[management] log user agent and return request id (#6380)

client/firewall/iptables/acl_linux.go

@@ -3,6 +3,7 @@ package iptables
 import (
 	"errors"
 	"fmt"
+	"maps"
 	"net"
 	"slices"
 
@@ -421,12 +422,17 @@ func (m *aclManager) updateState() {
 	currentState.Lock()
 	defer currentState.Unlock()
 
+	// Clone the maps so the persisted state holds a private snapshot. The
+	// live maps keep being mutated by subsequent rule operations while the
+	// state manager marshals the state from its periodic-save goroutine.
+	// Sharing them by reference races the two and aborts the process with a
+	// concurrent map iteration and write.
 	if m.v6 {
-		currentState.ACLEntries6 = m.entries
-		currentState.ACLIPsetStore6 = m.ipsetStore
+		currentState.ACLEntries6 = maps.Clone(m.entries)
+		currentState.ACLIPsetStore6 = m.ipsetStore.clone()
 	} else {
-		currentState.ACLEntries = m.entries
-		currentState.ACLIPsetStore = m.ipsetStore
+		currentState.ACLEntries = maps.Clone(m.entries)
+		currentState.ACLIPsetStore = m.ipsetStore.clone()
 	}
 
 	if err := m.stateManager.UpdateState(currentState); err != nil {

client/firewall/iptables/router_linux.go

@@ -4,6 +4,7 @@ package iptables
 
 import (
 	"fmt"
+	"maps"
 	"net/netip"
 	"strconv"
 	"strings"
@@ -749,11 +750,17 @@ func (r *router) updateState() {
 	currentState.Lock()
 	defer currentState.Unlock()
 
+	// Clone the rule map so the persisted state holds a private snapshot. The
+	// live map keeps being mutated by subsequent rule operations while the
+	// state manager marshals the state from its periodic-save goroutine.
+	// Sharing it by reference races the two and aborts the process with a
+	// concurrent map iteration and write. The ipset counter guards itself
+	// during marshaling, so it can be shared directly.
 	if r.v6 {
-		currentState.RouteRules6 = r.rules
+		currentState.RouteRules6 = maps.Clone(r.rules)
 		currentState.RouteIPsetCounter6 = r.ipsetCounter
 	} else {
-		currentState.RouteRules = r.rules
+		currentState.RouteRules = maps.Clone(r.rules)
 		currentState.RouteIPsetCounter = r.ipsetCounter
 	}
 

client/firewall/iptables/rulestore_linux.go

@@ -1,6 +1,9 @@
 package iptables
 
-import "encoding/json"
+import (
+	"encoding/json"
+	"maps"
+)
 
 type ipList struct {
 	ips map[string]struct{}
@@ -19,6 +22,14 @@ func (s *ipList) addIP(ip string) {
 	s.ips[ip] = struct{}{}
 }
 
+// clone returns a deep copy of the ipList with its own ips map.
+func (s *ipList) clone() *ipList {
+	if s == nil {
+		return nil
+	}
+	return &ipList{ips: maps.Clone(s.ips)}
+}
+
 // MarshalJSON implements json.Marshaler
 func (s *ipList) MarshalJSON() ([]byte, error) {
 	return json.Marshal(struct {
@@ -55,6 +66,19 @@ func newIpsetStore() *ipsetStore {
 	}
 }
 
+// clone returns a deep copy of the ipsetStore with its own ipsets map and
+// independent ipList entries.
+func (s *ipsetStore) clone() *ipsetStore {
+	if s == nil {
+		return nil
+	}
+	cloned := &ipsetStore{ipsets: make(map[string]*ipList, len(s.ipsets))}
+	for name, list := range s.ipsets {
+		cloned.ipsets[name] = list.clone()
+	}
+	return cloned
+}
+
 func (s *ipsetStore) ipset(ipsetName string) (*ipList, bool) {
 	r, ok := s.ipsets[ipsetName]
 	return r, ok

client/internal/dns/server.go

@@ -777,13 +777,24 @@ func (s *DefaultServer) applyHostConfig() {
 // context is released rather than leaked until GC.
 func (s *DefaultServer) registerFallback() {
 	originalNameservers := s.hostManager.getOriginalNameservers()
-	if len(originalNameservers) == 0 {
+
+	serverIP := s.service.RuntimeIP()
+	var servers []netip.AddrPort
+	for _, ns := range originalNameservers {
+		if ns == serverIP {
+			log.Debugf("skipping original nameserver %s as it is the same as the server IP %s", ns, serverIP)
+			continue
+		}
+		servers = append(servers, netip.AddrPortFrom(ns, DefaultPort))
+	}
+
+	if len(servers) == 0 {
 		log.Debugf("no fallback upstreams to register; clearing PriorityFallback handler")
 		s.clearFallback()
 		return
 	}
 
-	log.Infof("registering original nameservers %v as upstream handlers with priority %d", originalNameservers, PriorityFallback)
+	log.Infof("registering original nameservers %v as upstream handlers with priority %d", servers, PriorityFallback)
 
 	handler, err := newUpstreamResolver(
 		s.ctx,
@@ -797,11 +808,6 @@ func (s *DefaultServer) registerFallback() {
 		return
 	}
 	handler.selectedRoutes = s.selectedRoutes
-
-	var servers []netip.AddrPort
-	for _, ns := range originalNameservers {
-		servers = append(servers, netip.AddrPortFrom(ns, DefaultPort))
-	}
 	handler.addRace(servers)
 
