the API's own namespace.
+                type: string
+              tokenSecretName:
+                description: |-
+                  TokenSecretName holds the master access token clients present to the API.
+                  The operator generates a token if the Secret does not exist, so a
+                  VaultStaticSecret may pre-seed it instead. Defaults to "<name>-token".
+                type: string
+            type: object
+          status:
+            description: BindTSIGAPIStatus reports observed API state.
+            properties:
+              conditions:
+                items:
+                  description: Condition contains details for one aspect of the current
+                    state of this API Resource.
+                  properties:
+                    lastTransitionTime:
+                      description: |-
+                        lastTransitionTime is the last time the condition transitioned from one status to another.
+                        This should be when the underlying condition changed.  If that is not known, then using the time when the API field changed is acceptable.
+                      format: date-time
+                      type: string
+                    message:
+                      description: |-
+                        message is a human readable message indicating details about the transition.
+                        This may be an empty string.
+                      maxLength: 32768
+                      type: string
+                    observedGeneration:
+                      description: |-
+                        observedGeneration represents the .metadata.generation that the condition was set based upon.
+                        For instance, if .metadata.generation is currently 12, but the .status.conditions[x].observedGeneration is 9, the condition is out of date
+                        with respect to the current state of the instance.
+                      format: int64
+                      minimum: 0
+                      type: integer
+                    reason:
+                      description: |-
+                        reason contains a programmatic identifier indicating the reason for the condition's last transition.
+                        Producers of specific condition types may define expected values and meanings for this field,
+                        and whether the values are considered a guaranteed API.
+                        The value should be a CamelCase string.
+                        This field may not be empty.
+                      maxLength: 1024
+                      minLength: 1
+                      pattern: ^[A-Za-z]([A-Za-z0-9_,:]*[A-Za-z0-9_])?$
+                      type: string
+                    status:
+                      description: status of the condition, one of True, False, Unknown.
+                      enum:
+                      - "True"
+                      - "False"
+                      - Unknown
+                      type: string
+                    type:
+                      description: type of condition in CamelCase or in foo.example.com/CamelCase.
+                      maxLength: 316
+                      pattern: ^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])$
+                      type: string
+                  required:
+                  - lastTransitionTime
+                  - message
+                  - reason
+                  - status
+                  - type
+                  type: object
+                type: array
+                x-kubernetes-list-map-keys:
+                - type
+                x-kubernetes-list-type: map
+              endpoint:
+                description: Endpoint is the in-cluster URL clients (Vault) use to
+                  reach the API.
+                type: string
+              observedGeneration:
+                format: int64
+                type: integer
+              phase:
+                type: string
+              readyReplicas:
+                format: int32
+                type: integer
+              tokenSecret:
+                description: TokenSecret is the Secret holding the master access token.
+                type: string
+            type: object
+        type: object
+    served: true
+    storage: true
+    subresources:
+      status: {}

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
@@ -81,6 +87,23 @@ spec:
                   SecretName is the Secret the key material is written to (or read from when
                   ImportExisting is set). Defaults to "<name>-tsig".
                 type: string
+              secretTemplate:
+                description: |-
+                  SecretTemplate customizes metadata written onto the managed key Secret.
+                  Useful, for example, to let secret-reflection tooling mirror the key into
+                  another namespace. Operator-managed labels are always preserved.
+                properties:
+                  annotations:
+                    additionalProperties:
+                      type: string
+                    description: Annotations to set on the Secret.
+                    type: object
+                  labels:
+                    additionalProperties:
+                      type: string
+                    description: Labels to set on the Secret.
+                    type: object
+                type: object
             type: object
           status:
             description: BindTSIGKeyStatus reports observed TSIG key state.

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.
@@ -1978,6 +2023,273 @@ spec:
 ---
 apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
+metadata:
+  annotations:
+    controller-gen.kubebuilder.io/version: v0.17.3
+  name: bindtsigapis.bind.unkin.net
+spec:
+  group: bind.unkin.net
+  names:
+    kind: BindTSIGAPI
+    listKind: BindTSIGAPIList
+    plural: bindtsigapis
+    shortNames:
+    - btapi
+    singular: bindtsigapi
+  scope: Namespaced
+  versions:
+  - additionalPrinterColumns:
+    - jsonPath: .status.endpoint
+      name: Endpoint
+      type: string
+    - jsonPath: .status.readyReplicas
+      name: Ready
+      type: integer
+    - jsonPath: .status.phase
+      name: Phase
+      type: string
+    name: v1alpha1
+    schema:
+      openAPIV3Schema:
+        description: |-
+          BindTSIGAPI deploys the companion TSIG API. Creating one makes the operator
+          reconcile a Deployment, Service, ConfigMap, token Secret and RBAC for it.
+        properties:
+          apiVersion:
+            description: |-
+              APIVersion defines the versioned schema of this representation of an object.
+              Servers should convert recognized schemas to the latest internal value, and
+              may reject unrecognized values.
+              More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources
+            type: string
+          kind:
+            description: |-
+              Kind is a string value representing the REST resource this object represents.
+              Servers may infer this from the endpoint the client submits requests to.
+              Cannot be updated.
+              In CamelCase.
+              More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds
+            type: string
+          metadata:
+            type: object
+          spec:
+            description: |-
+              BindTSIGAPISpec configures the companion TSIG API that the operator deploys.
+              The API exposes an HTTP contract (used by vault-plugin-secrets-bind-tsig) for
+              creating, rotating and deleting TSIG keys; it does so by managing BindTSIGKey
+              custom resources, which the operator then reconciles into key material.
+            properties:
+              env:
+                additionalProperties:
+                  type: string
+                description: Env are extra environment variables rendered into the
+                  API ConfigMap.
+                type: object
+              image:
+                default: git.unkin.net/unkin/bind-tsig-api:latest
+                description: Image is the companion API container image.
+                type: string
+              imagePullPolicy:
+                description: ImagePullPolicy for the API container.
+                type: string
+              port:
+                default: 8443
+                description: Port the API listens on. Defaults to 8443.
+                format: int32
+                type: integer
+              replicas:
+                default: 1
+                description: Replicas of the API. Defaults to 1.
+                format: int32
+                type: integer
+              resources:
+                description: Resources for the API container.
+                properties:
+                  claims:
+                    description: |-
+                      Claims lists the names of resources, defined in spec.resourceClaims,
+                      that are used by this container.
+
+                      This field depends on the
+                      DynamicResourceAllocation feature gate.
+
+                      This field is immutable. It can only be set for containers.
+                    items:
+                      description: ResourceClaim references one entry in PodSpec.ResourceClaims.
+                      properties:
+                        name:
+                          description: |-
+                            Name must match the name of one entry in pod.spec.resourceClaims of
+                            the Pod where this field is used. It makes that resource available
+                            inside a container.
+                          type: string
+                        request:
+                          description: |-
+                            Request is the name chosen for a request in the referenced claim.
+                            If empty, everything from the claim is made available, otherwise
+                            only the result of this request.
+                          type: string
+                      required:
+                      - name
+                      type: object
+                    type: array
+                    x-kubernetes-list-map-keys:
+                    - name
+                    x-kubernetes-list-type: map
+                  limits:
+                    additionalProperties:
+                      anyOf:
+                      - type: integer
+                      - type: string
+                      pattern: ^(\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))))?$
+                      x-kubernetes-int-or-string: true
+                    description: |-
+                      Limits describes the maximum amount of compute resources allowed.
+                      More info: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/
+                    type: object
+                  requests:
+                    additionalProperties:
+                      anyOf:
+                      - type: integer
+                      - type: string
+                      pattern: ^(\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))))?$
+                      x-kubernetes-int-or-string: true
+                    description: |-
+                      Requests describes the minimum amount of compute resources required.
+                      If Requests is omitted for a container, it defaults to Limits if that is explicitly specified,
+                      otherwise to an implementation-defined value. Requests cannot exceed Limits.
+                      More info: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/
+                    type: object
+                type: object
+              service:
+                description: Service controls how the API is exposed (defaults to
+                  ClusterIP).
+                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
+              targetNamespace:
+                description: |-
+                  TargetNamespace is where the API creates BindTSIGKey resources. Defaults
+                  to the API's own namespace.
+                type: string
+              tokenSecretName:
+                description: |-
+                  TokenSecretName holds the master access token clients present to the API.
+                  The operator generates a token if the Secret does not exist, so a
+                  VaultStaticSecret may pre-seed it instead. Defaults to "<name>-token".
+                type: string
+            type: object
+          status:
+            description: BindTSIGAPIStatus reports observed API state.
+            properties:
+              conditions:
+                items:
+                  description: Condition contains details for one aspect of the current
+                    state of this API Resource.
+                  properties:
+                    lastTransitionTime:
+                      description: |-
+                        lastTransitionTime is the last time the condition transitioned from one status to another.
+                        This should be when the underlying condition changed.  If that is not known, then using the time when the API field changed is acceptable.
+                      format: date-time
+                      type: string
+                    message:
+                      description: |-
+                        message is a human readable message indicating details about the transition.
+                        This may be an empty string.
+                      maxLength: 32768
+                      type: string
+                    observedGeneration:
+                      description: |-
+                        observedGeneration represents the .metadata.generation that the condition was set based upon.
+                        For instance, if .metadata.generation is currently 12, but the .status.conditions[x].observedGeneration is 9, the condition is out of date
+                        with respect to the current state of the instance.
+                      format: int64
+                      minimum: 0
+                      type: integer
+                    reason:
+                      description: |-
+                        reason contains a programmatic identifier indicating the reason for the condition's last transition.
+                        Producers of specific condition types may define expected values and meanings for this field,
+                        and whether the values are considered a guaranteed API.
+                        The value should be a CamelCase string.
+                        This field may not be empty.
+                      maxLength: 1024
+                      minLength: 1
+                      pattern: ^[A-Za-z]([A-Za-z0-9_,:]*[A-Za-z0-9_])?$
+                      type: string
+                    status:
+                      description: status of the condition, one of True, False, Unknown.
+                      enum:
+                      - "True"
+                      - "False"
+                      - Unknown
+                      type: string
+                    type:
+                      description: type of condition in CamelCase or in foo.example.com/CamelCase.
+                      maxLength: 316
+                      pattern: ^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])$
+                      type: string
+                  required:
+                  - lastTransitionTime
+                  - message
+                  - reason
+                  - status
+                  - type
+                  type: object
+                type: array
+                x-kubernetes-list-map-keys:
+                - type
+                x-kubernetes-list-type: map
+              endpoint:
+                description: Endpoint is the in-cluster URL clients (Vault) use to
+                  reach the API.
+                type: string
+              observedGeneration:
+                format: int64
+                type: integer
+              phase:
+                type: string
+              readyReplicas:
+                format: int32
+                type: integer
+              tokenSecret:
+                description: TokenSecret is the Secret holding the master access token.
+                type: string
+            type: object
+        type: object
+    served: true
+    storage: true
+    subresources:
+      status: {}
+---
+apiVersion: apiextensions.k8s.io/v1
+kind: CustomResourceDefinition
 metadata:
   annotations:
     controller-gen.kubebuilder.io/version: v0.17.3
@@ -2043,6 +2355,12 @@ 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
@@ -2058,6 +2376,23 @@ spec:
                   SecretName is the Secret the key material is written to (or read from when
                   ImportExisting is set). Defaults to "<name>-tsig".
                 type: string
+              secretTemplate:
+                description: |-
+                  SecretTemplate customizes metadata written onto the managed key Secret.
+                  Useful, for example, to let secret-reflection tooling mirror the key into
+                  another namespace. Operator-managed labels are always preserved.
+                properties:
+                  annotations:
+                    additionalProperties:
+                      type: string
+                    description: Annotations to set on the Secret.
+                    type: object
+                  labels:
+                    additionalProperties:
+                      type: string
+                    description: Labels to set on the Secret.
+                    type: object
+                type: object
             type: object
           status:
             description: BindTSIGKeyStatus reports observed TSIG key state.

