Reinject captured first packet on lazy peer activation and keep peer alive across endpoint swap

private services on a custom domain didn't resolve on clients — the synthesized DNS zone was anchored to the cluster, and the account's custom domains weren't even
  loaded.

- account.go — SynthesizePrivateServiceZones now keys zones by a resolved apex (privateServiceDomainZone): cluster suffix → registered account.Domains (filtered by matching
  TargetCluster, longest wins) → skip if none. One zone per apex; custom-domain services group under their registered domain.
- sql_store.go — GetAccount now loads account.Domains on both loaders (gorm Preload("Domains") + pgx goroutine via ListCustomDomains; errChan buffer bumped 12→16). This was
  the reason the deploy didn't work — the relation was empty in prod.
- Tests — custom-domain zone synthesis cases (apex resolution, free+custom separation, sibling collapse, cluster mismatch, mixed cluster/custom/public) + GetAccount
  domain-preload tests on sqlite and Postgres.

.github/workflows/release.yml

@@ -29,10 +29,10 @@ jobs:
           persist-credentials: false
 
       - name: Generate FreeBSD port diff
-        run: bash release_files/freebsd-port-diff.sh
+        run: bash -x release_files/freebsd-port-diff.sh
 
       - name: Generate FreeBSD port issue body
-        run: bash release_files/freebsd-port-issue-body.sh
+        run: bash -x release_files/freebsd-port-issue-body.sh
 
       - name: Check if diff was generated
         id: check_diff

client/iface/wgproxy/bind/proxy.go

@@ -136,6 +136,15 @@ func (p *ProxyBind) CloseConn() error {
 	return p.close()
 }
 
+// InjectPacket writes b to the remote peer over the underlying transport.
+func (p *ProxyBind) InjectPacket(b []byte) error {
+	if p.remoteConn == nil {
+		return fmt.Errorf("proxy not started")
+	}
+	_, err := p.remoteConn.Write(b)
+	return err
+}
+
 func (p *ProxyBind) close() error {
 	if p.remoteConn == nil {
 		return nil

client/iface/wgproxy/ebpf/wrapper.go

@@ -219,6 +219,15 @@ func (p *ProxyWrapper) RedirectAs(endpoint *net.UDPAddr) {
 	p.pausedCond.L.Unlock()
 }
 
+// InjectPacket writes b to the remote peer over the underlying transport.
+func (p *ProxyWrapper) InjectPacket(b []byte) error {
+	if p.remoteConn == nil {
+		return fmt.Errorf("proxy not started")
+	}
+	_, err := p.remoteConn.Write(b)
+	return err
+}
+
 // CloseConn close the remoteConn and automatically remove the conn instance from the map
 func (p *ProxyWrapper) CloseConn() error {
 	if p.cancel == nil {

client/iface/wgproxy/proxy.go

@@ -18,4 +18,8 @@ type Proxy interface {
 	RedirectAs(endpoint *net.UDPAddr)
 	CloseConn() error
 	SetDisconnectListener(disconnected func())
+
+	// InjectPacket writes a raw packet directly to the remote peer over the underlying transport,
+	// bypassing WireGuard. Used to replay the captured lazyconn handshake initiation.
+	InjectPacket(b []byte) error
 }

client/iface/wgproxy/udp/proxy.go

@@ -147,6 +147,15 @@ func (p *WGUDPProxy) RedirectAs(endpoint *net.UDPAddr) {
 	p.sendPkg = p.srcFakerConn.SendPkg
 }
 
+// InjectPacket writes b to the remote peer over the underlying transport.
+func (p *WGUDPProxy) InjectPacket(b []byte) error {
+	if p.remoteConn == nil {
+		return fmt.Errorf("proxy not started")
+	}
+	_, err := p.remoteConn.Write(b)
+	return err
+}
+
 // CloseConn close the localConn
 func (p *WGUDPProxy) CloseConn() error {
 	if p.cancel == nil {

client/internal/engine_test.go

@@ -224,6 +224,10 @@ func (m *MockWGIface) LastActivities() map[string]monotime.Time {
 	return nil
 }
 
+func (m *MockWGIface) MTU() uint16 {
+	return 1280
+}
+
 func (m *MockWGIface) SetPresharedKey(peerKey string, psk wgtypes.Key, updateOnly bool) error {
 	return nil
 }

client/internal/iface_common.go

@@ -44,4 +44,5 @@ type wgIfaceBase interface {
 	FullStats() (*configurer.Stats, error)
 	LastActivities() map[string]monotime.Time
 	SetPresharedKey(peerKey string, psk wgtypes.Key, updateOnly bool) error
+	MTU() uint16
 }

client/internal/lazyconn/activity/lazy_conn.go

@@ -4,15 +4,21 @@ import (
 	"context"
 	"io"
 	"net"
+	"sync"
 	"time"
 )
 