 	prev := s.fallbackHandler

client/internal/routemanager/manager.go

@@ -700,6 +700,13 @@ func resolveURLsToIPs(urls []string) []net.IP {
 
 // updateRouteSelectorFromManagement updates the route selector based on the isSelected status from the management server
 func (m *DefaultManager) updateRouteSelectorFromManagement(clientRoutes route.HAMap) {
+	// An explicit user "deselect all" must not be overridden by management auto-apply.
+	// Auto-applying an exit node here would call SelectRoutes, which clears the
+	// deselect-all flag and re-enables every route the user turned off.
+	if m.routeSelector.IsDeselectAll() {
+		return
+	}
+
 	exitNodeInfo := m.collectExitNodeInfo(clientRoutes)
 	if len(exitNodeInfo.allIDs) == 0 {
 		return

client/internal/routemanager/selector_management_test.go

@@ -0,0 +1,71 @@
+package routemanager
+
+import (
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/routeselector"
+	"github.com/netbirdio/netbird/route"
+)
+
+func exitNodeRoutes(netID route.NetID, skipAutoApply bool) route.HAMap {
+	haID := route.HAUniqueID(string(netID) + "|0.0.0.0/0")
+	return route.HAMap{
+		haID: []*route.Route{
+			{
+				ID:            "r-" + route.ID(netID),
+				NetID:         netID,
+				Network:       netip.MustParsePrefix("0.0.0.0/0"),
+				NetworkType:   route.IPv4Network,
+				Enabled:       true,
+				SkipAutoApply: skipAutoApply,
+			},
+		},
+	}
+}
+
+func TestUpdateRouteSelectorFromManagement(t *testing.T) {
+	t.Run("management auto-apply selects exit node without user selection", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		routes := exitNodeRoutes("exit1", false)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.True(t, m.routeSelector.IsSelected("exit1"), "auto-apply exit node should be selected")
+		require.Len(t, m.routeSelector.FilterSelectedExitNodes(routes), 1, "selected exit node should pass the filter")
+	})
+
+	t.Run("management SkipAutoApply leaves exit node deselected", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		routes := exitNodeRoutes("exit1", true)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.False(t, m.routeSelector.IsSelected("exit1"), "SkipAutoApply exit node should not be selected")
+		require.Empty(t, m.routeSelector.FilterSelectedExitNodes(routes), "deselected exit node should be filtered out")
+	})
+
+	t.Run("user selection is not overridden by management", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		require.NoError(t, m.routeSelector.SelectRoutes([]route.NetID{"exit1"}, true, []route.NetID{"exit1"}))
+		routes := exitNodeRoutes("exit1", true)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.True(t, m.routeSelector.IsSelected("exit1"), "explicit user selection must survive a management sync that wants to skip auto-apply")
+		require.Len(t, m.routeSelector.FilterSelectedExitNodes(routes), 1, "user-selected exit node should pass the filter")
+	})
+
+	t.Run("deselect-all is preserved across a management sync", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		m.routeSelector.DeselectAllRoutes()
+		routes := exitNodeRoutes("exit1", false)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.True(t, m.routeSelector.IsDeselectAll(), "an explicit deselect-all must not be cleared by management auto-apply")
+		require.Empty(t, m.routeSelector.FilterSelectedExitNodes(routes), "no routes should be selected while deselect-all is set")
+	})
+}

client/internal/routeselector/routeselector.go

@@ -116,6 +116,14 @@ func (rs *RouteSelector) DeselectAllRoutes() {
 	clear(rs.selectedRoutes)
 }
 
+// IsDeselectAll reports whether the user has explicitly deselected all routes.
+func (rs *RouteSelector) IsDeselectAll() bool {
+	rs.mu.RLock()
+	defer rs.mu.RUnlock()
+
+	return rs.deselectAll
+}
+
 // IsSelected checks if a specific route is selected.
 func (rs *RouteSelector) IsSelected(routeID route.NetID) bool {
 	rs.mu.RLock()

formatter/hook/hook.go

@@ -99,6 +99,9 @@ func addFields(entry *logrus.Entry) {
 	if ctxAccountID, ok := entry.Context.Value(context.AccountIDKey).(string); ok {
 		entry.Data[context.AccountIDKey] = ctxAccountID
 	}
+	if ctxUserAgent, ok := entry.Context.Value(context.UserAgentKey).(string); ok {
+		entry.Data[context.UserAgentKey] = ctxUserAgent
+	}
 	if ctxInitiatorID, ok := entry.Context.Value(context.UserIDKey).(string); ok {
 		entry.Data[context.UserIDKey] = ctxInitiatorID
 	}

management/internals/shared/grpc/proxy.go

@@ -666,8 +666,10 @@ func (s *ProxyServiceServer) sender(conn *proxyConnection, errChan chan<- error)
 		case resp := <-conn.sendChan:
 			if err := conn.sendResponse(resp); err != nil {
 				errChan <- err
+				log.WithContext(conn.ctx).Tracef("Failed to send response to proxy %s: %v", conn.proxyID, err)
 				return
 			}
+			log.WithContext(conn.ctx).Tracef("Send response to proxy %s", conn.proxyID)
 		case <-conn.ctx.Done():
 			return
 		}

management/server/context/keys.go

@@ -12,6 +12,7 @@ const (
 	RoleKey      = nbcontext.RoleKey
 	UserIDKey    = nbcontext.UserIDKey
 	PeerIDKey    = nbcontext.PeerIDKey
+	UserAgentKey = nbcontext.UserAgentKey
 )
 
 // RoleFromContext returns the role stored in ctx, or empty string and false if absent.

management/server/telemetry/http_api_metrics.go

@@ -21,6 +21,8 @@ const (
 	httpRequestCounterPrefix  = "management.http.request.counter"
 	httpResponseCounterPrefix = "management.http.response.counter"
 	httpRequestDurationPrefix = "management.http.request.duration.ms"
+
+	RequestIDHeader = "X-Request-Id"
 )
 