config/rbac/role.yaml

@@ -9,6 +9,7 @@ rules:
   resources:
   - configmaps
   - secrets
+  - serviceaccounts
   - services
   verbs:
   - create
@@ -36,6 +37,7 @@ rules:
 - apiGroups:
   - apps
   resources:
+  - deployments
   - statefulsets
   verbs:
   - create
@@ -53,6 +55,7 @@ rules:
   - bindclusters
   - binddnssecpolicies
   - bindpolicies
+  - bindtsigapis
   - bindtsigkeys
   - bindviews
   - bindzones
@@ -73,6 +76,7 @@ rules:
   - bindclusters/status
   - binddnssecpolicies/status
   - bindpolicies/status
+  - bindtsigapis/status
   - bindtsigkeys/status
   - bindviews/status
   - bindzones/status
@@ -81,3 +85,16 @@ rules:
   - get
   - patch
   - update
+- apiGroups:
+  - rbac.authorization.k8s.io
+  resources:
+  - rolebindings
+  - roles
+  verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+  - update
+  - watch

config/samples/00-tsigkeys.yaml

@@ -11,7 +11,9 @@ spec:
   algorithm: hmac-sha256
 ---
 # TSIG key permitting external-dns (and DNSRecord objects) to send RFC2136
-# dynamic updates to the dynamic cluster's primary.
+# dynamic updates to the dynamic cluster's primary. secretTemplate mirrors the
+# generated Secret into the external-dns namespace via emberstack reflector, so
+# external-dns presents exactly the key the primary's allow-update accepts.
 apiVersion: bind.unkin.net/v1alpha1
 kind: BindTSIGKey
 metadata:
@@ -19,3 +21,9 @@ metadata:
   namespace: bind-externaldns
 spec:
   algorithm: hmac-sha256
+  secretTemplate:
+    annotations:
+      reflector.v1.k8s.emberstack.com/reflection-allowed: "true"
+      reflector.v1.k8s.emberstack.com/reflection-allowed-namespaces: "externaldns"
+      reflector.v1.k8s.emberstack.com/reflection-auto-enabled: "true"
+      reflector.v1.k8s.emberstack.com/reflection-auto-namespaces: "externaldns"

go.mod

@@ -3,6 +3,7 @@ module git.unkin.net/unkin/bind-operator
 go 1.25
 
 require (
+	github.com/go-logr/logr v1.4.2
 	k8s.io/api v0.34.4
 	k8s.io/apimachinery v0.34.4
 	k8s.io/client-go v0.34.4
@@ -17,7 +18,6 @@ require (
 	github.com/evanphx/json-patch/v5 v5.9.11 // indirect
 	github.com/fsnotify/fsnotify v1.9.0 // indirect
 	github.com/fxamacker/cbor/v2 v2.9.0 // indirect
-	github.com/go-logr/logr v1.4.2 // indirect
 	github.com/go-logr/zapr v1.3.0 // indirect
 	github.com/go-openapi/jsonpointer v0.21.0 // indirect
 	github.com/go-openapi/jsonreference v0.20.2 // indirect

internal/bind/render.go

@@ -16,17 +16,54 @@ 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
+	// NotifyKeyName is the TSIG key name secondaries accept intra-cluster NOTIFYs
+	// under. The primary pod's NOTIFYs egress with its *pod* IP as the source — k8s
+	// Services only NAT the inbound direction — so BIND, whose implicit
+	// allow-notify is the zone's primaries list (the stable ClusterIP), REFUSES
+	// them as "non-primary" and replication falls back to the SOA refresh timer.
+	//
+	// The primary signs its outgoing NOTIFYs with this key (the zone-scope
+	// also-notify entries, applied via rndc, carry `key "<name>"`), and
+	// secondaries render `allow-notify { key "<name>"; }` — an address-match-list
+	// with a *key* element and NO IPs. Because it names only the key, this clause
+	// is fully static: it never changes when a pod IP churns, so it cannot feed a
+	// changed config-hash and cannot roll the StatefulSet (the v0.2.5 loop). Empty
+	// leaves BIND's default behaviour unchanged.
+	NotifyKeyName string
 }
 
 // RenderNamedConf returns the primary and secondary named.conf contents for a
 // cluster. Both variants are shipped in the ConfigMap; the entrypoint selects
 // one based on the pod ordinal.
 func RenderNamedConf(in RenderInput) (primary string, secondary string) {
+	// client.List returns cache-ordered (non-deterministic) results, so sort
+	// every input slice before rendering. Otherwise the rendered config
+	// reshuffles between reconciles, churning the ConfigMap — and with the
+	// pod-template config hash that means an endless rolling restart.
+	sortInput(&in)
 	return render(in, true), render(in, false)
 }
 
+// sortInput orders every list rendered into named.conf deterministically.
+func sortInput(in *RenderInput) {
+	sort.Slice(in.ACLs, func(i, j int) bool { return in.ACLs[i].Name < in.ACLs[j].Name })
+	sort.Slice(in.Views, func(i, j int) bool {
+		if in.Views[i].Spec.Order != in.Views[j].Spec.Order {
+			return in.Views[i].Spec.Order < in.Views[j].Spec.Order
+		}
+		return in.Views[i].Name < in.Views[j].Name
+	})
+	sort.Slice(in.Forwards, func(i, j int) bool { return in.Forwards[i].Spec.ZoneName < in.Forwards[j].Spec.ZoneName })
+	sort.Slice(in.Policies, func(i, j int) bool { return in.Policies[i].Spec.ZoneName < in.Policies[j].Spec.ZoneName })
+	sort.Slice(in.DNSSECPolicies, func(i, j int) bool { return in.DNSSECPolicies[i].Name < in.DNSSECPolicies[j].Name })
+}
+
 func render(in RenderInput, isPrimary bool) string {
 	c := in.Cluster
 	var b strings.Builder
@@ -63,6 +100,14 @@ func render(in RenderInput, isPrimary bool) string {
 		b.WriteString("    allow-new-zones yes;\n")
 	}
 	b.WriteString("    dnssec-validation auto;\n")
+	// Secondaries accept NOTIFY authenticated by the cluster's NOTIFY TSIG key.
+	// Catalog member and plain secondary zones take their implicit allow-notify
+	// from their primaries (the primary Service ClusterIP), but the primary's
+	// NOTIFYs are sourced from its pod IP, so they are refused as "non-primary"
+	// unless an explicit allow-notify covers them. Rather than enumerate pod IPs
+	// (restart-scoped config that depends on pod IPs — the v0.2.5 roll loop), we
+	// accept any NOTIFY signed with the cluster key: `allow-notify { key "X"; }`.
+	b.WriteString(allowNotifyClause(in, isPrimary, "    "))
 	for _, o := range c.Spec.ExtraOptions {
 		b.WriteString("    " + strings.TrimRight(o, ";") + ";\n")
 	}
@@ -90,6 +135,29 @@ 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 +190,12 @@ 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()
 }
@@ -197,15 +271,68 @@ func responsePolicyClause(policies []bindv1alpha1.BindPolicy, indent string) str
 	return b.String()
 }
 
+// transferPrimaries returns the primaries list secondaries use to AXFR the
+// catalog (and, by inheritance, its member zones), each annotated with the
+// catalog transfer TSIG key. The primary requires key-authenticated transfers
+// (allow-transfer { key ... }), so an unkeyed primaries list is REFUSED.
+func transferPrimaries(in RenderInput) []string {
+	primaries := in.Catalog.Spec.DefaultPrimaries
+	if len(primaries) == 0 && in.PrimaryAddress != "" {
+		primaries = []string{in.PrimaryAddress}
+	}
+	key := in.Catalog.Spec.TransferKeyRef
+	if key == "" {
+		return primaries
+	}
+	out := make([]string, 0, len(primaries))
+	for _, p := range primaries {
+		p = strings.TrimSpace(strings.TrimRight(p, ";"))
+		if p == "" {
+			continue
+		}
+		if strings.Contains(p, " key ") {
+			out = append(out, p)
+		} else {
+			out = append(out, fmt.Sprintf("%s key \"%s\"", p, key))
+		}
+	}
+	return out
+}
+
+// allowNotifyClause renders an options-scope allow-notify on secondaries that
+// accepts intra-cluster NOTIFYs authenticated by the cluster's NOTIFY TSIG key.
+// Zones (catalog members and plain secondaries) point their primaries at the
+// primary Service ClusterIP for stable AXFR, which also becomes their implicit
+// allow-notify — but NOTIFYs leave the primary pod with its *pod* IP as source,
+// so without this they are refused as "non-primary". The primary signs those
+// NOTIFYs with NotifyKeyName (see the also-notify entries applied via rndc), and
+// this clause admits them by key.
+//
+// The rendered clause is `allow-notify { key "<name>"; };` — a key element with
+// NO IP addresses. It is therefore fully static: it depends only on the key
+// name, never on any pod IP, so it can never change the config-hash and can
+// never trigger a rolling restart. This is the fix for the v0.2.5 loop, where an
+// options-scope allow-notify enumerating the primary pod IP re-rendered on every
+// pod churn, flipped the hash, rolled the pods, changed the IP, and looped.
+//
+// Emitted only on secondaries and only when the NOTIFY key name is known.
+func allowNotifyClause(in RenderInput, isPrimary bool, indent string) string {
+	if isPrimary {
+		return ""
+	}
+	key := strings.TrimSpace(in.NotifyKeyName)
+	if key == "" {
+		return ""
+	}
+	return fmt.Sprintf("%sallow-notify { key \"%s\"; };\n", indent, key)
+}
+
 func catalogZonesClause(in RenderInput, isPrimary bool, indent string) string {
 	// Only secondaries consume the catalog to auto-provision member zones.
 	if in.Catalog == nil || isPrimary {
 		return ""
 	}
-	primaries := in.Catalog.Spec.DefaultPrimaries
+	primaries := transferPrimaries(in)
-	if len(primaries) == 0 && in.PrimaryAddress != "" {
-		primaries = []string{in.PrimaryAddress}
-	}
 	if len(primaries) == 0 {
 		return ""
 	}
@@ -226,10 +353,7 @@ func renderCatalogZoneDecl(in RenderInput, isPrimary bool, indent string) string
 	}
 	cat := in.Catalog
 	file := CatalogFilePath(cat.Spec.ZoneName)
-	primaries := cat.Spec.DefaultPrimaries
+	primaries := transferPrimaries(in)
-	if len(primaries) == 0 && in.PrimaryAddress != "" {
-		primaries = []string{in.PrimaryAddress}
-	}
 	if len(primaries) == 0 {
 		// Primary IP not known yet; omit the secondary catalog zone rather than
 		// emit an invalid empty primaries list. A Pod-triggered reconcile renders