 // lazyConn detects activity when WireGuard attempts to send packets.
-// It does not deliver packets, only signals that activity occurred.
+// It does not deliver packets, only signals that activity occurred. The first packet WG-go writes
+// is captured so the lazyconn manager can reinject it through the real bind endpoint, avoiding the
+// handshake-retry wait when the real connection comes up.
 type lazyConn struct {
 	activityCh chan struct{}
 	ctx        context.Context
 	cancel     context.CancelFunc
+
+	mu             sync.Mutex
+	capturedPacket []byte
 }
 
 // newLazyConn creates a new lazyConn for activity detection.
@@ -31,12 +37,19 @@ func (c *lazyConn) Read(_ []byte) (n int, err error) {
 	return 0, io.EOF
 }
 
-// Write signals activity detection when ICEBind routes packets to this endpoint.
+// Write signals activity detection when ICEBind routes packets to this endpoint. The first packet
+// is cloned and stored so it can be replayed on the real endpoint once ICE/relay comes up.
 func (c *lazyConn) Write(b []byte) (n int, err error) {
 	if c.ctx.Err() != nil {
 		return 0, io.EOF
 	}
 
+	c.mu.Lock()
+	if c.capturedPacket == nil && len(b) > 0 {
+		c.capturedPacket = append([]byte(nil), b...)
+	}
+	c.mu.Unlock()
+
 	select {
 	case c.activityCh <- struct{}{}:
 	default:
@@ -45,6 +58,13 @@ func (c *lazyConn) Write(b []byte) (n int, err error) {
 	return len(b), nil
 }
 
+// CapturedPacket returns a copy of the first packet written to this connection, or nil.
+func (c *lazyConn) CapturedPacket() []byte {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.capturedPacket
+}
+
 // ActivityChan returns the channel that signals when activity is detected.
 func (c *lazyConn) ActivityChan() <-chan struct{} {
 	return c.activityCh

client/internal/lazyconn/activity/listener_bind.go

@@ -118,16 +118,23 @@ func (d *BindListener) ReadPackets() {
 		d.peerCfg.Log.Infof("exit from activity listener")
 	}
 
-	d.peerCfg.Log.Debugf("removing lazy endpoint for peer %s", d.peerCfg.PublicKey)
-	if err := d.wgIface.RemovePeer(d.peerCfg.PublicKey); err != nil {
-		d.peerCfg.Log.Errorf("failed to remove endpoint: %s", err)
-	}
-
+	// Leave the peer in place. ConfigureWGEndpoint will UpdatePeer with the real endpoint;
+	// removing the peer here wipes wireguard-go's staged queue and drops the user packet that
+	// triggered activation.
 	_ = d.lazyConn.Close()
 	d.bind.RemoveEndpoint(d.fakeIP)
 	d.done.Done()
 }
 
+// CapturedPacket returns the first packet that triggered activity, or nil if none was captured.
+// Safe to call after ReadPackets returns.
+func (d *BindListener) CapturedPacket() []byte {
+	if d.lazyConn == nil {
+		return nil
+	}
+	return d.lazyConn.CapturedPacket()
+}
+
 // Close stops the listener and cleans up resources.
 func (d *BindListener) Close() {
 	d.peerCfg.Log.Infof("closing activity listener (LazyConn)")

client/internal/lazyconn/activity/listener_bind_test.go

@@ -68,6 +68,10 @@ func (m *MockWGIfaceBind) GetBind() device.EndpointManager {
 	return m.endpointMgr
 }
 
+func (m *MockWGIfaceBind) MTU() uint16 {
+	return 1280
+}
+
 func TestBindListener_Creation(t *testing.T) {
 	mockEndpointMgr := newMockEndpointManager()
 	mockIface := &MockWGIfaceBind{endpointMgr: mockEndpointMgr}
@@ -207,8 +211,9 @@ func TestManager_BindMode(t *testing.T) {
 	require.NoError(t, err)
 
 	select {
-	case peerConnID := <-mgr.OnActivityChan:
-		assert.Equal(t, cfg.PeerConnID, peerConnID, "Received peer connection ID should match")
+	case ev := <-mgr.OnActivityChan:
+		assert.Equal(t, cfg.PeerConnID, ev.PeerConnID, "Received peer connection ID should match")
+		assert.Equal(t, []byte{0x01, 0x02, 0x03}, ev.FirstPacket, "First packet should be captured for reinjection")
 	case <-time.After(2 * time.Second):
 		t.Fatal("timeout waiting for activity notification")
 	}
@@ -266,8 +271,8 @@ func TestManager_BindMode_MultiplePeers(t *testing.T) {
 	receivedPeers := make(map[peerid.ConnID]bool)
 	for i := 0; i < 2; i++ {
 		select {
-		case peerConnID := <-mgr.OnActivityChan:
-			receivedPeers[peerConnID] = true
+		case ev := <-mgr.OnActivityChan:
+			receivedPeers[ev.PeerConnID] = true
 		case <-time.After(2 * time.Second):
 			t.Fatal("timeout waiting for activity notifications")
 		}

client/internal/lazyconn/activity/listener_udp.go

@@ -8,6 +8,7 @@ import (
 
 	log "github.com/sirupsen/logrus"
 
