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

internal/bind/render.go

@@ -22,15 +22,20 @@ type RenderInput struct {
 	Forwards []bindv1alpha1.BindZone
 	// PrimaryAddress is the in-cluster address secondaries transfer from.
 	PrimaryAddress string
-	// PrimaryPodAddresses are the primary pod's own IP(s). Secondaries transfer
-	// from PrimaryAddress (the stable primary Service ClusterIP) but 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 ClusterIP), REFUSES them as "non-primary" and
-	// replication falls back to the SOA refresh timer. Secondaries render these
-	// into an options-scope allow-notify so intra-cluster NOTIFYs are accepted
-	// immediately. Empty leaves BIND's default behaviour unchanged.
-	PrimaryPodAddresses []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
@@ -95,11 +100,13 @@ func render(in RenderInput, isPrimary bool) string {
 		b.WriteString("    allow-new-zones yes;\n")
 	}
 	b.WriteString("    dnssec-validation auto;\n")
-	// Secondaries accept NOTIFY from the primary's pod IP(s). 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 an options-scope allow-notify covering the pod IP(s) is
-	// needed or every NOTIFY is refused and replication waits for the SOA refresh.
+	// 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")
@@ -293,36 +300,31 @@ func transferPrimaries(in RenderInput) []string {
 }
 
 // allowNotifyClause renders an options-scope allow-notify on secondaries that
-// permits the primary pod IP(s). 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".
-// Emitted only on secondaries and only when the primary pod IP(s) are known.
+// 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 ""
 	}
-	addrs := make([]string, 0, len(in.PrimaryPodAddresses)+1)
-	seen := map[string]bool{}
-	for _, a := range in.PrimaryPodAddresses {
-		a = strings.TrimSpace(strings.TrimRight(a, ";"))
-		if a == "" || seen[a] {
-			continue
-		}
-		seen[a] = true
-		addrs = append(addrs, a)
-	}
-	// Keep the transfer address (the Service ClusterIP, or the pod IP when no
-	// primary Service exists) in the set: an explicit allow-notify replaces the
-	// implicit primaries-derived default, so it must still cover that source.
-	if a := strings.TrimSpace(strings.TrimRight(in.PrimaryAddress, ";")); a != "" && !seen[a] {
-		addrs = append(addrs, a)
-	}
-	if len(addrs) == 0 {
+	key := strings.TrimSpace(in.NotifyKeyName)
+	if key == "" {
 		return ""
 	}
-	sort.Strings(addrs)
-	return fmt.Sprintf("%sallow-notify { %s };\n", indent, terminate(addrs))
+	return fmt.Sprintf("%sallow-notify { key \"%s\"; };\n", indent, key)
 }
 
 func catalogZonesClause(in RenderInput, isPrimary bool, indent string) string {

internal/bind/render_test.go

@@ -98,31 +98,87 @@ func TestRenderCatalogPrimariesCarryTransferKey(t *testing.T) {
 	}
 }
 
-func TestRenderSecondaryAllowNotifyPrimaryPodIP(t *testing.T) {
+func TestRenderSecondaryAllowNotifyByKey(t *testing.T) {
 	// Secondaries transfer from the primary Service ClusterIP but the primary's
-	// NOTIFYs arrive from its pod IP, so an options allow-notify must cover the
-	// pod IP (and keep the transfer address) or BIND refuses them as non-primary.
+	// 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",
-		PrimaryPodAddresses: []string{"10.42.3.197"},
+		NotifyKeyName:  "externaldns-key",
 	}
 	primary, secondary := RenderNamedConf(in)
-	if !strings.Contains(secondary, "allow-notify { 10.42.3.197; 10.43.5.5; };") {
-		t.Fatalf("secondary allow-notify must cover the primary pod IP and transfer address:\n%s", secondary)
+	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 TestRenderSecondaryAllowNotifyOmittedWhenPodIPUnknown(t *testing.T) {
-	// With no primary pod IP known there is nothing to add beyond BIND's implicit
-	// primaries-derived default; emit nothing rather than a bare/duplicate clause.
-	in := RenderInput{Cluster: newCluster(bindv1alpha1.ModeAuthoritative)}
+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 primary addresses are known:\n%s", secondary)
+		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)
 	}
 }
 

internal/controller/bindcluster_controller.go

@@ -176,14 +176,6 @@ func (r *BindClusterReconciler) reconcileConfigMap(ctx context.Context, c *bindv
 	// 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)}
-	// Secondaries transfer from the stable primary Service ClusterIP, but the
-	// primary pod's NOTIFYs are sourced from its pod IP, which BIND refuses as
-	// "non-primary" unless it appears in allow-notify. Render the primary pod IP
-	// so intra-cluster NOTIFYs are accepted immediately (the Pod watch re-renders
-	// the ConfigMap when the pod IP changes across restarts).
-	if ip := primaryPodIP(ctx, r.Client, c); ip != "" {
-		in.PrimaryPodAddresses = []string{ip}
-	}
 
 	var acls bindv1alpha1.BindACLList
 	if err := r.List(ctx, &acls, client.InNamespace(c.Namespace)); err == nil {
@@ -239,6 +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,

internal/controller/bindzone_controller.go

@@ -88,7 +88,13 @@ func (r *BindZoneReconciler) Reconcile(ctx context.Context, req ctrl.Request) (c
 		notifyTargets = secondaryPodIPs(ctx, r.Client, cluster)
 	}
 
-	zoneConfig, err := r.buildZoneConfig(ctx, &zone, r.zoneTransferKeyRef(ctx, &zone, cluster), notifyTargets)
+	// 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())
 	}
@@ -142,8 +148,10 @@ func (r *BindZoneReconciler) Reconcile(ctx context.Context, req ctrl.Request) (c
 
 // buildZoneConfig renders the inner clause passed to rndc addzone/modzone.
 // transferKey, when set, is the catalog transfer TSIG key name; catalog member
-// primary zones must allow AXFR with it so secondaries can pull them.
-func (r *BindZoneReconciler) buildZoneConfig(ctx context.Context, zone *bindv1alpha1.BindZone, transferKey string, notifyTargets []string) (string, error) {
+// 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
@@ -164,9 +172,13 @@ func (r *BindZoneReconciler) buildZoneConfig(ctx context.Context, zone *bindv1al
 		}
 		// 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.
+		// 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 }", terminateInline(notifyTargets)))
+			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")

internal/controller/bindzone_notify_test.go

@@ -19,15 +19,32 @@ func TestBuildZoneConfigPrimaryAlsoNotify(t *testing.T) {
 		},
 	}
 
-	cfg, err := r.buildZoneConfig(context.Background(), zone, "transfer-key", []string{"10.42.2.6", "10.42.1.5"})
+	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)
 	}
-	if !strings.Contains(cfg, "also-notify { 10.42.2.6; 10.42.1.5; }") {
-		t.Errorf("expected also-notify with the secondary IPs 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)
 	}
 }
 
@@ -38,7 +55,7 @@ func TestBuildZoneConfigPrimaryNoNotifyTargets(t *testing.T) {
 		Spec: bindv1alpha1.BindZoneSpec{ZoneName: "main.unkin.net", Type: bindv1alpha1.ZonePrimary},
 	}
 
-	cfg, err := r.buildZoneConfig(context.Background(), zone, "transfer-key", nil)
+	cfg, err := r.buildZoneConfig(context.Background(), zone, "transfer-key", nil, "externaldns-key")
 	if err != nil {
 		t.Fatalf("buildZoneConfig: %v", err)
 	}

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/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, " ")
+}