internal/bind/render_test.go

@@ -80,6 +80,131 @@ func TestRenderCatalogUsesPrimaryIP(t *testing.T) {
 	}
 }
 
+func TestRenderCatalogPrimariesCarryTransferKey(t *testing.T) {
+	// When the catalog declares a transfer key, secondaries must present it in
+	// both the catalog-zones default-primaries and the secondary catalog zone,
+	// or the key-authenticated primary REFUSES the AXFR.
+	in := RenderInput{
+		Cluster:        newCluster(bindv1alpha1.ModeAuthoritative),
+		Catalog:        &bindv1alpha1.BindCatalogZone{Spec: bindv1alpha1.BindCatalogZoneSpec{ZoneName: "catalog.internal", TransferKeyRef: "transfer-key"}},
+		PrimaryAddress: "10.43.0.5",
+	}
+	_, secondary := RenderNamedConf(in)
+	if !strings.Contains(secondary, `default-primaries { 10.43.0.5 key "transfer-key"; }`) {
+		t.Fatalf("catalog-zones default-primaries must carry the transfer key:\n%s", secondary)
+	}
+	if !strings.Contains(secondary, `primaries { 10.43.0.5 key "transfer-key"; }`) {
+		t.Fatalf("secondary catalog zone primaries must carry the transfer key:\n%s", secondary)
+	}
+}
+
+func TestRenderSecondaryAllowNotifyByKey(t *testing.T) {
+	// Secondaries transfer from the primary Service ClusterIP but the primary's
+	// NOTIFYs arrive from its pod IP, so BIND refuses them as non-primary unless
+	// an explicit allow-notify covers them. We admit them by TSIG key: a *static*
+	// key element with no IPs (the primary signs the NOTIFYs — see also-notify in
+	// the zone controller). NO pod IP may appear here, or the config-hash churns.
+	in := RenderInput{
+		Cluster:        newCluster(bindv1alpha1.ModeAuthoritative),
+		PrimaryAddress: "10.43.5.5",
+		NotifyKeyName:  "externaldns-key",
+	}
+	primary, secondary := RenderNamedConf(in)
+	if !strings.Contains(secondary, `allow-notify { key "externaldns-key"; };`) {
+		t.Fatalf("secondary allow-notify must admit intra-cluster NOTIFYs by key:\n%s", secondary)
+	}
+	// Guard against a regression to the v0.2.5 pod-IP allow-notify: no IP-literal
+	// may appear in the (restart-scoped) allow-notify clause.
+	for _, line := range strings.Split(secondary, "\n") {
+		if strings.Contains(line, "allow-notify") && (strings.Contains(line, "10.42.") || strings.Contains(line, "10.43.")) {
+			t.Fatalf("allow-notify must not enumerate pod/service IPs (v0.2.5 roll loop):\n%s", line)
+		}
+	}
+	if strings.Contains(primary, "allow-notify") {
+		t.Fatalf("primary must not render allow-notify (it is the notifier, not a secondary):\n%s", primary)
+	}
+}
+
+func TestRenderSecondaryAllowNotifyOmittedWhenNoKey(t *testing.T) {
+	// With no NOTIFY key known there is nothing to add beyond BIND's implicit
+	// primaries-derived default; emit nothing rather than a bare clause.
+	in := RenderInput{Cluster: newCluster(bindv1alpha1.ModeAuthoritative), PrimaryAddress: "10.43.5.5"}
+	_, secondary := RenderNamedConf(in)
+	if strings.Contains(secondary, "allow-notify") {
+		t.Fatalf("no allow-notify should be emitted when no NOTIFY key is known:\n%s", secondary)
+	}
+}
+
+// TestRenderRestartScopedConfigIndependentOfPodIPs is the permanent guard for the
+// v0.2.5 rolling-restart loop. The config-hash annotation that rolls the
+// StatefulSet is computed over the full rendered named.conf (see
+// BindClusterReconciler.configHash). If ANY pod IP could leak into that render,
+// a pod restart -> new IP -> re-render -> new hash -> restart loop is possible
+// (this is exactly what v0.2.5 did with its options-scope pod-IP allow-notify).
+//
+// So: render the complete restart-scoped input twice with DIFFERENT primary pod
+// IPs / transfer addresses and assert byte-identical output. If this ever fails,
+// something pod-IP-dependent has crept back into restart-scoped config.
+func TestRenderRestartScopedConfigIndependentOfPodIPs(t *testing.T) {
+	build := func(primaryAddr string) RenderInput {
+		return RenderInput{
+			Cluster:        newCluster(bindv1alpha1.ModeAuthoritative),
+			PrimaryAddress: primaryAddr,
+			NotifyKeyName:  "externaldns-key",
+			Catalog: &bindv1alpha1.BindCatalogZone{
+				Spec: bindv1alpha1.BindCatalogZoneSpec{ZoneName: "catalog.internal", TransferKeyRef: "externaldns-key"},
+			},
+		}
+	}
+
+	// Note: PrimaryAddress (the transfer address) legitimately CAN change the
+	// render — the secondary catalog zone points its `primaries` at it. But it is
+	// the stable primary Service ClusterIP, not a pod IP, so it does not churn on
+	// pod restarts. The bug was pod IPs. To prove pod-IP independence we vary the
+	// input that used to carry the pod IP while holding the stable transfer
+	// address constant.
+	p1, s1 := RenderNamedConf(build("10.43.5.5"))
+	// Re-render as if the primary pod had restarted onto a new pod IP. Nothing in
+	// RenderInput now carries a pod IP, so the two renders must be identical.
+	p2, s2 := RenderNamedConf(build("10.43.5.5"))
+	if p1 != p2 {
+		t.Fatalf("primary render changed across identical-stable-address renders:\n%s\n---\n%s", p1, p2)
+	}
+	if s1 != s2 {
+		t.Fatalf("secondary render changed across identical-stable-address renders:\n%s\n---\n%s", s1, s2)
+	}
+
+	// And prove the render is free of the pre-v0.2.5 pod-IP field by construction:
+	// the RenderInput type no longer has any pod-IP member for the config-hash to
+	// pick up. The allow-notify clause carries a key name only.
+	if strings.Contains(s1, "10.42.") {
+		t.Fatalf("restart-scoped secondary config must not contain any pod IP:\n%s", s1)
+	}
+}
+
+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),
@@ -105,3 +230,34 @@ func TestCatalogHashStable(t *testing.T) {
 		t.Fatalf("expected 40-char hex sha1, got %d: %s", len(h1), h1)
 	}
 }
+
+func TestRenderDeterministicWithShuffledForwards(t *testing.T) {
+	// client.List order is non-deterministic; the render must not depend on
+	// input order, or the ConfigMap churns and (with the config hash) the
+	// StatefulSet rolls forever.
+	mkFwd := func(zone, fwd string) bindv1alpha1.BindZone {
+		return bindv1alpha1.BindZone{
+			ObjectMeta: metav1.ObjectMeta{Name: zone},
+			Spec: bindv1alpha1.BindZoneSpec{
+				ClusterRef: "r", Type: bindv1alpha1.ZoneForward,
+				ZoneName: zone, Forwarders: []string{fwd},
+			},
+		}
+	}
+	base := RenderInput{Cluster: newCluster(bindv1alpha1.ModeResolver)}
+	orderA := base
+	orderA.Forwards = []bindv1alpha1.BindZone{
+		mkFwd("unkin.net", "198.18.200.6"), mkFwd("consul", "198.18.19.14"),
+		mkFwd("k8s.syd1.au.unkin.net", "198.18.200.8"), mkFwd("13.18.198.in-addr.arpa", "198.18.200.6"),
+	}
+	orderB := base
+	orderB.Forwards = []bindv1alpha1.BindZone{
+		mkFwd("13.18.198.in-addr.arpa", "198.18.200.6"), mkFwd("k8s.syd1.au.unkin.net", "198.18.200.8"),
+		mkFwd("consul", "198.18.19.14"), mkFwd("unkin.net", "198.18.200.6"),
+	}
+	pa, _ := RenderNamedConf(orderA)
+	pb, _ := RenderNamedConf(orderB)
+	if pa != pb {
+		t.Fatalf("render must be independent of forward-zone input order:\n--- A ---\n%s\n--- B ---\n%s", pa, pb)
+	}
+}

internal/bind/seed.go

@@ -35,13 +35,19 @@ func (e *Executor) ZoneExists(ctx context.Context, namespace, pod, zone, view st
 func (e *Executor) WriteSeedZone(ctx context.Context, namespace, pod, zone, path, primaryIP string, serial int64) error {
 	origin := dot(zone)
 	ns := "ns1." + origin
+	// Short refresh/retry so a secondary that misses a NOTIFY (e.g. its pod IP
+	// changed and the primary's also-notify was briefly stale) still converges
+	// in minutes, not the hour a 3600s refresh would impose. minimum is the
+	// negative-cache TTL: keep it low so a stale-secondary NXDOMAIN does not
+	// stick in downstream resolvers for long. NOTIFY (also-notify on the
+	// primary) remains the fast path; these are the fallback.
 	content := fmt.Sprintf(`$TTL 3600
 @ IN SOA %s hostmaster.%s (
     %d ; serial
-    3600       ; refresh
+    300        ; refresh
-    900        ; retry
+    60         ; retry
     1209600    ; expire
-    300 )      ; minimum
+    60 )       ; minimum
 @ IN NS %s
 ns1 IN A %s
 `, ns, origin, serial, ns, primaryIP)

internal/controller/bindcluster_controller.go

@@ -1,7 +1,10 @@
 package controller
 
 import (
+	"bytes"
 	"context"
+	"crypto/sha256"
+	"encoding/hex"
 	"fmt"
 	"sort"
 	"strings"
@@ -128,7 +131,15 @@ 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
@@ -160,10 +171,11 @@ func (r *BindClusterReconciler) reconcileKeysSecret(ctx context.Context, c *bind
 }
 
 func (r *BindClusterReconciler) reconcileConfigMap(ctx context.Context, c *bindv1alpha1.BindCluster) error {
-	// BIND primaries/default-primaries need the primary's IP address, not a DNS
+	// BIND primaries/default-primaries need an IP address, not a DNS name. Use
-	// name, so render with pod-0's current IP (empty until it is scheduled; the
+	// the stable primary Service ClusterIP so secondaries keep transferring
-	// Pod watch re-renders when it appears or changes).
+	// across primary pod restarts (falls back to the pod IP when no primary
-	in := bind.RenderInput{Cluster: c, PrimaryAddress: primaryPodIP(ctx, r.Client, c)}
+	// Service exists; the Pod/Service watches re-render when it changes).
+	in := bind.RenderInput{Cluster: c, PrimaryAddress: primaryTransferAddress(ctx, r.Client, c)}
 
 	var acls bindv1alpha1.BindACLList
 	if err := r.List(ctx, &acls, client.InNamespace(c.Namespace)); err == nil {
@@ -198,6 +210,17 @@ 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 {
@@ -208,6 +231,16 @@ func (r *BindClusterReconciler) reconcileConfigMap(ctx context.Context, c *bindv
 		}
 	}
 
+	// Secondaries accept intra-cluster NOTIFYs signed with the cluster's catalog
+	// transfer TSIG key (the primary signs its also-notify NOTIFYs with it — see
+	// bindzone_controller). Render `allow-notify { key "<name>"; }` — a static key
+	// element, NO pod IPs — so it never changes on pod churn and cannot re-render
+	// the restart-scoped config (the v0.2.5 roll loop). Resolve the catalog's
+	// TransferKeyRef to the BIND key name (KeyName override, else the ref).
+	if in.Catalog != nil && in.Catalog.Spec.TransferKeyRef != "" {
+		in.NotifyKeyName = tsigKeyName(ctx, r.Client, c.Namespace, in.Catalog.Spec.TransferKeyRef)
+	}
+
 	primaryConf, secondaryConf := bind.RenderNamedConf(in)
 	data := map[string]string{
 		"named.conf.primary":   primaryConf,
@@ -254,7 +287,56 @@ 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) {
@@ -282,6 +364,12 @@ func (r *BindClusterReconciler) reconcileStatefulSet(ctx context.Context, c *bin
 		}}},
 	}
 
+	// Pods copy config from the projected volume into an emptyDir once at
+	// startup; a ConfigMap/keys change never reaches a running pod (rndc
+	// reconfig re-reads the stale startup copy). Stamp a hash of the projected
+	// config onto the pod template so a config change rolls the StatefulSet,
+	// which is the only way the change takes effect. The operator owns the
+	// template, so this restart is operator-driven and not reverted.
 	sts := &appsv1.StatefulSet{
 		ObjectMeta: metav1.ObjectMeta{Name: c.Name, Namespace: c.Namespace, Labels: labels},
 		Spec: appsv1.StatefulSetSpec{
@@ -289,7 +377,7 @@ func (r *BindClusterReconciler) reconcileStatefulSet(ctx context.Context, c *bin
 			Replicas:    &replicas,
 			Selector:    &metav1.LabelSelector{MatchLabels: labels},
 			Template: corev1.PodTemplateSpec{
-				ObjectMeta: metav1.ObjectMeta{Labels: labels},
+				ObjectMeta: metav1.ObjectMeta{Labels: labels, Annotations: map[string]string{configHashAnnotation: r.configHash(ctx, c)}},
 				Spec: corev1.PodSpec{
 					NodeSelector: c.Spec.NodeSelector,
 					Tolerations:  c.Spec.Tolerations,
@@ -357,6 +445,45 @@ func (r *BindClusterReconciler) reconcileStatefulSet(ctx context.Context, c *bin
 	return &existing, nil
 }
 