+	"github.com/netbirdio/netbird/client/iface/bufsize"
 	"github.com/netbirdio/netbird/client/internal/lazyconn"
 )
 
@@ -20,6 +21,8 @@ type UDPListener struct {
 	done     sync.Mutex
 
 	isClosed atomic.Bool
+
+	capturedPacket []byte
 }
 
 // NewUDPListener creates a listener that detects activity via UDP socket reads.
@@ -46,9 +49,13 @@ func NewUDPListener(wgIface WgInterface, cfg lazyconn.PeerConfig) (*UDPListener,
 }
 
 // ReadPackets blocks reading from the UDP socket until activity is detected or the listener is closed.
+// The first packet that triggers activity is captured so it can be reinjected through the real
+// transport once it is established. Without this, kernel WireGuard's handshake initiation would be
+// dropped and WG would only retry after REKEY_TIMEOUT.
 func (d *UDPListener) ReadPackets() {
+	buf := make([]byte, int(d.wgIface.MTU())+bufsize.WGBufferOverhead)
 	for {
-		n, remoteAddr, err := d.conn.ReadFromUDP(make([]byte, 1))
+		n, remoteAddr, err := d.conn.ReadFromUDP(buf)
 		if err != nil {
 			if d.isClosed.Load() {
 				d.peerCfg.Log.Infof("exit from activity listener")
@@ -62,20 +69,24 @@ func (d *UDPListener) ReadPackets() {
 			d.peerCfg.Log.Warnf("received %d bytes from %s, too short", n, remoteAddr)
 			continue
 		}
-		d.peerCfg.Log.Infof("activity detected")
+		d.capturedPacket = append([]byte(nil), buf[:n]...)
+		d.peerCfg.Log.Infof("activity detected, captured %d bytes for reinjection", n)
 		break
 	}
 
-	d.peerCfg.Log.Debugf("removing lazy endpoint: %s", d.endpoint.String())
-	if err := d.wgIface.RemovePeer(d.peerCfg.PublicKey); err != nil {
-		d.peerCfg.Log.Errorf("failed to remove endpoint: %s", err)
-	}
-
-	// Ignore close error as it may return "use of closed network connection" if already closed.
+	// Leave the peer in place. ConfigureWGEndpoint will UpdatePeer with the real endpoint;
+	// removing the peer here wipes kernel WG's staged queue and drops the user packet that
+	// triggered activation.
 	_ = d.conn.Close()
 	d.done.Unlock()
 }
 
+// CapturedPacket returns the first packet that triggered activity, or nil if none was captured.
+// Safe to call after ReadPackets returns.
+func (d *UDPListener) CapturedPacket() []byte {
+	return d.capturedPacket
+}
+
 // Close stops the listener and cleans up resources.
 func (d *UDPListener) Close() {
 	d.peerCfg.Log.Infof("closing activity listener: %s", d.conn.LocalAddr().String())

client/internal/lazyconn/activity/manager.go

@@ -19,6 +19,14 @@ import (
 type listener interface {
 	ReadPackets()
 	Close()
+	CapturedPacket() []byte
+}
+
+// Event reports activity on a managed peer. FirstPacket is the bytes that triggered activation,
+// captured for reinjection through the real transport.
+type Event struct {
+	PeerConnID  peerid.ConnID
+	FirstPacket []byte
 }
 
 type WgInterface interface {
@@ -26,10 +34,11 @@ type WgInterface interface {
 	UpdatePeer(peerKey string, allowedIps []netip.Prefix, keepAlive time.Duration, endpoint *net.UDPAddr, preSharedKey *wgtypes.Key) error
 	IsUserspaceBind() bool
 	Address() wgaddr.Address
+	MTU() uint16
 }
 
 type Manager struct {
-	OnActivityChan chan peerid.ConnID
+	OnActivityChan chan Event
 
 	wgIface WgInterface
 
@@ -41,7 +50,7 @@ type Manager struct {
 
 func NewManager(wgIface WgInterface) *Manager {
 	m := &Manager{
-		OnActivityChan: make(chan peerid.ConnID, 1),
+		OnActivityChan: make(chan Event, 1),
 		wgIface:        wgIface,
 		peers:          make(map[peerid.ConnID]listener),
 		done:           make(chan struct{}),
@@ -116,12 +125,12 @@ func (m *Manager) waitForTraffic(l listener, peerConnID peerid.ConnID) {
 	delete(m.peers, peerConnID)
 	m.mu.Unlock()
 