 // WrappedResponseWriter is a wrapper for http.ResponseWriter that allows the
@@ -172,6 +174,10 @@ func (m *HTTPMiddleware) Handler(h http.Handler) http.Handler {
 		reqID := xid.New().String()
 		//nolint
 		ctx = context.WithValue(ctx, nbContext.RequestIDKey, reqID)
+		//nolint
+		ctx = context.WithValue(ctx, nbContext.UserAgentKey, r.UserAgent())
+
+		rw.Header().Set(RequestIDHeader, reqID)
 
 		log.WithContext(ctx).Tracef("HTTP request %v: %v %v", reqID, r.Method, r.URL)
 

management/server/types/networkmap_components.go

@@ -557,7 +557,6 @@ func (c *NetworkMapComponents) getRoutingPeerRoutes(peerID string) (enabledRoute
 	return enabledRoutes, disabledRoutes
 }
 
-
 func (c *NetworkMapComponents) filterRoutesByGroups(routes []*route.Route, groupListMap LookupMap) []*route.Route {
 	var filteredRoutes []*route.Route
 	for _, r := range routes {
@@ -628,9 +627,14 @@ func (c *NetworkMapComponents) getDefaultPermit(r *route.Route, includeIPv6 bool
 
 	rules := []*RouteFirewallRule{&rule}
 
-	if includeIPv6 && r.IsDynamic() {
+	isDefaultV4 := r.Network.Addr().Is4() && r.Network.Bits() == 0
+	if includeIPv6 && (r.IsDynamic() || isDefaultV4) {
 		ruleV6 := rule
 		ruleV6.SourceRanges = []string{"::/0"}
+		if isDefaultV4 {
+			ruleV6.Destination = "::/0"
+			ruleV6.RouteID = r.ID + "-v6-default"
+		}
 		rules = append(rules, &ruleV6)
 	}
 

management/server/types/networkmap_components_correctness_test.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"net"
 	"net/netip"
+	"slices"
 	"testing"
 	"time"
 
@@ -1029,6 +1030,48 @@ func TestComponents_RouteDefaultPermit(t *testing.T) {
 	assert.True(t, hasDefaultPermit, "route without ACG should have default permit rule with 0.0.0.0/0 source")
 }
 
+// TestComponents_ExitNodeDefaultPermitIPv6 verifies that a default exit node route
+// (0.0.0.0/0) without AccessControlGroups also emits an IPv6 default permit rule
+// (::/0 source and destination) for peers that support IPv6, mirroring the route
+// the client installs. Without it, IPv6 traffic is routed to the exit node but
+// dropped at the forward chain.
+func TestComponents_ExitNodeDefaultPermitIPv6(t *testing.T) {
+	account, validatedPeers := scalableTestAccount(20, 2)
+
+	routingPeerID := "peer-5"
+	routingPeer := account.Peers[routingPeerID]
+	routingPeer.IPv6 = netip.MustParseAddr("fd00::5")
+	routingPeer.Meta.Capabilities = append(routingPeer.Meta.Capabilities, nbpeer.PeerCapabilityIPv6Overlay)
+
+	account.Routes["route-exit"] = &route.Route{
+		ID: "route-exit", Network: netip.MustParsePrefix("0.0.0.0/0"),
+		PeerID: routingPeerID, Peer: routingPeer.Key,
+		Enabled: true, Groups: []string{"group-all"}, PeerGroups: []string{"group-0"},
+		AccessControlGroups: []string{},
+		AccountID:           "test-account",
+	}
+
+	nm := componentsNetworkMap(account, routingPeerID, validatedPeers)
+	require.NotNil(t, nm)
+
+	hasV4 := false
+	hasV6 := false
+	for _, rfr := range nm.RoutesFirewallRules {
+		switch rfr.Destination {
+		case "0.0.0.0/0":
+			if slices.Contains(rfr.SourceRanges, "0.0.0.0/0") {
+				hasV4 = true
+			}
+		case "::/0":
+			if slices.Contains(rfr.SourceRanges, "::/0") {
+				hasV6 = true
+			}
+		}
+	}
+	assert.True(t, hasV4, "exit node route should have an IPv4 default permit rule (0.0.0.0/0)")
+	assert.True(t, hasV6, "exit node route should have an IPv6 default permit rule (::/0)")
+}
+
 // ──────────────────────────────────────────────────────────────────────────────
 // 15. MULTIPLE ROUTERS PER NETWORK
 // ──────────────────────────────────────────────────────────────────────────────

proxy/cmd/proxy/cmd/root.go

@@ -249,6 +249,7 @@ func runServer(cmd *cobra.Command, args []string) error {
 		Private:                  private,
 		MaxDialTimeout:           maxDialTimeout,
 		MaxSessionIdleTimeout:    maxSessionIdleTimeout,
+		MappingBatchWatchdog:     envDurationOrDefault("NB_PROXY_MAPPING_BATCH_WATCHDOG", 0),
 		GeoDataDir:               geoDataDir,
 		CrowdSecAPIURL:           crowdsecAPIURL,
 		CrowdSecAPIKey:           crowdsecAPIKey,

proxy/internal/roundtrip/netbird.go

@@ -28,6 +28,10 @@ import (
 
 const deviceNamePrefix = "ingress-proxy-"
 
+const clientStopTimeout = 30 * time.Second
+
+const createProxyPeerTimeout = 30 * time.Second
+
 // backendKey identifies a backend by its host:port from the target URL.
 type backendKey string
 