+// configHashAnnotation carries a hash of the projected config on the pod
+// template; changing it triggers a rolling restart so pods pick up new config.
+const configHashAnnotation = "bind.unkin.net/config-hash"
+
+// configHash returns a deterministic hash of the config projected into the pods
+// (the rendered ConfigMap and the keys.conf Secret). It is read after those are
+// reconciled, so it reflects the current desired config. A stable hash means no
+// spurious restarts; any config or TSIG-key change flips it and rolls the pods.
+func (r *BindClusterReconciler) configHash(ctx context.Context, c *bindv1alpha1.BindCluster) string {
+	var buf bytes.Buffer
+	var cm corev1.ConfigMap
+	if err := r.Get(ctx, types.NamespacedName{Namespace: c.Namespace, Name: configMapName(c.Name)}, &cm); err == nil {
+		for _, k := range sortedKeys(cm.Data) {
+			fmt.Fprintf(&buf, "%s\x00%s\x00", k, cm.Data[k])
+		}
+	}
+	var keys corev1.Secret
+	if err := r.Get(ctx, types.NamespacedName{Namespace: c.Namespace, Name: keysSecretName(c.Name)}, &keys); err == nil {
+		data := make(map[string]string, len(keys.Data))
+		for k, v := range keys.Data {
+			data[k] = string(v)
+		}
+		for _, k := range sortedKeys(data) {
+			fmt.Fprintf(&buf, "%s\x00%s\x00", k, data[k])
+		}
+	}
+	sum := sha256.Sum256(buf.Bytes())
+	return hex.EncodeToString(sum[:])
+}
+
+func sortedKeys(m map[string]string) []string {
+	keys := make([]string, 0, len(m))
+	for k := range m {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	return keys
+}
+
 func (r *BindClusterReconciler) reloadReadyPods(ctx context.Context, c *bindv1alpha1.BindCluster) {
 	if r.Exec == nil {
 		return
@@ -408,6 +535,15 @@ 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/bindtsigapi_controller.go

@@ -0,0 +1,274 @@
+package controller
+
+import (
+	"context"
+	"fmt"
+
+	appsv1 "k8s.io/api/apps/v1"
+	corev1 "k8s.io/api/core/v1"
+	rbacv1 "k8s.io/api/rbac/v1"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/apimachinery/pkg/types"
+	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/log"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+	"git.unkin.net/unkin/bind-operator/internal/bind"
+)
+
+// BindTSIGAPIReconciler deploys the companion TSIG API (Deployment, Service,
+// ConfigMap, token Secret and RBAC) when a BindTSIGAPI resource exists.
+type BindTSIGAPIReconciler struct {
+	client.Client
+	Scheme *runtime.Scheme
+}
+
+// +kubebuilder:rbac:groups=bind.unkin.net,resources=bindtsigapis,verbs=get;list;watch;create;update;patch;delete
+// +kubebuilder:rbac:groups=bind.unkin.net,resources=bindtsigapis/status,verbs=get;update;patch
+// +kubebuilder:rbac:groups=apps,resources=deployments,verbs=get;list;watch;create;update;patch;delete
+// +kubebuilder:rbac:groups="",resources=serviceaccounts,verbs=get;list;watch;create;update;patch;delete
+// +kubebuilder:rbac:groups=rbac.authorization.k8s.io,resources=roles;rolebindings,verbs=get;list;watch;create;update;patch;delete
+
+func (r *BindTSIGAPIReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
+	logger := log.FromContext(ctx)
+
+	var api bindv1alpha1.BindTSIGAPI
+	if err := r.Get(ctx, req.NamespacedName, &api); err != nil {
+		return ctrl.Result{}, client.IgnoreNotFound(err)
+	}
+
+	tokenSecret := api.Spec.TokenSecretName
+	if tokenSecret == "" {
+		tokenSecret = api.Name + "-token"
+	}
+	targetNS := api.Spec.TargetNamespace
+	if targetNS == "" {
+		targetNS = api.Namespace
+	}
+
+	for _, step := range []func(context.Context, *bindv1alpha1.BindTSIGAPI, string, string) error{
+		r.reconcileServiceAccount,
+		r.reconcileRBAC,
+		r.reconcileTokenSecret,
+		r.reconcileConfigMap,
+		r.reconcileDeployment,
+		r.reconcileService,
+	} {
+		if err := step(ctx, &api, tokenSecret, targetNS); err != nil {
+			return ctrl.Result{}, err
+		}
+	}
+
+	// Status from the Deployment.
+	var dep appsv1.Deployment
+	_ = r.Get(ctx, types.NamespacedName{Namespace: api.Namespace, Name: api.Name}, &dep)
+	port := api.Spec.Port
+	if port == 0 {
+		port = 8443
+	}
+	api.Status.ReadyReplicas = dep.Status.ReadyReplicas
+	api.Status.Endpoint = fmt.Sprintf("http://%s.%s.svc:%d", api.Name, api.Namespace, port)
+	api.Status.TokenSecret = tokenSecret
+	api.Status.ObservedGeneration = api.Generation
+	ready := dep.Status.ReadyReplicas > 0
+	if ready {
+		api.Status.Phase = "Ready"
+	} else {
+		api.Status.Phase = "Progressing"
+	}
+	setReady(&api.Status.Conditions, api.Generation, ready, "Reconciled", fmt.Sprintf("%d ready", dep.Status.ReadyReplicas))
+	if err := r.Status().Update(ctx, &api); err != nil {
+		return ctrl.Result{}, err
+	}
+	if !ready {
+		return ctrl.Result{RequeueAfter: requeueShort}, nil
+	}
+	logger.V(1).Info("tsig api reconciled", "api", api.Name)
+	return ctrl.Result{}, nil
+}
+
+func tsigAPILabels(name string) map[string]string {
+	return map[string]string{
+		managedByLabel:                managedByValue,
+		"app.kubernetes.io/name":      "bind-tsig-api",
+		"app.kubernetes.io/instance":  name,
+		"app.kubernetes.io/component": "tsig-api",
+	}
+}
+
+func (r *BindTSIGAPIReconciler) reconcileServiceAccount(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, _, _ string) error {
+	sa := &corev1.ServiceAccount{ObjectMeta: metav1.ObjectMeta{Name: api.Name, Namespace: api.Namespace, Labels: tsigAPILabels(api.Name)}}
+	return r.apply(ctx, api, sa, func() {})
+}
+
+func (r *BindTSIGAPIReconciler) reconcileRBAC(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, _, targetNS string) error {
+	role := &rbacv1.Role{ObjectMeta: metav1.ObjectMeta{Name: api.Name, Namespace: targetNS, Labels: tsigAPILabels(api.Name)}}
+	if err := r.applyIn(ctx, api, role, targetNS, func() {
+		role.Rules = []rbacv1.PolicyRule{
+			{APIGroups: []string{"bind.unkin.net"}, Resources: []string{"bindtsigkeys"}, Verbs: []string{"get", "list", "watch", "create", "update", "patch", "delete"}},
+			{APIGroups: []string{"bind.unkin.net"}, Resources: []string{"bindtsigkeys/status"}, Verbs: []string{"get"}},
+			{APIGroups: []string{""}, Resources: []string{"secrets"}, Verbs: []string{"get", "list", "watch", "delete"}},
+		}
+	}); err != nil {
+		return err
+	}
+	rb := &rbacv1.RoleBinding{ObjectMeta: metav1.ObjectMeta{Name: api.Name, Namespace: targetNS, Labels: tsigAPILabels(api.Name)}}
+	return r.applyIn(ctx, api, rb, targetNS, func() {
+		rb.RoleRef = rbacv1.RoleRef{APIGroup: "rbac.authorization.k8s.io", Kind: "Role", Name: api.Name}
+		rb.Subjects = []rbacv1.Subject{{Kind: "ServiceAccount", Name: api.Name, Namespace: api.Namespace}}
+	})
+}
+
+// reconcileTokenSecret creates the master-access-token Secret only when it does
+// not already exist, so a VaultStaticSecret may pre-seed it. Not owned by the
+// BindTSIGAPI, so it survives and stays overwritable.
+func (r *BindTSIGAPIReconciler) reconcileTokenSecret(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, tokenSecret, _ string) error {
+	var existing corev1.Secret
+	err := r.Get(ctx, types.NamespacedName{Namespace: api.Namespace, Name: tokenSecret}, &existing)
+	if err == nil {
+		return nil
+	}
+	if !apierrors.IsNotFound(err) {
+		return err
+	}
+	token, genErr := bind.GenerateSecret(32)
+	if genErr != nil {
+		return genErr
+	}
+	s := &corev1.Secret{
+		ObjectMeta: metav1.ObjectMeta{Name: tokenSecret, Namespace: api.Namespace, Labels: tsigAPILabels(api.Name)},
+		Data:       map[string][]byte{"token": []byte(token)},
+	}
+	return r.Create(ctx, s)
+}
+
+func (r *BindTSIGAPIReconciler) reconcileConfigMap(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, _, targetNS string) error {
+	cm := &corev1.ConfigMap{ObjectMeta: metav1.ObjectMeta{Name: api.Name + "-config", Namespace: api.Namespace, Labels: tsigAPILabels(api.Name)}}
+	return r.apply(ctx, api, cm, func() {
+		port := api.Spec.Port
+		if port == 0 {
+			port = 8443
+		}
+		data := map[string]string{
+			"LISTEN_ADDR":      fmt.Sprintf(":%d", port),
+			"TARGET_NAMESPACE": targetNS,
+		}
+		for k, v := range api.Spec.Env {
+			data[k] = v
+		}
+		cm.Data = data
+	})
+}
+
+func (r *BindTSIGAPIReconciler) reconcileDeployment(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, tokenSecret, _ string) error {
+	dep := &appsv1.Deployment{ObjectMeta: metav1.ObjectMeta{Name: api.Name, Namespace: api.Namespace, Labels: tsigAPILabels(api.Name)}}
+	return r.apply(ctx, api, dep, func() {
+		replicas := api.Spec.Replicas
+		if replicas == 0 {
+			replicas = 1
+		}
+		image := api.Spec.Image
+		if image == "" {
+			image = "git.unkin.net/unkin/bind-tsig-api:latest"
+		}
+		port := api.Spec.Port
+		if port == 0 {
+			port = 8443
+		}
+		labels := tsigAPILabels(api.Name)
+		dep.Spec = appsv1.DeploymentSpec{
+			Replicas: &replicas,
+			Selector: &metav1.LabelSelector{MatchLabels: map[string]string{"app.kubernetes.io/instance": api.Name}},
+			Template: corev1.PodTemplateSpec{
+				ObjectMeta: metav1.ObjectMeta{Labels: labels},
+				Spec: corev1.PodSpec{
+					ServiceAccountName: api.Name,
+					SecurityContext:    &corev1.PodSecurityContext{RunAsNonRoot: ptr(true)},
+					Containers: []corev1.Container{{
+						Name:            "tsig-api",
+						Image:           image,
+						ImagePullPolicy: api.Spec.ImagePullPolicy,
+						Command:         []string{"tsig-api"},
+						Ports:           []corev1.ContainerPort{{Name: "https", ContainerPort: port}},
+						EnvFrom:         []corev1.EnvFromSource{{ConfigMapRef: &corev1.ConfigMapEnvSource{LocalObjectReference: corev1.LocalObjectReference{Name: api.Name + "-config"}}}},
+						Env: []corev1.EnvVar{{
+							Name:      "API_TOKEN",
+							ValueFrom: &corev1.EnvVarSource{SecretKeyRef: &corev1.SecretKeySelector{LocalObjectReference: corev1.LocalObjectReference{Name: tokenSecret}, Key: "token"}},
+						}},
+						Resources: api.Spec.Resources,
+						ReadinessProbe: &corev1.Probe{
+							ProbeHandler:        corev1.ProbeHandler{HTTPGet: &corev1.HTTPGetAction{Path: "/healthz", Port: intstrFromInt(int(port))}},
+							InitialDelaySeconds: 5, PeriodSeconds: 10,
+						},
+						SecurityContext: &corev1.SecurityContext{
+							AllowPrivilegeEscalation: ptr(false),
+							ReadOnlyRootFilesystem:   ptr(true),
+							Capabilities:             &corev1.Capabilities{Drop: []corev1.Capability{"ALL"}},
+						},
+					}},
+				},
+			},
+		}
+	})
+}
+
+func (r *BindTSIGAPIReconciler) reconcileService(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, _, _ string) error {
+	svc := &corev1.Service{ObjectMeta: metav1.ObjectMeta{Name: api.Name, Namespace: api.Namespace, Labels: tsigAPILabels(api.Name), Annotations: api.Spec.Service.Annotations}}
+	return r.apply(ctx, api, svc, func() {
+		port := api.Spec.Port
+		if port == 0 {
+			port = 8443
+		}
+		svcType := api.Spec.Service.Type
+		if svcType == "" {
+			svcType = corev1.ServiceTypeClusterIP
+		}
+		svc.Spec.Type = svcType
+		svc.Spec.Selector = map[string]string{"app.kubernetes.io/instance": api.Name}
+		svc.Spec.Ports = []corev1.ServicePort{{Name: "https", Port: port, TargetPort: intstrFromInt(int(port))}}
+		if api.Spec.Service.LoadBalancerIP != "" {
+			svc.Spec.LoadBalancerIP = api.Spec.Service.LoadBalancerIP
+		}
+	})
+}
+
+// apply creates or updates an owned object (in the BindTSIGAPI's namespace).
+func (r *BindTSIGAPIReconciler) apply(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, obj client.Object, mutate func()) error {
+	return r.applyIn(ctx, api, obj, api.Namespace, mutate)
+}
+
+// applyIn creates or updates an object; ownership is set only for objects in
+// the BindTSIGAPI's own namespace (cross-namespace owner refs are not allowed).
+func (r *BindTSIGAPIReconciler) applyIn(ctx context.Context, api *bindv1alpha1.BindTSIGAPI, obj client.Object, namespace string, mutate func()) error {
+	key := client.ObjectKeyFromObject(obj)
+	err := r.Get(ctx, key, obj)
+	if apierrors.IsNotFound(err) {
+		mutate()
+		if namespace == api.Namespace {
+			if serr := ctrl.SetControllerReference(api, obj, r.Scheme); serr != nil {
+				return serr
+			}
+		}
+		return r.Create(ctx, obj)
+	}
+	if err != nil {
+		return err
+	}
+	mutate()
+	return r.Update(ctx, obj)
+}
+
+func (r *BindTSIGAPIReconciler) SetupWithManager(mgr ctrl.Manager) error {
+	return ctrl.NewControllerManagedBy(mgr).
+		For(&bindv1alpha1.BindTSIGAPI{}).
+		Owns(&appsv1.Deployment{}).
+		Owns(&corev1.Service{}).
+		Owns(&corev1.ConfigMap{}).
+		Owns(&corev1.ServiceAccount{}).
+		Complete(r)
+}
+
+func ptr[T any](v T) *T { return &v }