-	m.notify(peerConnID)
+	m.notify(Event{PeerConnID: peerConnID, FirstPacket: l.CapturedPacket()})
 }
 
-func (m *Manager) notify(peerConnID peerid.ConnID) {
+func (m *Manager) notify(ev Event) {
 	select {
 	case <-m.done:
-	case m.OnActivityChan <- peerConnID:
+	case m.OnActivityChan <- ev:
 	}
 }

client/internal/lazyconn/activity/manager_test.go

@@ -44,6 +44,10 @@ func (m MocWGIface) Address() wgaddr.Address {
 	}
 }
 
+func (m MocWGIface) MTU() uint16 {
+	return 1280
+}
+
 // GetPeerListener is a test helper to access listeners
 func (m *Manager) GetPeerListener(peerConnID peerid.ConnID) (listener, bool) {
 	m.mu.Lock()
@@ -86,9 +90,9 @@ func TestManager_MonitorPeerActivity(t *testing.T) {
 	}
 
 	select {
-	case peerConnID := <-mgr.OnActivityChan:
-		if peerConnID != peerCfg1.PeerConnID {
-			t.Fatalf("unexpected peerConnID: %v", peerConnID)
+	case ev := <-mgr.OnActivityChan:
+		if ev.PeerConnID != peerCfg1.PeerConnID {
+			t.Fatalf("unexpected peerConnID: %v", ev.PeerConnID)
 		}
 	case <-time.After(1 * time.Second):
 	}

client/internal/lazyconn/manager/manager.go

@@ -130,8 +130,8 @@ func (m *Manager) Start(ctx context.Context) {
 		select {
 		case <-ctx.Done():
 			return
-		case peerConnID := <-m.activityManager.OnActivityChan:
-			m.onPeerActivity(peerConnID)
+		case ev := <-m.activityManager.OnActivityChan:
+			m.onPeerActivity(ev)
 		case peerIDs := <-m.inactivityManager.InactivePeersChan():
 			m.onPeerInactivityTimedOut(peerIDs)
 		}
@@ -513,13 +513,13 @@ func (m *Manager) checkHaGroupActivity(haGroup route.HAUniqueID, peerID string,
 	return false
 }
 
-func (m *Manager) onPeerActivity(peerConnID peerid.ConnID) {
+func (m *Manager) onPeerActivity(ev activity.Event) {
 	m.managedPeersMu.Lock()
 	defer m.managedPeersMu.Unlock()
 
-	mp, ok := m.managedPeersByConnID[peerConnID]
+	mp, ok := m.managedPeersByConnID[ev.PeerConnID]
 	if !ok {
-		log.Errorf("peer not found by conn id: %v", peerConnID)
+		log.Errorf("peer not found by conn id: %v", ev.PeerConnID)
 		return
 	}
 
@@ -536,7 +536,7 @@ func (m *Manager) onPeerActivity(peerConnID peerid.ConnID) {
 
 	m.activateHAGroupPeers(mp.peerCfg)
 
-	m.peerStore.PeerConnOpen(m.engineCtx, mp.peerCfg.PublicKey)
+	m.peerStore.PeerConnOpenWithFirstPacket(m.engineCtx, mp.peerCfg.PublicKey, ev.FirstPacket)
 }
 
 func (m *Manager) onPeerInactivityTimedOut(peerIDs map[string]struct{}) {

client/internal/lazyconn/wgiface.go

@@ -17,4 +17,5 @@ type WGIface interface {
 	IsUserspaceBind() bool
 	Address() wgaddr.Address
 	LastActivities() map[string]monotime.Time
+	MTU() uint16
 }

