Merge pull request 'ci: add buildkit_config CA trust for artifactapi push' (#18) from benvin/buildx-ca-config into main

Reviewed-on: #18

.woodpecker/build.yaml

@@ -3,15 +3,15 @@ when:
 
 steps:
   - name: docker-build-operator
-    image: woodpeckerci/plugin-docker-buildx
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/plugin-docker-buildx:latest
     settings:
-      repo: git.unkin.net/unkin/bind-operator
+      repo: artifactapi.k8s.syd1.au.unkin.net/docker-internal/bind-operator
       dockerfile: Dockerfile.operator
       dry_run: true
 
   - name: docker-build-tsig-api
-    image: woodpeckerci/plugin-docker-buildx
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/plugin-docker-buildx:latest
     settings:
-      repo: git.unkin.net/unkin/bind-tsig-api
+      repo: artifactapi.k8s.syd1.au.unkin.net/docker-internal/bind-tsig-api
       dockerfile: Dockerfile.tsigapi
       dry_run: true

.woodpecker/docker.yaml

@@ -4,27 +4,27 @@ when:
 
 steps:
   - name: docker-operator
-    image: woodpeckerci/plugin-docker-buildx
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/plugin-docker-buildx:latest
     settings:
-      registry: git.unkin.net
-      repo: git.unkin.net/unkin/bind-operator
+      registry: artifactapi.k8s.syd1.au.unkin.net
+      repo: artifactapi.k8s.syd1.au.unkin.net/docker-internal/bind-operator
       dockerfile: Dockerfile.operator
-      username: droneci
-      password:
-        from_secret: DRONECI_PASSWORD
+      buildkit_config: |
+        [registry."artifactapi.k8s.syd1.au.unkin.net"]
+          ca = ["/etc/docker/certs.d/artifactapi.k8s.syd1.au.unkin.net/ca.crt"]
       tags:
         - ${CI_COMMIT_TAG}
         - latest
 
   - name: docker-tsig-api
-    image: woodpeckerci/plugin-docker-buildx
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/plugin-docker-buildx:latest
     settings:
-      registry: git.unkin.net
-      repo: git.unkin.net/unkin/bind-tsig-api
+      registry: artifactapi.k8s.syd1.au.unkin.net
+      repo: artifactapi.k8s.syd1.au.unkin.net/docker-internal/bind-tsig-api
       dockerfile: Dockerfile.tsigapi
-      username: droneci
-      password:
-        from_secret: DRONECI_PASSWORD
+      buildkit_config: |
+        [registry."artifactapi.k8s.syd1.au.unkin.net"]
+          ca = ["/etc/docker/certs.d/artifactapi.k8s.syd1.au.unkin.net/ca.crt"]
       tags:
         - ${CI_COMMIT_TAG}
         - latest

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. |
@@ -85,7 +85,8 @@ kubectl apply -f config/samples/
 
 Woodpecker runs `pre-commit` (gofmt + vet), `test`, and a dry-run image `build`
 on pull requests; pushing a `v*` tag builds and pushes
-`git.unkin.net/unkin/bind-operator` to the Gitea registry.
+`artifactapi.k8s.syd1.au.unkin.net/docker-internal/bind-operator` (and
+`bind-tsig-api`) to the artifactapi local docker registry.
 
 ## Notes & caveats
 

api/v1alpha1/bindtsigkey_types.go

@@ -41,6 +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

@@ -691,7 +691,7 @@ 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)
 }
 
@@ -748,6 +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.
@@ -1208,3 +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
+}

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

@@ -87,6 +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

@@ -2376,6 +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/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

@@ -22,15 +22,48 @@ type RenderInput struct {
 	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
@@ -67,6 +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")
 	}
@@ -230,15 +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
-	if len(primaries) == 0 && in.PrimaryAddress != "" {
-		primaries = []string{in.PrimaryAddress}
-	}
+	primaries := transferPrimaries(in)
 	if len(primaries) == 0 {
 		return ""
 	}
@@ -259,10 +353,7 @@ func renderCatalogZoneDecl(in RenderInput, isPrimary bool, indent string) string
 	}
 	cat := in.Catalog
 	file := CatalogFilePath(cat.Spec.ZoneName)
-	primaries := cat.Spec.DefaultPrimaries
-	if len(primaries) == 0 && in.PrimaryAddress != "" {
-		primaries = []string{in.PrimaryAddress}
-	}
+	primaries := transferPrimaries(in)
 	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 +80,108 @@ 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{
@@ -128,3 +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
-    900        ; retry
+    300        ; refresh
+    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"
@@ -168,10 +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
-	// name, so render with pod-0's current IP (empty until it is scheduled; the
-	// Pod watch re-renders when it appears or changes).
-	in := bind.RenderInput{Cluster: c, PrimaryAddress: primaryPodIP(ctx, r.Client, c)}
+	// BIND primaries/default-primaries need an IP address, not a DNS name. Use
+	// the stable primary Service ClusterIP so secondaries keep transferring
+	// across primary pod restarts (falls back to the pod IP when no primary
+	// 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 {
@@ -227,6 +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,
@@ -350,6 +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{
@@ -357,7 +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,
@@ -425,6 +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

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

@@ -80,7 +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())
 	}
@@ -133,7 +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
@@ -145,8 +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")
@@ -202,6 +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"
@@ -108,6 +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/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, " ")
+}