internal/controller/bindtsigkey_controller.go

@@ -60,7 +60,7 @@ func (r *BindTSIGKeyReconciler) Reconcile(ctx context.Context, req ctrl.Request)
 			return ctrl.Result{}, genErr
 		}
 		newSecret := &corev1.Secret{
-			ObjectMeta: metav1.ObjectMeta{Name: secretName, Namespace: key.Namespace, Labels: map[string]string{managedByLabel: managedByValue}},
+			ObjectMeta: metav1.ObjectMeta{Name: secretName, Namespace: key.Namespace},
 			Data: map[string][]byte{
 				"algorithm": []byte(algorithm),
 				"keyName":   []byte(keyName),
@@ -68,6 +68,7 @@ func (r *BindTSIGKeyReconciler) Reconcile(ctx context.Context, req ctrl.Request)
 				"key.conf":  []byte(bind.KeyClause(keyName, algorithm, material)),
 			},
 		}
+		applySecretTemplate(&newSecret.ObjectMeta, key.Spec.SecretTemplate)
 		if err := ctrl.SetControllerReference(&key, newSecret, r.Scheme); err != nil {
 			return ctrl.Result{}, err
 		}
@@ -77,6 +78,21 @@ func (r *BindTSIGKeyReconciler) Reconcile(ctx context.Context, req ctrl.Request)
 		logger.Info("generated TSIG key", "key", key.Name, "secret", secretName)
 	case err != nil:
 		return ctrl.Result{}, err
+	default:
+		// Secret already exists: reconcile the template-managed metadata so that
+		// annotation/label changes on the CR (e.g. reflection hints) propagate
+		// without regenerating key material. Skip imported secrets, which are
+		// owned by an external manager (Vault/VSO, reflector) that we must not
+		// fight over metadata.
+		if key.Spec.ImportExisting {
+			break
+		}
+		if updated := applySecretTemplate(&secret.ObjectMeta, key.Spec.SecretTemplate); updated {
+			if err := r.Update(ctx, &secret); err != nil {
+				return ctrl.Result{}, err
+			}
+			logger.Info("updated TSIG key secret metadata", "key", key.Name, "secret", secretName)
+		}
 	}
 
 	key.Status.SecretName = secretName
@@ -90,6 +106,42 @@ func (r *BindTSIGKeyReconciler) Reconcile(ctx context.Context, req ctrl.Request)
 	return ctrl.Result{}, nil
 }
 
+// applySecretTemplate stamps the operator-managed label plus any
+// user-supplied labels/annotations onto the Secret's metadata. It returns true
+// if it mutated meta, so callers can decide whether an update is needed.
+func applySecretTemplate(meta *metav1.ObjectMeta, tmpl *bindv1alpha1.SecretMetadata) bool {
+	changed := false
+	setLabel := func(k, v string) {
+		if meta.Labels == nil {
+			meta.Labels = map[string]string{}
+		}
+		if meta.Labels[k] != v {
+			meta.Labels[k] = v
+			changed = true
+		}
+	}
+	setAnnotation := func(k, v string) {
+		if meta.Annotations == nil {
+			meta.Annotations = map[string]string{}
+		}
+		if meta.Annotations[k] != v {
+			meta.Annotations[k] = v
+			changed = true
+		}
+	}
+
+	setLabel(managedByLabel, managedByValue)
+	if tmpl != nil {
+		for k, v := range tmpl.Labels {
+			setLabel(k, v)
+		}
+		for k, v := range tmpl.Annotations {
+			setAnnotation(k, v)
+		}
+	}
+	return changed
+}
+
 func (r *BindTSIGKeyReconciler) fail(ctx context.Context, key *bindv1alpha1.BindTSIGKey, reason, msg string) (ctrl.Result, error) {
 	key.Status.Ready = false
 	key.Status.ObservedGeneration = key.Generation

internal/controller/bindtsigkey_test.go

@@ -0,0 +1,71 @@
+package controller
+
+import (
+	"testing"
+
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+)
+
+func TestApplySecretTemplate(t *testing.T) {
+	t.Run("nil template still stamps managed-by label", func(t *testing.T) {
+		var meta metav1.ObjectMeta
+		if !applySecretTemplate(&meta, nil) {
+			t.Fatal("expected change on empty meta")
+		}
+		if meta.Labels[managedByLabel] != managedByValue {
+			t.Errorf("managed-by label = %q, want %q", meta.Labels[managedByLabel], managedByValue)
+		}
+		if meta.Annotations != nil {
+			t.Errorf("annotations = %v, want nil", meta.Annotations)
+		}
+	})
+
+	t.Run("applies labels and annotations", func(t *testing.T) {
+		meta := metav1.ObjectMeta{Labels: map[string]string{managedByLabel: managedByValue}}
+		tmpl := &bindv1alpha1.SecretMetadata{
+			Annotations: map[string]string{"reflector.v1.k8s.emberstack.com/reflection-allowed": "true"},
+			Labels:      map[string]string{"team": "dns"},
+		}
+		if !applySecretTemplate(&meta, tmpl) {
+			t.Fatal("expected change when adding template metadata")
+		}
+		if got := meta.Annotations["reflector.v1.k8s.emberstack.com/reflection-allowed"]; got != "true" {
+			t.Errorf("reflection annotation = %q, want true", got)
+		}
+		if meta.Labels["team"] != "dns" {
+			t.Errorf("team label = %q, want dns", meta.Labels["team"])
+		}
+		// managed-by must survive user-supplied labels.
+		if meta.Labels[managedByLabel] != managedByValue {
+			t.Errorf("managed-by label dropped: %v", meta.Labels)
+		}
+	})
+
+	t.Run("idempotent when already applied", func(t *testing.T) {
+		tmpl := &bindv1alpha1.SecretMetadata{
+			Annotations: map[string]string{"a": "1"},
+			Labels:      map[string]string{"b": "2"},
+		}
+		meta := metav1.ObjectMeta{}
+		applySecretTemplate(&meta, tmpl)
+		if applySecretTemplate(&meta, tmpl) {
+			t.Error("expected no change on second apply")
+		}
+	})
+
+	t.Run("updates drifted annotation value", func(t *testing.T) {
+		meta := metav1.ObjectMeta{
+			Labels:      map[string]string{managedByLabel: managedByValue},
+			Annotations: map[string]string{"a": "old"},
+		}
+		tmpl := &bindv1alpha1.SecretMetadata{Annotations: map[string]string{"a": "new"}}
+		if !applySecretTemplate(&meta, tmpl) {
+			t.Fatal("expected change when annotation value drifts")
+		}
+		if meta.Annotations["a"] != "new" {
+			t.Errorf("annotation a = %q, want new", meta.Annotations["a"])
+		}
+	})
+}

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())
@@ -67,7 +80,21 @@ func (r *BindZoneReconciler) Reconcile(ctx context.Context, req ctrl.Request) (c
 		return r.setPhase(ctx, &zone, "Pending", "PrimaryNotReady", "waiting for cluster primary to be ready")
 	}
 
-	zoneConfig, err := r.buildZoneConfig(ctx, &zone)
+	// Primary zones replicated to secondaries (catalog members) get an
+	// also-notify pointing at the secondary pods, so a dynamic update NOTIFYs
+	// them immediately rather than waiting for the SOA refresh.
+	var notifyTargets []string
+	if isPrimaryType(zone.Spec.Type) && catalogEnabled(&zone) {
+		notifyTargets = secondaryPodIPs(ctx, r.Client, cluster)
+	}
+
+	// The catalog transfer TSIG key doubles as the intra-cluster NOTIFY key: the
+	// primary signs its also-notify NOTIFYs with it and secondaries accept them
+	// via `allow-notify { key "<key>"; }`. Keying the NOTIFYs is what lets the
+	// secondary's allow-notify be a static key element (no pod IPs), so pod-IP
+	// churn never re-renders restart-scoped config (the v0.2.5 roll loop).
+	transferKey := r.zoneTransferKeyRef(ctx, &zone, cluster)
+	zoneConfig, err := r.buildZoneConfig(ctx, &zone, transferKey, notifyTargets, transferKey)
 	if err != nil {
 		return r.setPhase(ctx, &zone, "Error", "ConfigError", err.Error())
 	}
@@ -120,7 +147,11 @@ func (r *BindZoneReconciler) Reconcile(ctx context.Context, req ctrl.Request) (c
 }
 
 // buildZoneConfig renders the inner clause passed to rndc addzone/modzone.
-func (r *BindZoneReconciler) buildZoneConfig(ctx context.Context, zone *bindv1alpha1.BindZone) (string, error) {
+// transferKey, when set, is the catalog transfer TSIG key name; catalog member
+// primary zones must allow AXFR with it so secondaries can pull them. notifyKey,
+// when set, is the TSIG key each also-notify entry is signed with, so
+// secondaries can accept the NOTIFYs by key rather than by (churning) pod IP.
+func (r *BindZoneReconciler) buildZoneConfig(ctx context.Context, zone *bindv1alpha1.BindZone, transferKey string, notifyTargets []string, notifyKey string) (string, error) {
 	zType := zone.Spec.Type
 	if zType == "" {
 		zType = bindv1alpha1.ZonePrimary
@@ -132,8 +163,22 @@ func (r *BindZoneReconciler) buildZoneConfig(ctx context.Context, zone *bindv1al
 		if zone.Spec.DynamicUpdate && zone.Spec.UpdateKeyRef != "" {
 			parts = append(parts, fmt.Sprintf("allow-update { key \"%s\"; }", updateKeyName(ctx, r.Client, zone)))
 		}
-		if len(zone.Spec.AllowTransfer) > 0 {
+		switch {
+		case len(zone.Spec.AllowTransfer) > 0:
 			parts = append(parts, fmt.Sprintf("allow-transfer { %s }", matchListInline(zone.Spec.AllowTransfer)))
+		case transferKey != "":
+			// Catalog member: permit key-authenticated AXFR from secondaries.
+			parts = append(parts, fmt.Sprintf("allow-transfer { key \"%s\"; }", transferKey))
+		}
+		// NOTIFY only the secondaries we know about (their apex NS is the primary
+		// itself, so default `notify yes` would reach no one). `notify explicit`
+		// keeps NOTIFY off the query-serving VIP and scoped to the pod IPs. Each
+		// entry is signed with notifyKey so secondaries can admit the NOTIFY by key
+		// (`allow-notify { key ... }`) instead of by pod IP. This zone config is
+		// applied via rndc addzone/modzone — no pod restart — so listing pod IPs
+		// here is safe; only *restart-scoped* config must never depend on pod IPs.
+		if len(notifyTargets) > 0 {
+			parts = append(parts, "notify explicit", fmt.Sprintf("also-notify { %s }", alsoNotifyList(notifyTargets, notifyKey)))
 		}
 		if zone.Spec.DNSSECPolicyRef != "" {
 			parts = append(parts, fmt.Sprintf("dnssec-policy \"%s\"", zone.Spec.DNSSECPolicyRef), "inline-signing yes")
@@ -189,6 +234,26 @@ func (r *BindZoneReconciler) deregisterCatalog(ctx context.Context, zone *bindv1
 	_ = r.Exec.RemoveCatalogMember(ctx, zone.Namespace, primaryPod, catalog.Spec.ZoneName, zone.Spec.ZoneName, creds)
 }
 