@@ -162,6 +166,7 @@ type NetBird struct {
 
 	clientsMux     sync.RWMutex
 	clients        map[types.AccountID]*clientEntry
+	lifecycleMu    sync.Map
 	initLogOnce    sync.Once
 	statusNotifier statusNotifier
 	// readyHandler runs after the embedded client for an account reports
@@ -177,6 +182,10 @@ type NetBird struct {
 	// (i.e. when a new client was actually created, not when an existing one
 	// was reused). The duration covers keygen + gRPC CreateProxyPeer + embed.New.
 	OnAddPeer func(d time.Duration, err error)
+
+	// startClient runs the post-create client startup. Nil uses runClientStartup;
+	// tests override it to avoid a real embed client.Start.
+	startClient func(accountID types.AccountID, client *embed.Client)
 }
 
 // ClientDebugInfo contains debug information about a client.
@@ -200,31 +209,20 @@ type skipTLSVerifyContextKey struct{}
 func (n *NetBird) AddPeer(ctx context.Context, accountID types.AccountID, key ServiceKey, authToken string, serviceID types.ServiceID) error {
 	si := serviceInfo{serviceID: serviceID}
 
-	n.clientsMux.Lock()
+	if n.registerExistingClient(accountID, key, si) {
+		return nil
+	}
 
-	entry, exists := n.clients[accountID]
-	if exists {
-		entry.services[key] = si
-		started := entry.started
-		n.clientsMux.Unlock()
-
-		n.logger.WithFields(log.Fields{
-			"account_id":  accountID,
-			"service_key": key,
-		}).Debug("registered service with existing client")
-
-		if started && n.statusNotifier != nil {
-			// Use a background context, not the caller's: the management
-			// connection notification must land even if the request /
-			// stream that triggered this registration is cancelled.
-			// Mirrors the async runClientStartup path.
-			if err := n.statusNotifier.NotifyStatus(context.Background(), accountID, serviceID, true); err != nil {
-				n.logger.WithFields(log.Fields{
-					"account_id":  accountID,
-					"service_key": key,
-				}).WithError(err).Warn("failed to notify status for existing client")
-			}
+	lifecycle := n.accountLifecycle(accountID)
+	lifecycle.Lock()
+	transferred := false
+	defer func() {
+		if !transferred {
+			lifecycle.Unlock()
 		}
+	}()
+
+	if n.registerExistingClient(accountID, key, si) {
 		return nil
 	}
 
@@ -234,10 +232,10 @@ func (n *NetBird) AddPeer(ctx context.Context, accountID types.AccountID, key Se
 		n.OnAddPeer(time.Since(createStart), err)
 	}
 	if err != nil {
-		n.clientsMux.Unlock()
 		return err
 	}
 
+	n.clientsMux.Lock()
 	n.clients[accountID] = entry
 	n.clientsMux.Unlock()
 
@@ -246,17 +244,64 @@ func (n *NetBird) AddPeer(ctx context.Context, accountID types.AccountID, key Se
 		"service_key": key,
 	}).Info("created new client for account")
 
-	// Attempt to start the client in the background; if this fails we will
-	// retry on the first request via RoundTrip. runClientStartup uses its
-	// own background context so the caller's request-scoped ctx can't
-	// cancel the inbound bring-up.
-	go n.runClientStartup(accountID, entry.client)
+	transferred = true
+	go func() {
+		defer lifecycle.Unlock()
+		n.startClientStartup(accountID, entry.client)
+	}()
 
 	return nil
 }
 
+func (n *NetBird) startClientStartup(accountID types.AccountID, client *embed.Client) {
+	if n.startClient != nil {
+		n.startClient(accountID, client)
+		return
+	}
+	n.runClientStartup(accountID, client)
+}
+
+// registerExistingClient registers the service against an already-present
+// client for the account and returns true when it did. It notifies management
+// of the new service when the client is already started.
+func (n *NetBird) registerExistingClient(accountID types.AccountID, key ServiceKey, si serviceInfo) bool {
+	n.clientsMux.Lock()
+	entry, exists := n.clients[accountID]
+	if !exists {
+		n.clientsMux.Unlock()
+		return false
+	}
+	entry.services[key] = si
+	started := entry.started
+	n.clientsMux.Unlock()
+
+	n.logger.WithFields(log.Fields{
+		"account_id":  accountID,
+		"service_key": key,
+	}).Debug("registered service with existing client")
+
+	if started && n.statusNotifier != nil {
+		if err := n.statusNotifier.NotifyStatus(context.Background(), accountID, si.serviceID, true); err != nil {
+			n.logger.WithFields(log.Fields{
+				"account_id":  accountID,
+				"service_key": key,
+			}).WithError(err).Warn("failed to notify status for existing client")
+		}
+	}
+	return true
+}
+
+// accountLifecycle returns the per-account lifecycle mutex, serialising client
+// creation against teardown so a slow client.Stop cannot race a new
+// client.Start for the same account, without blocking clientsMux.
+func (n *NetBird) accountLifecycle(accountID types.AccountID) *sync.Mutex {
+	mu, _ := n.lifecycleMu.LoadOrStore(accountID, &sync.Mutex{})
+	return mu.(*sync.Mutex)
+}
+
 // createClientEntry generates a WireGuard keypair, authenticates with management,
-// and creates an embedded NetBird client. Must be called with clientsMux held.
+// and creates an embedded NetBird client. Must be called with the account's
+// lifecycle mutex held.
 func (n *NetBird) createClientEntry(ctx context.Context, accountID types.AccountID, key ServiceKey, authToken string, si serviceInfo) (*clientEntry, error) {
 	serviceID := si.serviceID
 	n.logger.WithFields(log.Fields{
@@ -276,7 +321,9 @@ func (n *NetBird) createClientEntry(ctx context.Context, accountID types.Account
 		"public_key": publicKey.String(),
 	}).Debug("authenticating new proxy peer with management")
 