client/internal/peer/conn.go

@@ -136,6 +136,51 @@ type Conn struct {
 	// Connection stage timestamps for metrics
 	metricsRecorder MetricsRecorder
 	metricsStages   *MetricsStages
+
+	// pendingFirstPacket is the lazyconn-captured handshake init, replayed once the real
+	// transport is up.
+	pendingFirstPacket []byte
+}
+
+// SetPendingFirstPacket stashes a packet to be replayed once the connection is established.
+func (conn *Conn) SetPendingFirstPacket(b []byte) {
+	conn.mu.Lock()
+	conn.pendingFirstPacket = append([]byte(nil), b...)
+	conn.mu.Unlock()
+}
+
+// takePendingFirstPacket returns and clears the stashed first packet. Caller must hold conn.mu.
+func (conn *Conn) takePendingFirstPacket() []byte {
+	pkt := conn.pendingFirstPacket
+	conn.pendingFirstPacket = nil
+	return pkt
+}
+
+// injectPendingFirstPacket replays the captured handshake through the proxy if present, else
+// directly through the ICE conn. Caller must hold conn.mu.
+func (conn *Conn) injectPendingFirstPacket(proxy wgproxy.Proxy, directConn net.Conn) {
+	pkt := conn.takePendingFirstPacket()
+	if len(pkt) == 0 {
+		return
+	}
+
+	switch {
+	case proxy != nil:
+		if err := proxy.InjectPacket(pkt); err != nil {
+			conn.Log.Debugf("failed to reinject captured first packet via proxy: %v", err)
+			return
+		}
+	case directConn != nil:
+		if _, err := directConn.Write(pkt); err != nil {
+			conn.Log.Debugf("failed to reinject captured first packet via direct conn: %v", err)
+			return
+		}
+	default:
+		conn.Log.Debugf("no transport available to reinject captured first packet")
+		return
+	}
+
+	conn.Log.Debugf("reinjected captured first packet (%d bytes)", len(pkt))
 }
 
 // NewConn creates a new not opened Conn to the remote peer.
@@ -423,6 +468,8 @@ func (conn *Conn) onICEConnectionIsReady(priority conntype.ConnPriority, iceConn
 		conn.wgProxyRelay.RedirectAs(ep)
 	}
 
+	conn.injectPendingFirstPacket(wgProxy, iceConnInfo.RemoteConn)
+
 	conn.currentConnPriority = priority
 	conn.statusICE.SetConnected()
 	conn.updateIceState(iceConnInfo, updateTime)
@@ -546,6 +593,8 @@ func (conn *Conn) onRelayConnectionIsReady(rci RelayConnInfo) {
 
 	wgConfigWorkaround()
 
+	conn.injectPendingFirstPacket(wgProxy, nil)
+
 	conn.rosenpassRemoteKey = rci.rosenpassPubKey
 	conn.currentConnPriority = conntype.Relay
 	conn.statusRelay.SetConnected()

client/internal/peerstore/store.go

@@ -81,6 +81,12 @@ func (s *Store) PeerConn(pubKey string) (*peer.Conn, bool) {
 }
 
 func (s *Store) PeerConnOpen(ctx context.Context, pubKey string) {
+	s.PeerConnOpenWithFirstPacket(ctx, pubKey, nil)
+}
+
+// PeerConnOpenWithFirstPacket opens the peer connection and stashes a first packet to be
+// reinjected once the real transport is established.
+func (s *Store) PeerConnOpenWithFirstPacket(ctx context.Context, pubKey string, firstPacket []byte) {
 	s.peerConnsMu.RLock()
 	defer s.peerConnsMu.RUnlock()
 
@@ -88,11 +94,13 @@ func (s *Store) PeerConnOpen(ctx context.Context, pubKey string) {
 	if !ok {
 		return
 	}
+	if len(firstPacket) > 0 {
+		p.SetPendingFirstPacket(firstPacket)
+	}
 	// this can be blocked because of the connect open limiter semaphore
 	if err := p.Open(ctx); err != nil {
 		p.Log.Errorf("failed to open peer connection: %v", err)
 	}
-
 }
 
 func (s *Store) PeerConnIdle(pubKey string) {

management/server/store/sql_store.go

@@ -1216,6 +1216,7 @@ func (s *SqlStore) getAccountGorm(ctx context.Context, accountID string) (*types
 		Preload("NetworkResources").
 		Preload("Onboarding").
 		Preload("Services.Targets").
+		Preload("Domains").
 		Take(&account, idQueryCondition, accountID)
 	if result.Error != nil {
 		log.WithContext(ctx).Errorf("error when getting account %s from the store: %s", accountID, result.Error)
@@ -1302,7 +1303,7 @@ func (s *SqlStore) getAccountPgx(ctx context.Context, accountID string) (*types.
 	}
 
 	var wg sync.WaitGroup
-	errChan := make(chan error, 12)
+	errChan := make(chan error, 16)
 
 	wg.Add(1)
 	go func() {
@@ -1403,6 +1404,17 @@ func (s *SqlStore) getAccountPgx(ctx context.Context, accountID string) (*types.
 		account.Services = services
 	}()
 
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		domains, err := s.ListCustomDomains(ctx, accountID)
+		if err != nil {
+			errChan <- err
+			return
+		}
+		account.Domains = domains
+	}()
+
 	wg.Add(1)
 	go func() {
 		defer wg.Done()

management/server/store/sql_store_get_account_test.go

@@ -4,6 +4,8 @@ import (
 	"context"
 	"net"
 	"net/netip"
+	"os"
+	"runtime"
 	"testing"
 	"time"
 
@@ -21,6 +23,63 @@ import (
 	"github.com/netbirdio/netbird/route"
 )
 