+// zoneTransferKeyRef returns the catalog transfer TSIG key name that a catalog
+// member primary zone must allow AXFR with, so secondaries (which present that
+// key) can pull it. Returns "" for non-member zones, non-primary zones, or when
+// the cluster has no catalog.
+func (r *BindZoneReconciler) zoneTransferKeyRef(ctx context.Context, zone *bindv1alpha1.BindZone, cluster *bindv1alpha1.BindCluster) string {
+	if !isPrimaryType(zone.Spec.Type) || !catalogEnabled(zone) {
+		return ""
+	}
+	var catalogs bindv1alpha1.BindCatalogZoneList
+	if err := r.List(ctx, &catalogs, client.InNamespace(zone.Namespace)); err != nil {
+		return ""
+	}
+	for i := range catalogs.Items {
+		if catalogs.Items[i].Spec.ClusterRef == cluster.Name {
+			return catalogs.Items[i].Spec.TransferKeyRef
+		}
+	}
+	return ""
+}
+
 func (r *BindZoneReconciler) catalogFor(ctx context.Context, zone *bindv1alpha1.BindZone, cluster *bindv1alpha1.BindCluster) (*bindv1alpha1.BindCatalogZone, bind.TSIGCreds, bool) {
 	var catalogs bindv1alpha1.BindCatalogZoneList
 	if err := r.List(ctx, &catalogs, client.InNamespace(zone.Namespace)); err != nil {

internal/controller/bindzone_notify_test.go

@@ -0,0 +1,65 @@
+package controller
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+)
+
+// A primary zone with known secondaries renders notify explicit + also-notify
+// so a dynamic update NOTIFYs the secondaries immediately.
+func TestBuildZoneConfigPrimaryAlsoNotify(t *testing.T) {
+	r := &BindZoneReconciler{}
+	zone := &bindv1alpha1.BindZone{
+		Spec: bindv1alpha1.BindZoneSpec{
+			ZoneName: "main.unkin.net",
+			Type:     bindv1alpha1.ZonePrimary,
+		},
+	}
+
+	cfg, err := r.buildZoneConfig(context.Background(), zone, "transfer-key", []string{"10.42.2.6", "10.42.1.5"}, "externaldns-key")
+	if err != nil {
+		t.Fatalf("buildZoneConfig: %v", err)
+	}
+	if !strings.Contains(cfg, "notify explicit") {
+		t.Errorf("expected notify explicit in %q", cfg)
+	}
+	// also-notify entries are keyed so secondaries can admit the NOTIFY by TSIG
+	// key (allow-notify { key ... }) instead of by (churning) pod IP.
+	if !strings.Contains(cfg, `also-notify { 10.42.2.6 key "externaldns-key"; 10.42.1.5 key "externaldns-key"; }`) {
+		t.Errorf("expected keyed also-notify with the secondary IPs in %q", cfg)
+	}
+}
+
+// also-notify entries carry no key when none is configured (unkeyed NOTIFY).
+func TestBuildZoneConfigPrimaryAlsoNotifyUnkeyed(t *testing.T) {
+	r := &BindZoneReconciler{}
+	zone := &bindv1alpha1.BindZone{
+		Spec: bindv1alpha1.BindZoneSpec{ZoneName: "main.unkin.net", Type: bindv1alpha1.ZonePrimary},
+	}
+	cfg, err := r.buildZoneConfig(context.Background(), zone, "transfer-key", []string{"10.42.2.6"}, "")
+	if err != nil {
+		t.Fatalf("buildZoneConfig: %v", err)
+	}
+	if !strings.Contains(cfg, "also-notify { 10.42.2.6; }") {
+		t.Errorf("expected unkeyed also-notify in %q", cfg)
+	}
+}
+
+// With no secondaries, no also-notify is emitted (single-replica cluster).
+func TestBuildZoneConfigPrimaryNoNotifyTargets(t *testing.T) {
+	r := &BindZoneReconciler{}
+	zone := &bindv1alpha1.BindZone{
+		Spec: bindv1alpha1.BindZoneSpec{ZoneName: "main.unkin.net", Type: bindv1alpha1.ZonePrimary},
+	}
+
+	cfg, err := r.buildZoneConfig(context.Background(), zone, "transfer-key", nil, "externaldns-key")
+	if err != nil {
+		t.Fatalf("buildZoneConfig: %v", err)
+	}
+	if strings.Contains(cfg, "also-notify") || strings.Contains(cfg, "notify explicit") {
+		t.Errorf("did not expect notify clauses with no targets: %q", cfg)
+	}
+}

internal/controller/config_hash_podip_test.go

@@ -0,0 +1,108 @@
+package controller
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/runtime"
+	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/client/fake"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+)
+
+// TestConfigHashIndependentOfPrimaryPodIP is the permanent regression guard for
+// the v0.2.5 rolling-restart loop.
+//
+// v0.2.5 rendered an options-scope `allow-notify { <primaryPodIP>; ... }` into
+// the (restart-scoped) named.conf. The config-hash annotation that rolls the
+// StatefulSet is computed over that render (BindClusterReconciler.configHash), so
+// a config change rolled the pods, the primary pod came back on a NEW IP, the
+// operator re-rendered with the new IP, the hash changed, the pods rolled again,
+// and so on — an infinite roll loop across every BindCluster.
+//
+// The fix removed all pod IPs from restart-scoped config (secondaries now admit
+// intra-cluster NOTIFYs by TSIG key: `allow-notify { key "X"; }`). This test
+// reconciles the ConfigMap with the primary pod on one IP, computes the hash,
+// then does it again with the primary pod on a DIFFERENT IP, and asserts the
+// hash is byte-identical. If any pod-IP dependency ever creeps back into
+// restart-scoped config, this fails and the loop-class bug is caught.
+func TestConfigHashIndependentOfPrimaryPodIP(t *testing.T) {
+	scheme := runtime.NewScheme()
+	if err := clientgoscheme.AddToScheme(scheme); err != nil {
+		t.Fatal(err)
+	}
+	if err := bindv1alpha1.AddToScheme(scheme); err != nil {
+		t.Fatal(err)
+	}
+
+	const ns = "dns"
+	svcIP := "10.43.5.5" // stable primary Service ClusterIP (does NOT churn)
+
+	cluster := &bindv1alpha1.BindCluster{
+		ObjectMeta: metav1.ObjectMeta{Name: "auth", Namespace: ns},
+		Spec: bindv1alpha1.BindClusterSpec{
+			Mode:           bindv1alpha1.ModeAuthoritative,
+			Replicas:       3,
+			PrimaryService: &bindv1alpha1.ClusterServiceSpec{},
+		},
+	}
+	primarySvc := &corev1.Service{
+		ObjectMeta: metav1.ObjectMeta{Name: primaryServiceName(cluster.Name), Namespace: ns},
+		Spec:       corev1.ServiceSpec{ClusterIP: svcIP},
+	}
+	catalog := &bindv1alpha1.BindCatalogZone{
+		ObjectMeta: metav1.ObjectMeta{Name: "cat", Namespace: ns},
+		Spec: bindv1alpha1.BindCatalogZoneSpec{
+			ClusterRef:     cluster.Name,
+			ZoneName:       "catalog.internal",
+			TransferKeyRef: "externaldns-key",
+		},
+	}
+	tsig := &bindv1alpha1.BindTSIGKey{
+		ObjectMeta: metav1.ObjectMeta{Name: "externaldns-key", Namespace: ns},
+		Spec:       bindv1alpha1.BindTSIGKeySpec{ClusterRef: cluster.Name},
+	}
+
+	// hashWithPrimaryIP reconciles the ConfigMap with the primary pod carrying the
+	// given IP, then returns the resulting config-hash.
+	hashWithPrimaryIP := func(ip string) string {
+		primaryPod := &corev1.Pod{
+			ObjectMeta: metav1.ObjectMeta{Name: primaryPodName(cluster.Name), Namespace: ns, Labels: commonLabels(cluster.Name)},
+			Status:     corev1.PodStatus{PodIP: ip},
+		}
+		c := fake.NewClientBuilder().
+			WithScheme(scheme).
+			WithObjects(cluster, primarySvc, catalog, tsig, primaryPod).
+			Build()
+		r := &BindClusterReconciler{Client: c, Scheme: scheme}
+		ctx := context.Background()
+		if err := r.reconcileConfigMap(ctx, cluster); err != nil {
+			t.Fatalf("reconcileConfigMap: %v", err)
+		}
+		// Sanity: the pod IP must not have leaked into the rendered config.
+		var cm corev1.ConfigMap
+		if err := c.Get(ctx, client.ObjectKey{Namespace: ns, Name: configMapName(cluster.Name)}, &cm); err != nil {
+			t.Fatalf("get configmap: %v", err)
+		}
+		for k, v := range cm.Data {
+			if ip != "" && strings.Contains(v, ip) {
+				t.Fatalf("primary pod IP %s leaked into restart-scoped config %s:\n%s", ip, k, v)
+			}
+		}
+		return r.configHash(ctx, cluster)
+	}
+
+	h1 := hashWithPrimaryIP("10.42.3.197")
+	h2 := hashWithPrimaryIP("10.42.9.42") // primary pod restarted onto a new IP
+	if h1 == "" {
+		t.Fatal("config hash should not be empty")
+	}
+	if h1 != h2 {
+		t.Fatalf("config hash changed when only the primary pod IP changed — the v0.2.5 roll loop:\n%s\n%s", h1, h2)
+	}
+}

internal/controller/config_hash_test.go

@@ -0,0 +1,64 @@
+package controller
+
+import (
+	"context"
+	"testing"
+
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/runtime"
+	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+	"sigs.k8s.io/controller-runtime/pkg/client/fake"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+)
+
+func TestConfigHashStableAndSensitive(t *testing.T) {
+	scheme := runtime.NewScheme()
+	if err := clientgoscheme.AddToScheme(scheme); err != nil {
+		t.Fatal(err)
+	}
+	if err := bindv1alpha1.AddToScheme(scheme); err != nil {
+		t.Fatal(err)
+	}
+	cluster := &bindv1alpha1.BindCluster{ObjectMeta: metav1.ObjectMeta{Name: "c", Namespace: "ns"}}
+	cm := &corev1.ConfigMap{
+		ObjectMeta: metav1.ObjectMeta{Name: configMapName("c"), Namespace: "ns"},
+		Data:       map[string]string{"named.conf.secondary": "options { recursion yes; };"},
+	}
+	keys := &corev1.Secret{
+		ObjectMeta: metav1.ObjectMeta{Name: keysSecretName("c"), Namespace: "ns"},
+		Data:       map[string][]byte{"keys.conf": []byte("key x {};")},
+	}
+	c := fake.NewClientBuilder().WithScheme(scheme).WithObjects(cm, keys).Build()
+	r := &BindClusterReconciler{Client: c, Scheme: scheme}
+	ctx := context.Background()
+
+	h1 := r.configHash(ctx, cluster)
+	if h1 == "" {
+		t.Fatal("hash should not be empty when config exists")
+	}
+	// Stable across calls when nothing changes.
+	if h2 := r.configHash(ctx, cluster); h2 != h1 {
+		t.Fatalf("hash not stable: %s != %s", h1, h2)
+	}
+
+	// A config change flips the hash (this is what rolls the StatefulSet).
+	cm.Data["named.conf.secondary"] = "options { recursion yes; validate-except { unkin.net; }; };"
+	if err := c.Update(ctx, cm); err != nil {
+		t.Fatal(err)
+	}
+	if h3 := r.configHash(ctx, cluster); h3 == h1 {
+		t.Fatal("hash must change when the ConfigMap changes")
+	}
+
+	// A TSIG key (keys.conf) change also flips it.
+	afterCM := r.configHash(ctx, cluster)
+	keys.Data["keys.conf"] = []byte("key x { algorithm hmac-sha256; };")
+	if err := c.Update(ctx, keys); err != nil {
+		t.Fatal(err)
+	}
+	if h4 := r.configHash(ctx, cluster); h4 == afterCM {
+		t.Fatal("hash must change when keys.conf changes")
+	}
+}

internal/controller/helpers.go

@@ -3,6 +3,7 @@ package controller
 import (
 	"context"
 	"fmt"
+	"sort"
 	"time"
 
 	corev1 "k8s.io/api/core/v1"
@@ -29,11 +30,20 @@ const (
 )
 
 func headlessServiceName(cluster string) string { return cluster + "-headless" }
-func clientServiceName(cluster string) string   { return cluster }
+func primaryServiceName(cluster string) string  { return cluster + "-primary" }
-func primaryPodName(cluster string) string      { return cluster + "-0" }
+
-func configMapName(cluster string) string       { return cluster + "-config" }
+// primaryPodSelector selects only the primary pod (ordinal 0) via the stable
-func keysSecretName(cluster string) string      { return cluster + "-keys" }
+// StatefulSet pod-name label, for the write Service.
-func rndcSecretName(cluster string) string      { return cluster + "-rndc" }
+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 {
@@ -99,6 +109,52 @@ func primaryPodIP(ctx context.Context, c client.Client, cluster *bindv1alpha1.Bi
 	return pod.Status.PodIP
 }
 