-	resp, err := n.mgmtClient.CreateProxyPeer(ctx, &proto.CreateProxyPeerRequest{
+	createCtx, cancel := context.WithTimeout(ctx, createProxyPeerTimeout)
+	defer cancel()
+	resp, err := n.mgmtClient.CreateProxyPeer(createCtx, &proto.CreateProxyPeerRequest{
 		ServiceId:          string(serviceID),
 		AccountId:          string(accountID),
 		Token:              authToken,
@@ -444,6 +491,15 @@ func (n *NetBird) notifyClientReady(accountID types.AccountID, client *embed.Cli
 // RemovePeer unregisters a service from an account. The client is only stopped
 // when no services are using it anymore.
 func (n *NetBird) RemovePeer(ctx context.Context, accountID types.AccountID, key ServiceKey) error {
+	lifecycle := n.accountLifecycle(accountID)
+	lifecycle.Lock()
+	transferred := false
+	defer func() {
+		if !transferred {
+			lifecycle.Unlock()
+		}
+	}()
+
 	n.clientsMux.Lock()
 
 	entry, exists := n.clients[accountID]
@@ -466,17 +522,8 @@ func (n *NetBird) RemovePeer(ctx context.Context, accountID types.AccountID, key
 	delete(entry.services, key)
 
 	stopClient := len(entry.services) == 0
-	var client *embed.Client
-	var transport, insecureTransport *http.Transport
-	var inbound any
-	var stopHandler func(types.AccountID, any)
 	if stopClient {
 		n.logger.WithField("account_id", accountID).Info("stopping client, no more services")
-		client = entry.client
-		transport = entry.transport
-		insecureTransport = entry.insecureTransport
-		inbound = entry.inbound
-		stopHandler = n.stopHandler
 		delete(n.clients, accountID)
 	} else {
 		n.logger.WithFields(log.Fields{
@@ -490,19 +537,40 @@ func (n *NetBird) RemovePeer(ctx context.Context, accountID types.AccountID, key
 	n.notifyDisconnect(ctx, accountID, key, si.serviceID)
 
 	if stopClient {
-		if inbound != nil && stopHandler != nil {
-			stopHandler(accountID, inbound)
-		}
-		transport.CloseIdleConnections()
-		insecureTransport.CloseIdleConnections()
-		if err := client.Stop(ctx); err != nil {
-			n.logger.WithField("account_id", accountID).WithError(err).Warn("failed to stop netbird client")
-		}
+		transferred = true
+		go n.stopClientLocked(accountID, lifecycle, entry)
 	}
 
 	return nil
 }
 
+// stopClientLocked releases a client's resources off the caller's goroutine so a
+// slow client.Stop cannot wedge the mapping receive loop (which calls RemovePeer
+// synchronously). It unlocks lifecycle when done so a new client.Start for the
+// same account waits for this teardown.
+func (n *NetBird) stopClientLocked(accountID types.AccountID, lifecycle *sync.Mutex, entry *clientEntry) {
+	defer lifecycle.Unlock()
+
+	if entry.inbound != nil && n.stopHandler != nil {
+		n.stopHandler(accountID, entry.inbound)
+	}
+	if entry.transport != nil {
+		entry.transport.CloseIdleConnections()
+	}
+	if entry.insecureTransport != nil {
+		entry.insecureTransport.CloseIdleConnections()
+	}
+	if entry.client == nil {
+		return
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), clientStopTimeout)
+	defer cancel()
+	if err := entry.client.Stop(ctx); err != nil {
+		n.logger.WithField("account_id", accountID).WithError(err).Warn("failed to stop netbird client")
+	}
+}
+
 func (n *NetBird) notifyDisconnect(ctx context.Context, accountID types.AccountID, key ServiceKey, serviceID types.ServiceID) {
 	if n.statusNotifier == nil {
 		return

proxy/internal/roundtrip/netbird_test.go

@@ -6,6 +6,7 @@ import (
 	"net/netip"
 	"sync"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -22,6 +23,18 @@ func (m *mockMgmtClient) CreateProxyPeer(_ context.Context, _ *proto.CreateProxy
 	return &proto.CreateProxyPeerResponse{Success: true}, nil
 }
 
+// signalMgmtClient closes entered the first time CreateProxyPeer is called, so
+// tests can detect AddPeer reaching client creation.
+type signalMgmtClient struct {
+	entered chan struct{}
+	once    sync.Once
+}
+
+func (m *signalMgmtClient) CreateProxyPeer(_ context.Context, _ *proto.CreateProxyPeerRequest, _ ...grpc.CallOption) (*proto.CreateProxyPeerResponse, error) {
+	m.once.Do(func() { close(m.entered) })
+	return &proto.CreateProxyPeerResponse{Success: true}, nil
+}
+
 type mockStatusNotifier struct {
 	mu       sync.Mutex
 	statuses []statusCall
@@ -52,11 +65,15 @@ func (m *mockStatusNotifier) calls() []statusCall {
 // mockNetBird creates a NetBird instance for testing without actually connecting.
 // It uses an invalid management URL to prevent real connections.
 func mockNetBird() *NetBird {
-	return NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
 		MgmtAddr:     "http://invalid.test:9999",
 		WGPort:       0,
 		PreSharedKey: "",
 	}, nil, nil, &mockMgmtClient{})
+	// Skip the real embed client.Start, which would hang against the unreachable
+	// mgmt URL and (now that the lifecycle lock spans startup) serialise removes.
+	nb.startClient = func(types.AccountID, *embed.Client) {}
+	return nb
 }
 
 func TestNetBird_AddPeer_CreatesClientForNewAccount(t *testing.T) {
@@ -288,6 +305,7 @@ func TestNetBird_AddPeer_ExistingStartedClient_NotifiesStatus(t *testing.T) {
 		WGPort:       0,
 		PreSharedKey: "",
 	}, nil, notifier, &mockMgmtClient{})
+	nb.startClient = func(types.AccountID, *embed.Client) {}
 	accountID := types.AccountID("account-1")
 