+// TestGetAccount_LoadsCustomDomains verifies GetAccount populates account.Domains.
+// SynthesizePrivateServiceZones depends on this relation to anchor a custom-domain
+// private service's DNS zone; without the preload the relation is empty and the
+// service is silently skipped, so a custom domain never resolves on clients.
+func TestGetAccount_LoadsCustomDomains(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("The SQLite store is not properly supported by Windows yet")
+	}
+
+	store, cleanup, err := NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	defer cleanup()
+
+	assertGetAccountLoadsCustomDomains(t, store)
+}
+
+func TestPostgresql_GetAccount_LoadsCustomDomains(t *testing.T) {
+	if (os.Getenv("CI") == "true" && runtime.GOOS == "darwin") || runtime.GOOS == "windows" {
+		t.Skip("skip CI tests on darwin and windows")
+	}
+
+	t.Setenv("NETBIRD_STORE_ENGINE", string(types.PostgresStoreEngine))
+	store, cleanup, err := NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	assertGetAccountLoadsCustomDomains(t, store)
+}
+
+// assertGetAccountLoadsCustomDomains exercises both the gorm and pgx GetAccount
+// paths: it persists two custom domains and asserts the relation comes back
+// populated, which SynthesizePrivateServiceZones relies on.
+func assertGetAccountLoadsCustomDomains(t *testing.T, store Store) {
+	t.Helper()
+	ctx := context.Background()
+
+	accountID := "acct-custom-domains"
+	require.NoError(t, store.SaveAccount(ctx, newAccountWithId(ctx, accountID, "user-1", "")))
+
+	_, err := store.CreateCustomDomain(ctx, accountID, "example.com", "eu.proxy.netbird.io", true)
+	require.NoError(t, err, "creating the first custom domain must succeed")
+	_, err = store.CreateCustomDomain(ctx, accountID, "apps.acme.io", "us.proxy.netbird.io", false)
+	require.NoError(t, err, "creating the second custom domain must succeed")
+
+	account, err := store.GetAccount(ctx, accountID)
+	require.NoError(t, err)
+	require.Len(t, account.Domains, 2, "GetAccount must preload the account's custom domains")
+
+	byDomain := map[string]string{}
+	for _, d := range account.Domains {
+		require.NotNil(t, d)
+		byDomain[d.Domain] = d.TargetCluster
+	}
+	assert.Equal(t, "eu.proxy.netbird.io", byDomain["example.com"], "custom domain must carry its target cluster")
+	assert.Equal(t, "us.proxy.netbird.io", byDomain["apps.acme.io"], "custom domain must carry its target cluster")
+}
+
 // TestGetAccount_ComprehensiveFieldValidation validates that GetAccount properly loads
 // all fields and nested objects from the database, including deeply nested structures.
 func TestGetAccount_ComprehensiveFieldValidation(t *testing.T) {

management/server/types/account.go

@@ -273,7 +273,7 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 	}
 
 	peerGroups := a.GetPeerGroups(peerID)
-	zonesByCluster := map[string]*nbdns.CustomZone{}
+	zonesByApex := map[string]*nbdns.CustomZone{}
 
 	for _, svc := range a.Services {
 		if svc == nil || !svc.Enabled || !svc.Private {
@@ -290,19 +290,24 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 			continue
 		}
 
-		zone, exists := zonesByCluster[svc.ProxyCluster]
+		serviceDomainZone := a.privateServiceDomainZone(svc)
+		if serviceDomainZone == "" {
+			continue
+		}
+
+		zone, exists := zonesByApex[serviceDomainZone]
 		if !exists {
 			// NonAuthoritative makes this a match-only zone: queries for
 			// names without an explicit record fall through to the
 			// upstream resolver instead of returning NXDOMAIN. Without
 			// it, adding a single private service would black-hole every
-			// other name under the cluster apex.
+			// other name under the zone apex.
 			zone = &nbdns.CustomZone{
-				Domain:           dns.Fqdn(svc.ProxyCluster),
+				Domain:           dns.Fqdn(serviceDomainZone),
 				Records:          []nbdns.SimpleRecord{},
 				NonAuthoritative: true,
 			}
-			zonesByCluster[svc.ProxyCluster] = zone
+			zonesByApex[serviceDomainZone] = zone
 		}
 
 		emitted := 0
@@ -340,8 +345,8 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 		}
 	}
 
-	out := make([]nbdns.CustomZone, 0, len(zonesByCluster))
-	for _, zone := range zonesByCluster {
+	out := make([]nbdns.CustomZone, 0, len(zonesByApex))
+	for _, zone := range zonesByApex {
 		if len(zone.Records) == 0 {
 			continue
 		}
@@ -357,6 +362,33 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 	return out
 }
 