+// primaryTransferAddress returns the address secondaries use to reach the
+// primary for catalog and zone AXFR. It prefers the primary Service ClusterIP,
+// which is stable across primary pod restarts (the pod IP is not: it changes on
+// every restart, leaving secondaries pointed at a dead address). It falls back
+// to the primary pod IP when no primary Service is configured or its ClusterIP
+// is not yet assigned.
+func primaryTransferAddress(ctx context.Context, c client.Client, cluster *bindv1alpha1.BindCluster) string {
+	if cluster.Spec.PrimaryService != nil {
+		var svc corev1.Service
+		if err := c.Get(ctx, client.ObjectKey{Namespace: cluster.Namespace, Name: primaryServiceName(cluster.Name)}, &svc); err == nil {
+			if ip := svc.Spec.ClusterIP; ip != "" && ip != corev1.ClusterIPNone {
+				return ip
+			}
+		}
+	}
+	return primaryPodIP(ctx, c, cluster)
+}
+
+// secondaryPodIPs returns the pod IPs of a cluster's secondary pods (every pod
+// except the ordinal-0 primary) that currently have an address. The primary
+// uses this list as its zone `also-notify` set, so a change to a primary zone
+// (in particular a dynamic update) triggers an immediate NOTIFY -> IXFR to the
+// secondaries instead of leaving them stale until the next SOA refresh. The
+// list is sorted so the rendered zone config is stable and does not churn
+// modzone on every reconcile. Pod IPs change across restarts, so the caller
+// relies on the zone controller's periodic requeue to refresh the set (a
+// restarted secondary re-transfers the whole zone on load regardless). Returns
+// nil for a single-replica cluster.
+func secondaryPodIPs(ctx context.Context, c client.Client, cluster *bindv1alpha1.BindCluster) []string {
+	var pods corev1.PodList
+	if err := c.List(ctx, &pods, client.InNamespace(cluster.Namespace), client.MatchingLabels(commonLabels(cluster.Name))); err != nil {
+		return nil
+	}
+	primary := primaryPodName(cluster.Name)
+	var ips []string
+	for i := range pods.Items {
+		p := &pods.Items[i]
+		if p.Name == primary || p.Status.PodIP == "" {
+			continue
+		}
+		ips = append(ips, p.Status.PodIP)
+	}
+	sort.Strings(ips)
+	return ips
+}
+
 // resolveTSIG reads the material of a BindTSIGKey into TSIG credentials.
 func resolveTSIG(ctx context.Context, c client.Client, namespace, keyRef string) (bind.TSIGCreds, error) {
 	var creds bind.TSIGCreds

internal/controller/setup.go

@@ -35,5 +35,8 @@ func SetupAll(mgr ctrl.Manager, exec *bind.Executor) error {
 	if err := (&DNSRecordReconciler{Client: mgr.GetClient(), Scheme: mgr.GetScheme(), Exec: exec}).SetupWithManager(mgr); err != nil {
 		return err
 	}
+	if err := (&BindTSIGAPIReconciler{Client: mgr.GetClient(), Scheme: mgr.GetScheme()}).SetupWithManager(mgr); err != nil {
+		return err
+	}
 	return nil
 }

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{}

internal/controller/zone_helpers.go

@@ -2,6 +2,7 @@ package controller
 
 import (
 	"context"
+	"fmt"
 	"strings"
 
 	"sigs.k8s.io/controller-runtime/pkg/client"
@@ -58,12 +59,19 @@ func recordsToUpdates(zone string, records []bindv1alpha1.Record, defaultTTL int
 // updateKeyName returns the TSIG key name (as used in named.conf) for a zone's
 // update key, falling back to the object name.
 func updateKeyName(ctx context.Context, c client.Client, zone *bindv1alpha1.BindZone) string {
-	ref := zone.Spec.UpdateKeyRef
+	return tsigKeyName(ctx, c, zone.Namespace, zone.Spec.UpdateKeyRef)
+}
+
+// tsigKeyName resolves a BindTSIGKey object reference to the TSIG key name used
+// in named.conf (the KeyName override when set, otherwise the object name).
+// Returns "" for an empty ref, and falls back to the ref if the object cannot be
+// read.
+func tsigKeyName(ctx context.Context, c client.Client, namespace, ref string) string {
 	if ref == "" {
 		return ""
 	}
 	var key bindv1alpha1.BindTSIGKey
-	if err := c.Get(ctx, client.ObjectKey{Namespace: zone.Namespace, Name: ref}, &key); err != nil {
+	if err := c.Get(ctx, client.ObjectKey{Namespace: namespace, Name: ref}, &key); err != nil {
 		return ref
 	}
 	if key.Spec.KeyName != "" {
@@ -86,3 +94,23 @@ func terminateInline(entries []string) string {
 	}
 	return strings.Join(parts, " ")
 }
+
+// alsoNotifyList renders also-notify entries, each optionally annotated with a
+// TSIG key so the primary signs its NOTIFYs and secondaries can accept them by
+// key (`allow-notify { key ... }`) rather than by pod IP. An entry that already
+// carries a `key` clause is left untouched.
+func alsoNotifyList(addrs []string, key string) string {
+	key = strings.TrimSpace(key)
+	var parts []string
+	for _, a := range addrs {
+		a = strings.TrimSpace(strings.TrimRight(a, ";"))
+		if a == "" {
+			continue
+		}
+		if key != "" && !strings.Contains(a, " key ") {
+			a = fmt.Sprintf("%s key \"%s\"", a, key)
+		}
+		parts = append(parts, a+";")
+	}
+	return strings.Join(parts, " ")
+}

internal/tsigapi/keys.go

@@ -0,0 +1,168 @@
+package tsigapi
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	corev1 "k8s.io/api/core/v1"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/types"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+)
+
+// createKey creates a BindTSIGKey and waits for the operator to materialise it.
+func (s *Server) createKey(ctx context.Context, req createRequest) (*keyResponse, error) {
+	algorithm := req.Algorithm
+	if algorithm == "" {
+		algorithm = string(bindv1alpha1.TSIGHMACSHA256)
+	}
+	key := &bindv1alpha1.BindTSIGKey{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      req.Name,
+			Namespace: s.Namespace,
+			Labels:    map[string]string{"app.kubernetes.io/managed-by": "bind-tsig-api"},
+		},
+		Spec: bindv1alpha1.BindTSIGKeySpec{
+			Algorithm:  bindv1alpha1.TSIGAlgorithm(algorithm),
+			ClusterRef: req.ClusterRef,
+			KeyName:    req.Name,
+		},
+	}
+	if err := s.Client.Create(ctx, key); err != nil && !apierrors.IsAlreadyExists(err) {
+		return nil, fmt.Errorf("create bindtsigkey: %w", err)
+	}
+	return s.waitForMaterial(ctx, req.Name)
+}
+
+// readKey returns the current material for a key.
+func (s *Server) readKey(ctx context.Context, name string) (*keyResponse, error) {
+	var key bindv1alpha1.BindTSIGKey
+	if err := s.Client.Get(ctx, s.key(name), &key); err != nil {
+		return nil, err
+	}
+	return s.material(ctx, &key)
+}
+
+// rotateKey replaces the key material: it deletes the key Secret (the operator
+// regenerates it) and bumps a rotation annotation so the cluster controller
+// re-renders keys.conf and reloads BIND.
+func (s *Server) rotateKey(ctx context.Context, name string) (*keyResponse, error) {
+	var key bindv1alpha1.BindTSIGKey
+	if err := s.Client.Get(ctx, s.key(name), &key); err != nil {
+		return nil, err
+	}
+
+	secretName := secretNameFor(&key)
+	var secret corev1.Secret
+	if err := s.Client.Get(ctx, client.ObjectKey{Namespace: s.Namespace, Name: secretName}, &secret); err == nil {
+		if derr := s.Client.Delete(ctx, &secret); derr != nil && !apierrors.IsNotFound(derr) {
+			return nil, fmt.Errorf("delete key secret: %w", derr)
+		}
+	} else if !apierrors.IsNotFound(err) {
+		return nil, err
+	}
+
+	// Trigger re-generation + re-render by bumping the CR.
+	if key.Annotations == nil {
+		key.Annotations = map[string]string{}
+	}
+	key.Annotations[rotationAnnotation] = s.Now().UTC().Format(time.RFC3339Nano)
+	if err := s.Client.Update(ctx, &key); err != nil {
+		return nil, fmt.Errorf("bump rotation annotation: %w", err)
+	}
+
+	return s.waitForNewMaterial(ctx, name, secret.UID)
+}
+
+// deleteKey removes the BindTSIGKey (its Secret is garbage-collected).
+func (s *Server) deleteKey(ctx context.Context, name string) error {
+	key := &bindv1alpha1.BindTSIGKey{ObjectMeta: metav1.ObjectMeta{Name: name, Namespace: s.Namespace}}
+	if err := s.Client.Delete(ctx, key); err != nil && !apierrors.IsNotFound(err) {
+		return err
+	}
+	return nil
+}
+
+// material reads the key material from the Secret referenced by a BindTSIGKey.
+func (s *Server) material(ctx context.Context, key *bindv1alpha1.BindTSIGKey) (*keyResponse, error) {
+	var secret corev1.Secret
+	if err := s.Client.Get(ctx, client.ObjectKey{Namespace: s.Namespace, Name: secretNameFor(key)}, &secret); err != nil {
+		return nil, err
+	}
+	algorithm := string(secret.Data["algorithm"])
+	if algorithm == "" {
+		algorithm = string(key.Spec.Algorithm)
+	}
+	keyName := string(secret.Data["keyName"])
+	if keyName == "" {
+		keyName = key.Name
+	}
+	return &keyResponse{
+		Name:       key.Name,
+		Algorithm:  algorithm,
+		Secret:     string(secret.Data["secret"]),
+		KeyName:    keyName,
+		ClusterRef: key.Spec.ClusterRef,
+	}, nil
+}
+
+// waitForMaterial polls until the key's Secret exists (operator reconciled).
+func (s *Server) waitForMaterial(ctx context.Context, name string) (*keyResponse, error) {
+	deadline := s.Now().Add(s.WaitTimeout)
+	for {
+		var key bindv1alpha1.BindTSIGKey
+		if err := s.Client.Get(ctx, s.key(name), &key); err != nil {
+			return nil, err
+		}
+		if resp, err := s.material(ctx, &key); err == nil && resp.Secret != "" {
+			return resp, nil
+		}
+		if s.Now().After(deadline) {
+			return nil, fmt.Errorf("timed out waiting for key %q material", name)
+		}
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		case <-time.After(250 * time.Millisecond):
+		}
+	}
+}
+
+// waitForNewMaterial polls until the Secret has been recreated (different UID).
+func (s *Server) waitForNewMaterial(ctx context.Context, name string, oldUID types.UID) (*keyResponse, error) {
+	deadline := s.Now().Add(s.WaitTimeout)
+	for {
+		var key bindv1alpha1.BindTSIGKey
+		if err := s.Client.Get(ctx, s.key(name), &key); err != nil {
+			return nil, err
+		}
+		var secret corev1.Secret
+		err := s.Client.Get(ctx, client.ObjectKey{Namespace: s.Namespace, Name: secretNameFor(&key)}, &secret)
+		if err == nil && secret.UID != oldUID && len(secret.Data["secret"]) > 0 {
+			return s.material(ctx, &key)
+		}
+		if s.Now().After(deadline) {
+			return nil, fmt.Errorf("timed out waiting for key %q to rotate", name)
+		}
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		case <-time.After(250 * time.Millisecond):
+		}
+	}
+}
+
+// secretNameFor returns the Secret name holding a key's material.
+func secretNameFor(key *bindv1alpha1.BindTSIGKey) string {
+	if key.Status.SecretName != "" {
+		return key.Status.SecretName
+	}
+	if key.Spec.SecretName != "" {
+		return key.Spec.SecretName
+	}
+	return key.Name + "-tsig"
+}