 	// Add first service — creates a new client entry.
@@ -372,6 +390,117 @@ func TestNetBird_RemovePeer_NotifiesDisconnection(t *testing.T) {
 	assert.False(t, calls[0].connected)
 }
 
+// TestNetBird_RemovePeer_TeardownIsAsync proves the fix for the receive-loop
+// stall: RemovePeer must return promptly even when the client teardown blocks,
+// because teardown runs off the caller's goroutine. The receive loop calls
+// RemovePeer synchronously, so a blocking teardown inline would wedge it.
+func TestNetBird_RemovePeer_TeardownIsAsync(t *testing.T) {
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
+		MgmtAddr: "http://invalid.test:9999",
+	}, nil, &mockStatusNotifier{}, &mockMgmtClient{})
+
+	accountID := types.AccountID("acct-async-teardown")
+	key := DomainServiceKey("svc.example")
+
+	teardownEntered := make(chan struct{})
+	releaseTeardown := make(chan struct{})
+	nb.SetClientLifecycle(nil, func(types.AccountID, any) {
+		close(teardownEntered)
+		<-releaseTeardown
+	})
+
+	nb.clientsMux.Lock()
+	nb.clients[accountID] = &clientEntry{
+		services: map[ServiceKey]serviceInfo{key: {serviceID: types.ServiceID("svc-1")}},
+		started:  true,
+		inbound:  struct{}{},
+	}
+	nb.clientsMux.Unlock()
+
+	done := make(chan error, 1)
+	go func() { done <- nb.RemovePeer(context.Background(), accountID, key) }()
+
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(2 * time.Second):
+		t.Fatal("RemovePeer did not return while teardown was blocked — teardown is not async")
+	}
+
+	select {
+	case <-teardownEntered:
+	case <-time.After(2 * time.Second):
+		t.Fatal("teardown never ran")
+	}
+
+	close(releaseTeardown)
+}
+
+// TestNetBird_AddPeer_WaitsForTeardown proves the lifecycle lock serialises a
+// new client bringup behind an in-flight teardown for the same account, so a
+// slow client.Stop can never race a new client.Start for that account.
+//
+// It targets the handoff race specifically: AddPeer is launched immediately
+// after RemovePeer returns, WITHOUT waiting for the teardown goroutine to start.
+// This only passes if RemovePeer acquires the lifecycle lock synchronously
+// (before returning) and hands it to the teardown goroutine — if the goroutine
+// acquired the lock itself, AddPeer could win the lock in this window and start
+// a replacement client while the old teardown is still pending.
+func TestNetBird_AddPeer_WaitsForTeardown(t *testing.T) {
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
+		MgmtAddr: "http://invalid.test:9999",
+	}, nil, &mockStatusNotifier{}, &mockMgmtClient{})
+	nb.startClient = func(types.AccountID, *embed.Client) {}
+
+	accountID := types.AccountID("acct-serialize")
+	key := DomainServiceKey("svc.example")
+
+	addEntered := make(chan struct{})
+	releaseTeardown := make(chan struct{})
+	nb.SetClientLifecycle(nil, func(types.AccountID, any) {
+		// Block teardown until released. If AddPeer ever reaches createClientEntry
+		// (signalled via the mgmt client below) while we hold the lock, the lock
+		// failed to serialise and the test fails before we release.
+		<-releaseTeardown
+	})
+
+	nb.clientsMux.Lock()
+	nb.clients[accountID] = &clientEntry{
+		services: map[ServiceKey]serviceInfo{key: {serviceID: types.ServiceID("svc-1")}},
+		started:  true,
+		inbound:  struct{}{},
+	}
+	nb.clientsMux.Unlock()
+
+	// createClientEntry calls CreateProxyPeer; closing addEntered there tells us
+	// AddPeer got past the lifecycle lock and into client creation.
+	nb.mgmtClient = &signalMgmtClient{entered: addEntered}
+
+	require.NoError(t, nb.RemovePeer(context.Background(), accountID, key))
+
+	// Launch AddPeer with NO synchronisation against the teardown goroutine.
+	addReturned := make(chan struct{})
+	go func() {
+		_ = nb.AddPeer(context.Background(), accountID, DomainServiceKey("svc2.example"), "key-2", types.ServiceID("svc-2"))
+		close(addReturned)
+	}()
+
+	select {
+	case <-addEntered:
+		t.Fatal("AddPeer entered client creation while teardown held the lifecycle lock — handoff race not closed")
+	case <-addReturned:
+		t.Fatal("AddPeer completed while teardown held the lifecycle lock — not serialised")
+	case <-time.After(300 * time.Millisecond):
+	}
+
+	close(releaseTeardown)
+	select {
+	case <-addReturned:
+	case <-time.After(2 * time.Second):
+		t.Fatal("AddPeer never completed after teardown released the lifecycle lock")
+	}
+}
+
 // TestNotifyClientReady_UsesBackgroundCtx pins the contract that the
 // post-Start hooks (readyHandler + statusNotifier.NotifyStatus) run on
 // a fresh context.Background() rather than inheriting the AddPeer

proxy/lifecycle.go

@@ -114,6 +114,10 @@ type Config struct {
 	MaxDialTimeout time.Duration
 	// MaxSessionIdleTimeout caps the per-service session idle timeout.
 	MaxSessionIdleTimeout time.Duration
+	// MappingBatchWatchdog bounds how long a single mapping batch may spend
+	// being applied before the receive loop reconnects to resync. Zero falls
+	// back to the internal default.
+	MappingBatchWatchdog time.Duration
 