+// privateServiceDomainZone returns the DNS zone name for the given private service domain by
+// looking at the proxy cluster domain then the custom domains.
+func (a *Account) privateServiceDomainZone(svc *service.Service) string {
+	if domainFromSuffix(svc.Domain, svc.ProxyCluster) {
+		return svc.ProxyCluster
+	}
+
+	// Longest matching custom domain wins
+	zoneName := ""
+	for _, d := range a.Domains {
+		if d == nil || d.TargetCluster != svc.ProxyCluster {
+			continue
+		}
+		if domainFromSuffix(svc.Domain, d.Domain) && len(d.Domain) > len(zoneName) {
+			zoneName = d.Domain
+		}
+	}
+	return zoneName
+}
+
+func domainFromSuffix(domain, suffix string) bool {
+	if suffix == "" {
+		return false
+	}
+	return domain == suffix || strings.HasSuffix(domain, "."+suffix)
+}
+
 // peerInDistributionGroups reports whether any of the peer's groups
 // matches the service's bearer-auth distribution_groups.
 func peerInDistributionGroups(peerGroups LookupMap, distributionGroups []string) bool {

management/server/types/account_private_zones_test.go

@@ -11,6 +11,7 @@ import (
 	"github.com/stretchr/testify/require"
 
 	nbdns "github.com/netbirdio/netbird/dns"
+	proxydomain "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain"
 	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
 	nbpeer "github.com/netbirdio/netbird/management/server/peer"
 )
@@ -234,6 +235,113 @@ func TestPrivateZone_GetPeerNetworkMap_PeerOutsideGroups_OmitsSynthZone(t *testi
 	assert.False(t, ok, "peer outside the distribution_groups must not see the synth zone")
 }
 