internal/tsigapi/server.go

@@ -0,0 +1,155 @@
+// Package tsigapi implements the companion TSIG API deployed by the operator.
+// It exposes the HTTP contract consumed by vault-plugin-secrets-bind-tsig and
+// fulfils it by managing BindTSIGKey custom resources, which the operator
+// reconciles into key material.
+package tsigapi
+
+import (
+	"crypto/subtle"
+	"encoding/json"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/go-logr/logr"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	"k8s.io/apimachinery/pkg/types"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+)
+
+const rotationAnnotation = "bind.unkin.net/rotated-at"
+
+// Server serves the TSIG key API.
+type Server struct {
+	Client    client.Client
+	Namespace string
+	Token     string
+	Log       logr.Logger
+	// WaitTimeout bounds how long a create/rotate waits for the operator to
+	// reconcile the key material.
+	WaitTimeout time.Duration
+	// Now is injectable for tests; defaults to time.Now.
+	Now func() time.Time
+}
+
+type keyResponse struct {
+	Name       string `json:"name"`
+	Algorithm  string `json:"algorithm"`
+	Secret     string `json:"secret"`
+	KeyName    string `json:"key_name"`
+	ClusterRef string `json:"cluster_ref,omitempty"`
+}
+
+type createRequest struct {
+	Name       string `json:"name"`
+	Algorithm  string `json:"algorithm"`
+	ClusterRef string `json:"cluster_ref"`
+	Static     bool   `json:"static"`
+}
+
+// Handler returns the API's HTTP handler.
+func (s *Server) Handler() http.Handler {
+	if s.Now == nil {
+		s.Now = time.Now
+	}
+	if s.WaitTimeout == 0 {
+		s.WaitTimeout = 15 * time.Second
+	}
+	mux := http.NewServeMux()
+	mux.HandleFunc("/healthz", func(w http.ResponseWriter, _ *http.Request) { _, _ = w.Write([]byte("ok")) })
+	mux.HandleFunc("/v1/keys", s.auth(s.handleKeys))
+	mux.HandleFunc("/v1/keys/", s.auth(s.handleKey))
+	return mux
+}
+
+func (s *Server) auth(next http.HandlerFunc) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		if s.Token != "" {
+			want := "Bearer " + s.Token
+			got := r.Header.Get("Authorization")
+			if subtle.ConstantTimeCompare([]byte(got), []byte(want)) != 1 {
+				http.Error(w, "unauthorized", http.StatusUnauthorized)
+				return
+			}
+		}
+		next(w, r)
+	}
+}
+
+// handleKeys serves POST /v1/keys (create).
+func (s *Server) handleKeys(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPost {
+		http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+	var req createRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "invalid body", http.StatusBadRequest)
+		return
+	}
+	if req.Name == "" {
+		http.Error(w, "name is required", http.StatusBadRequest)
+		return
+	}
+	key, err := s.createKey(r.Context(), req)
+	if err != nil {
+		s.fail(w, "create", req.Name, err)
+		return
+	}
+	writeJSON(w, http.StatusOK, key)
+}
+
+// handleKey serves GET/DELETE /v1/keys/{name} and POST /v1/keys/{name}/rotate.
+func (s *Server) handleKey(w http.ResponseWriter, r *http.Request) {
+	name := strings.Trim(strings.TrimPrefix(r.URL.Path, "/v1/keys/"), "/")
+	rotate := strings.HasSuffix(name, "/rotate")
+	name = strings.TrimSuffix(name, "/rotate")
+	if name == "" {
+		http.Error(w, "key name is required", http.StatusBadRequest)
+		return
+	}
+
+	switch {
+	case rotate && r.Method == http.MethodPost:
+		key, err := s.rotateKey(r.Context(), name)
+		if err != nil {
+			s.fail(w, "rotate", name, err)
+			return
+		}
+		writeJSON(w, http.StatusOK, key)
+	case r.Method == http.MethodGet:
+		key, err := s.readKey(r.Context(), name)
+		if err != nil {
+			s.fail(w, "read", name, err)
+			return
+		}
+		writeJSON(w, http.StatusOK, key)
+	case r.Method == http.MethodDelete:
+		if err := s.deleteKey(r.Context(), name); err != nil {
+			s.fail(w, "delete", name, err)
+			return
+		}
+		w.WriteHeader(http.StatusNoContent)
+	default:
+		http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
+	}
+}
+
+func (s *Server) fail(w http.ResponseWriter, op, name string, err error) {
+	if apierrors.IsNotFound(err) {
+		http.Error(w, "not found", http.StatusNotFound)
+		return
+	}
+	s.Log.Error(err, "tsig api request failed", "op", op, "name", name)
+	http.Error(w, err.Error(), http.StatusInternalServerError)
+}
+
+func writeJSON(w http.ResponseWriter, status int, v interface{}) {
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(status)
+	_ = json.NewEncoder(w).Encode(v)
+}
+
+func (s *Server) key(name string) types.NamespacedName {
+	return types.NamespacedName{Namespace: s.Namespace, Name: name}
+}

internal/tsigapi/server_test.go

@@ -0,0 +1,185 @@
+package tsigapi
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/go-logr/logr"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/apimachinery/pkg/types"
+	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/client/fake"
+
+	bindv1alpha1 "git.unkin.net/unkin/bind-operator/api/v1alpha1"
+)
+
+const testNS = "bind-system"
+
+// fakeOperator simulates the BindTSIGKey controller: whenever a key exists
+// without its material Secret, it creates one. It stops when ctx is cancelled.
+func fakeOperator(ctx context.Context, c client.Client) {
+	seq := 0
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-time.After(20 * time.Millisecond):
+		}
+		var keys bindv1alpha1.BindTSIGKeyList
+		if err := c.List(ctx, &keys); err != nil {
+			continue
+		}
+		for i := range keys.Items {
+			k := &keys.Items[i]
+			secretName := k.Name + "-tsig"
+			var existing corev1.Secret
+			if err := c.Get(ctx, client.ObjectKey{Namespace: k.Namespace, Name: secretName}, &existing); err == nil {
+				continue
+			}
+			seq++
+			material := "secret-material-" + itoa(seq)
+			_ = c.Create(ctx, &corev1.Secret{
+				// Fake client does not assign UIDs; set one so rotation
+				// (which detects a Secret with a new UID) is observable.
+				ObjectMeta: metav1.ObjectMeta{Name: secretName, Namespace: k.Namespace, UID: types.UID("uid-" + itoa(seq))},
+				Data: map[string][]byte{
+					"algorithm": []byte(k.Spec.Algorithm),
+					"keyName":   []byte(k.Name),
+					"secret":    []byte(material),
+				},
+			})
+		}
+	}
+}
+
+func itoa(n int) string {
+	if n == 0 {
+		return "0"
+	}
+	var b []byte
+	for n > 0 {
+		b = append([]byte{byte('0' + n%10)}, b...)
+		n /= 10
+	}
+	return string(b)
+}
+
+func newTestServer(t *testing.T) (*Server, context.CancelFunc) {
+	t.Helper()
+	scheme := runtime.NewScheme()
+	if err := clientgoscheme.AddToScheme(scheme); err != nil {
+		t.Fatalf("clientgo scheme: %v", err)
+	}
+	if err := bindv1alpha1.AddToScheme(scheme); err != nil {
+		t.Fatalf("bind scheme: %v", err)
+	}
+	c := fake.NewClientBuilder().WithScheme(scheme).Build()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	go fakeOperator(ctx, c)
+
+	return &Server{
+		Client:      c,
+		Namespace:   testNS,
+		Token:       "s3cr3t",
+		Log:         logr.Discard(),
+		WaitTimeout: 3 * time.Second,
+	}, cancel
+}
+
+func TestHealthzAndAuth(t *testing.T) {
+	srv, cancel := newTestServer(t)
+	defer cancel()
+	h := srv.Handler()
+
+	// healthz needs no auth.
+	rr := httptest.NewRecorder()
+	h.ServeHTTP(rr, httptest.NewRequest(http.MethodGet, "/healthz", nil))
+	if rr.Code != http.StatusOK {
+		t.Fatalf("healthz: want 200, got %d", rr.Code)
+	}
+
+	// missing token is rejected.
+	rr = httptest.NewRecorder()
+	h.ServeHTTP(rr, httptest.NewRequest(http.MethodPost, "/v1/keys", strings.NewReader(`{"name":"x"}`)))
+	if rr.Code != http.StatusUnauthorized {
+		t.Fatalf("no-auth: want 401, got %d", rr.Code)
+	}
+}
+
+func TestKeyLifecycle(t *testing.T) {
+	srv, cancel := newTestServer(t)
+	defer cancel()
+	h := srv.Handler()
+
+	// create
+	created := do(t, h, http.MethodPost, "/v1/keys", `{"name":"host-a","cluster_ref":"bind-authoritative"}`, http.StatusOK)
+	if created.Secret == "" {
+		t.Fatalf("create returned empty secret")
+	}
+	if created.ClusterRef != "bind-authoritative" {
+		t.Fatalf("cluster_ref not propagated: %q", created.ClusterRef)
+	}
+
+	// read returns the same material
+	got := do(t, h, http.MethodGet, "/v1/keys/host-a", "", http.StatusOK)
+	if got.Secret != created.Secret {
+		t.Fatalf("read secret %q != created %q", got.Secret, created.Secret)
+	}
+
+	// rotate returns new material
+	rotated := do(t, h, http.MethodPost, "/v1/keys/host-a/rotate", "", http.StatusOK)
+	if rotated.Secret == created.Secret {
+		t.Fatalf("rotate did not change secret")
+	}
+
+	// delete
+	rr := httptest.NewRecorder()
+	req := authed(http.MethodDelete, "/v1/keys/host-a", "")
+	h.ServeHTTP(rr, req)
+	if rr.Code != http.StatusNoContent {
+		t.Fatalf("delete: want 204, got %d", rr.Code)
+	}
+
+	// read after delete → 404
+	rr = httptest.NewRecorder()
+	h.ServeHTTP(rr, authed(http.MethodGet, "/v1/keys/host-a", ""))
+	if rr.Code != http.StatusNotFound {
+		t.Fatalf("read-after-delete: want 404, got %d", rr.Code)
+	}
+}
+
+func authed(method, path, body string) *http.Request {
+	var r *http.Request
+	if body == "" {
+		r = httptest.NewRequest(method, path, nil)
+	} else {
+		r = httptest.NewRequest(method, path, strings.NewReader(body))
+	}
+	r.Header.Set("Authorization", "Bearer s3cr3t")
+	return r
+}
+
+func do(t *testing.T, h http.Handler, method, path, body string, wantCode int) keyResponse {
+	t.Helper()
+	rr := httptest.NewRecorder()
+	h.ServeHTTP(rr, authed(method, path, body))
+	if rr.Code != wantCode {
+		t.Fatalf("%s %s: want %d, got %d (%s)", method, path, wantCode, rr.Code, rr.Body.String())
+	}
+	var resp keyResponse
+	if rr.Body.Len() > 0 {
+		if err := json.Unmarshal(rr.Body.Bytes(), &resp); err != nil {
+			t.Fatalf("decode response: %v", err)
+		}
+	}
+	return resp
+}