 	// GeoDataDir is the directory containing GeoLite2 MMDB files.
 	GeoDataDir string
@@ -164,6 +168,7 @@ func New(ctx context.Context, cfg Config) *Server {
 		Private:                  cfg.Private,
 		MaxDialTimeout:           cfg.MaxDialTimeout,
 		MaxSessionIdleTimeout:    cfg.MaxSessionIdleTimeout,
+		MappingBatchWatchdog:     cfg.MappingBatchWatchdog,
 		GeoDataDir:               cfg.GeoDataDir,
 		CrowdSecAPIURL:           cfg.CrowdSecAPIURL,
 		CrowdSecAPIKey:           cfg.CrowdSecAPIKey,

proxy/mapping_stall_test.go

@@ -0,0 +1,282 @@
+package proxy
+
+import (
+	"context"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/metadata"
+
+	"github.com/netbirdio/netbird/proxy/internal/roundtrip"
+	"github.com/netbirdio/netbird/proxy/internal/types"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// blockingMgmtClient implements roundtrip's managementClient interface.
+// CreateProxyPeer parks until release is closed, signalling entry on entered.
+// This reproduces the confirmed real-world stall: createClientEntry calls
+// CreateProxyPeer synchronously while holding clientsMux, and the proxy's
+// receive loop calls that path synchronously inside processMappings.
+type blockingMgmtClient struct {
+	entered chan struct{}
+	once    sync.Once
+}
+
+func (b *blockingMgmtClient) CreateProxyPeer(ctx context.Context, _ *proto.CreateProxyPeerRequest, _ ...grpc.CallOption) (*proto.CreateProxyPeerResponse, error) {
+	b.once.Do(func() { close(b.entered) })
+	// Park until the caller's context is cancelled. In production this ctx is
+	// the gRPC mapping-stream context with no per-call timeout, so a slow or
+	// unresponsive CreateProxyPeer parks the receive loop here indefinitely.
+	<-ctx.Done()
+	return nil, ctx.Err()
+}
+
+// gatedMappingStream is a mock GetMappingUpdate client stream that hands out a
+// pre-seeded list of messages, then records how many times Recv advanced. It
+// lets the test observe whether the single-threaded receive loop ever gets
+// past the first (blocking) batch to pull the second message.
+type gatedMappingStream struct {
+	grpc.ClientStream
+	messages []*proto.GetMappingUpdateResponse
+	idx      int32
+}
+
+func (g *gatedMappingStream) Recv() (*proto.GetMappingUpdateResponse, error) {
+	i := int(atomic.LoadInt32(&g.idx))
+	if i >= len(g.messages) {
+		// Block instead of returning EOF so the loop doesn't exit; we only
+		// care whether the loop ever reaches this second Recv at all.
+		select {}
+	}
+	msg := g.messages[i]
+	atomic.AddInt32(&g.idx, 1)
+	return msg, nil
+}
+
+func (g *gatedMappingStream) deliveredCount() int32 { return atomic.LoadInt32(&g.idx) }
+
+func (g *gatedMappingStream) Header() (metadata.MD, error) { return nil, nil } //nolint:nilnil
+func (g *gatedMappingStream) Trailer() metadata.MD         { return nil }
+func (g *gatedMappingStream) CloseSend() error             { return nil }
+func (g *gatedMappingStream) Context() context.Context     { return context.Background() }
+func (g *gatedMappingStream) SendMsg(any) error            { return nil }
+func (g *gatedMappingStream) RecvMsg(any) error            { return nil }
+
+// noopNotifier satisfies roundtrip's statusNotifier interface.
+type noopNotifier struct{}
+
+func (noopNotifier) NotifyStatus(context.Context, types.AccountID, types.ServiceID, bool) error {
+	return nil
+}
+
+// noopProxyClient is a proto.ProxyServiceClient that no-ops the one method the
+// teardown unwind reaches (SendStatusUpdate, via notifyError when the parked
+// AddPeer is cancelled). The embedded nil interface satisfies the rest at
+// compile time; none of those methods are called by this test.
+type noopProxyClient struct {
+	proto.ProxyServiceClient
+}
+
+func (noopProxyClient) SendStatusUpdate(context.Context, *proto.SendStatusUpdateRequest, ...grpc.CallOption) (*proto.SendStatusUpdateResponse, error) {
+	return &proto.SendStatusUpdateResponse{}, nil
+}
+
+// TestMappingStream_StallsWhenApplyBlocks proves the deadlock: the proxy's
+// mapping receive loop processes batches strictly serially, so when applying
+// one batch blocks (here: createClientEntry parked on a synchronous
+// CreateProxyPeer call, exactly as observed in production), the loop never
+// advances to Recv the next batch. Management can keep sending updates onto
+// the stream with no error and no channel overflow, yet the proxy applies
+// nothing further — it is stuck.
+func TestMappingStream_StallsWhenApplyBlocks(t *testing.T) {
+	logger := log.New()
+	logger.SetLevel(log.PanicLevel)
+
+	mgmt := &blockingMgmtClient{
+		entered: make(chan struct{}),
+	}
+
+	nb := roundtrip.NewNetBird(
+		context.Background(),
+		"proxy-test",
+		"proxy.example.com",
+		roundtrip.ClientConfig{},
+		logger,
+		noopNotifier{},
+		mgmt,
+	)
+
+	s := &Server{
+		Logger:       logger,
+		netbird:      nb,
+		mgmtClient:   noopProxyClient{},
+		routerReady:  closedChan(),
+		lastMappings: make(map[types.ServiceID]*proto.ProxyMapping),
+	}
+
+	// First batch: a CREATED mapping for a brand-new account. addMapping ->
+	// netbird.AddPeer -> createClientEntry -> CreateProxyPeer, which blocks.