+func TestSynthesizePrivateServiceZones_CustomDomain_ZoneApexIsRegisteredDomain(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	// A custom-domain service: Domain is the custom FQDN, ProxyCluster
+	// is the cluster serving it, and account.Domains holds the registered
+	// custom domain. The synth zone apex must be the registered domain,
+	// not the cluster, or the client's match-only zone never intercepts
+	// the query.
+	account.Services[0].Domain = "app.example.com"
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 1, "custom-domain service must still produce one zone")
+	zone := zones[0]
+	assert.Equal(t, "example.com.", zone.Domain, "zone apex must be the registered custom domain, not the cluster or the service FQDN")
+	assert.True(t, zone.NonAuthoritative, "synth zone must remain match-only")
+	require.Len(t, zone.Records, 1, "custom-domain service yields one A record")
+	rec := zone.Records[0]
+	assert.Equal(t, "app.example.com.", rec.Name, "record name is the custom service FQDN")
+	assert.Equal(t, "100.64.0.99", rec.RData, "record points at the embedded proxy peer's tunnel IP")
+}
+
+func TestSynthesizePrivateServiceZones_CustomAndFreeDomain_SeparateZones(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+	account.Services = append(account.Services, &service.Service{
+		ID:           "svc-2",
+		AccountID:    "acct-1",
+		Name:         "custom",
+		Domain:       "app.example.com",
+		ProxyCluster: "eu.proxy.netbird.io",
+		Enabled:      true,
+		Private:      true,
+		Mode:         service.ModeHTTP,
+		AccessGroups: []string{"grp-admins"},
+	})
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 2, "a free-domain and a custom-domain service must not collapse into one zone")
+
+	free, ok := findCustomZone(zones, "eu.proxy.netbird.io")
+	require.True(t, ok, "free-domain service keeps the shared cluster-apex zone")
+	require.Len(t, free.Records, 1, "cluster zone carries only the free-domain record")
+	assert.Equal(t, "myapp.eu.proxy.netbird.io.", free.Records[0].Name, "cluster zone record is the free-domain FQDN")
+
+	custom, ok := findCustomZone(zones, "example.com")
+	require.True(t, ok, "custom-domain service gets its own zone at the registered custom domain apex")
+	require.Len(t, custom.Records, 1, "custom zone carries only the custom-domain record")
+	assert.Equal(t, "app.example.com.", custom.Records[0].Name, "custom zone record is the custom-domain FQDN")
+}
+
+func TestSynthesizePrivateServiceZones_TwoServicesSameCustomDomain_OneZone(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+	account.Services[0].Domain = "a.example.com"
+	account.Services = append(account.Services, &service.Service{
+		ID:           "svc-2",
+		AccountID:    "acct-1",
+		Name:         "bapp",
+		Domain:       "b.example.com",
+		ProxyCluster: "eu.proxy.netbird.io",
+		Enabled:      true,
+		Private:      true,
+		Mode:         service.ModeHTTP,
+		AccessGroups: []string{"grp-admins"},
+	})
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 1, "two services under the same registered custom domain must share one zone")
+	assert.Equal(t, "example.com.", zones[0].Domain, "shared zone apex is the registered custom domain")
+	require.Len(t, zones[0].Records, 2, "both services surface as records in the shared custom-domain zone")
+	names := []string{zones[0].Records[0].Name, zones[0].Records[1].Name}
+	assert.ElementsMatch(t, []string{"a.example.com.", "b.example.com."}, names, "both custom-domain service FQDNs must surface")
+}
+
+func TestSynthesizePrivateServiceZones_CustomDomainNotRegistered_NoZone(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	// Service domain is outside the cluster and no account.Domains entry
+	// covers it: there is no apex that would intercept the query, so the
+	// service must be skipped rather than emit an unmatchable record.
+	account.Services[0].Domain = "app.example.com"
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	assert.Empty(t, zones, "a custom-domain service with no registered domain apex must not produce a zone")
+}
+
+func TestSynthesizePrivateServiceZones_CustomDomainClusterMismatch_NoZone(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	// The registered custom domain matches the service FQDN by suffix but
+	// targets a different cluster than the service's ProxyCluster. It must
+	// be ignored, leaving no apex to intercept the query — otherwise the
+	// zone would point at this cluster's proxy peers under a domain owned
+	// by a different cluster.
+	account.Services[0].Domain = "app.example.com"
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "us.proxy.netbird.io", Validated: true},
+	}
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	assert.Empty(t, zones, "a custom domain targeting a different cluster must not anchor the service zone")
+}
+
 func TestSynthesizePrivateServiceZones_TwoServicesSameCluster_OneZone(t *testing.T) {
 	account := privateZoneTestAccount(t)
 	account.Services = append(account.Services, &service.Service{
@@ -254,3 +362,72 @@ func TestSynthesizePrivateServiceZones_TwoServicesSameCluster_OneZone(t *testing
 	names := []string{zones[0].Records[0].Name, zones[0].Records[1].Name}
 	assert.ElementsMatch(t, []string{"myapp.eu.proxy.netbird.io.", "anotherapp.eu.proxy.netbird.io."}, names, "both service domains must surface")
 }
+
+func TestSynthesizePrivateServiceZones_MixedClusterCustomAndPublic(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+
+	privateService := func(id, domain string) *service.Service {
+		return &service.Service{
+			ID:           id,
+			AccountID:    "acct-1",
+			Name:         id,
+			Domain:       domain,
+			ProxyCluster: "eu.proxy.netbird.io",
+			Enabled:      true,
+			Private:      true,
+			Mode:         service.ModeHTTP,
+			AccessGroups: []string{"grp-admins"},
+		}
+	}
+	publicService := func(id, domain string) *service.Service {
+		s := privateService(id, domain)
+		s.Private = false
+		return s
+	}
+
+	account.Services = []*service.Service{
+		// 3 private services under the cluster suffix.
+		privateService("cluster-1", "cluster1.eu.proxy.netbird.io"),
+		privateService("cluster-2", "cluster2.eu.proxy.netbird.io"),
+		privateService("cluster-3", "cluster3.eu.proxy.netbird.io"),
+		// 4 private services under the custom domain suffix.
+		privateService("custom-1", "custom1.example.com"),
+		privateService("custom-2", "custom2.example.com"),
+		privateService("custom-3", "custom3.example.com"),
+		privateService("custom-4", "custom4.example.com"),
+		// 2 public services, one per suffix, must not surface.
+		publicService("public-cluster", "public.eu.proxy.netbird.io"),
+		publicService("public-custom", "public.example.com"),
+	}
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 2, "one zone per apex: the cluster apex and the custom domain apex")
+
+	cluster, ok := findCustomZone(zones, "eu.proxy.netbird.io")
+	require.True(t, ok, "cluster-suffix services collapse into the cluster-apex zone")
+	clusterNames := recordNames(cluster)
+	assert.ElementsMatch(t,
+		[]string{"cluster1.eu.proxy.netbird.io.", "cluster2.eu.proxy.netbird.io.", "cluster3.eu.proxy.netbird.io."},
+		clusterNames,
+		"only the 3 private cluster services surface in the cluster zone (public one excluded)")
+
+	custom, ok := findCustomZone(zones, "example.com")
+	require.True(t, ok, "custom-suffix services collapse into the custom-domain-apex zone")
+	customNames := recordNames(custom)
+	assert.ElementsMatch(t,
+		[]string{"custom1.example.com.", "custom2.example.com.", "custom3.example.com.", "custom4.example.com."},
+		customNames,
+		"only the 4 private custom services surface in the custom zone (public one excluded)")
+}
+
+// recordNames returns the record names of a zone for order-independent assertions.
+func recordNames(zone nbdns.CustomZone) []string {
+	names := make([]string, 0, len(zone.Records))
+	for _, r := range zone.Records {
+		names = append(names, r.Name)
+	}
+	return names
+}