+	// Empty Path keeps setupHTTPMapping a no-op (it returns early), so the
+	// ONLY blocking point is the synchronous CreateProxyPeer in AddPeer —
+	// no routers/auth need wiring. The second batch exists only to detect
+	// whether the loop ever advances past the blocked first batch.
+	stream := &gatedMappingStream{
+		messages: []*proto.GetMappingUpdateResponse{
+			{
+				Mapping: []*proto.ProxyMapping{
+					{
+						Type:      proto.ProxyMappingUpdateType_UPDATE_TYPE_CREATED,
+						Id:        "svc-1",
+						AccountId: "acct-1",
+						AuthToken: "token-1",
+					},
+				},
+			},
+			{
+				Mapping: []*proto.ProxyMapping{
+					{
+						Type:      proto.ProxyMappingUpdateType_UPDATE_TYPE_CREATED,
+						Id:        "svc-2",
+						AccountId: "acct-2",
+						AuthToken: "token-2",
+					},
+				},
+			},
+		},
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	// Unblock the parked apply on teardown via ctx (CreateProxyPeer returns
+	// ctx.Err()), so the wedged loop goroutine unwinds before embed.New —
+	// avoiding any dependency on collaborators this test deliberately leaves
+	// nil. The deadlock is fully proven before this fires.
+	t.Cleanup(cancel)
+
+	loopDone := make(chan struct{})
+	syncDone := false
+	go func() {
+		defer close(loopDone)
+		_ = s.handleMappingStream(ctx, stream, &syncDone, time.Time{})
+	}()
+
+	// The loop must reach the blocking apply for the first batch.
+	select {
+	case <-mgmt.entered:
+	case <-time.After(2 * time.Second):
+		t.Fatal("receive loop never reached CreateProxyPeer for the first batch")
+	}
+
+	// THE DEADLOCK: while the first batch is parked in CreateProxyPeer, the
+	// single-threaded loop cannot advance. The second batch is never pulled,
+	// even though it is already available on the stream. Give it ample time.
+	// deliveredCount is atomic; syncDone is intentionally not read here because
+	// the loop goroutine owns it (reading it from the test would race).
+	time.Sleep(500 * time.Millisecond)
+	assert.Equal(t, int32(1), stream.deliveredCount(),
+		"loop must NOT consume the second batch while the first is blocked in apply — proxy is stuck")
+
+	select {
+	case <-loopDone:
+		t.Fatal("receive loop returned while it should be wedged in apply")
+	default:
+		// Still wedged, as expected.
+	}
+}
+
+// TestMappingStream_StallsWhenRemoveBlocks proves the deadlock for the REMOVE
+// path observed in production: a mapping remove tears down the account's last
+// embedded client via netbird.RemovePeer -> client.Stop -> Engine.Stop, whose
+// jobExecutorWG.Wait() is unbounded. Because the receive loop is single-
+// threaded, a blocked remove wedges the loop: no further mapping updates of any
+// kind (create/modify/remove) are applied, while management keeps sending them
+// successfully (no send error, no channel-full). Matches the reported symptom:
+// the last log line is a remove that stops a client, then silence.
+func TestMappingStream_StallsWhenRemoveBlocks(t *testing.T) {
+	logger := log.New()
+	logger.SetLevel(log.PanicLevel)
+
+	enteredRemove := make(chan struct{})
+	blockRemove := make(chan struct{})
+	var once sync.Once
+
+	s := &Server{
+		Logger:       logger,
+		mgmtClient:   noopProxyClient{},
+		routerReady:  closedChan(),
+		lastMappings: make(map[types.ServiceID]*proto.ProxyMapping),
+		// Stand in for netbird.RemovePeer -> client.Stop hanging on
+		// Engine.Stop's unbounded jobExecutorWG.Wait(). Only the first remove
+		// blocks; later removes return immediately so the recovery assertion
+		// can observe the loop advancing.
+		removePeer: func(ctx context.Context, _ types.AccountID, _ roundtrip.ServiceKey) error {
+			first := false
+			once.Do(func() {
+				first = true
+				close(enteredRemove)
+			})
+			if !first {
+				return nil
+			}
+			select {
+			case <-blockRemove:
+			case <-ctx.Done():
+			}
+			return nil
+		},
+	}
+
+	// Batch 1 removes a service (blocks in teardown). Batch 2 is a later update
+	// that must never be applied while the remove is wedged.
+	stream := &gatedMappingStream{
+		messages: []*proto.GetMappingUpdateResponse{
+			{
+				Mapping: []*proto.ProxyMapping{
+					{Type: proto.ProxyMappingUpdateType_UPDATE_TYPE_REMOVED, Id: "svc-1", AccountId: "acct-1"},
+				},
+			},
+			{
+				Mapping: []*proto.ProxyMapping{
+					{Type: proto.ProxyMappingUpdateType_UPDATE_TYPE_REMOVED, Id: "svc-2", AccountId: "acct-1"},
+				},
+			},
+		},
+	}
+
+	loopDone := make(chan struct{})
+	syncDone := false
+	go func() {
+		defer close(loopDone)
+		_ = s.handleMappingStream(context.Background(), stream, &syncDone, time.Time{})
+	}()
+
+	select {
+	case <-enteredRemove:
+	case <-time.After(2 * time.Second):
+		t.Fatal("receive loop never reached the blocking remove for the first batch")
+	}
+
+	// THE DEADLOCK: the loop is parked in the blocked remove and cannot advance.
+	// syncDone is owned by the loop goroutine, so it is not read here.
+	time.Sleep(500 * time.Millisecond)
+	assert.Equal(t, int32(1), stream.deliveredCount(),
+		"loop must NOT consume the second batch while the first remove is blocked — proxy is stuck")
+
+	select {
+	case <-loopDone:
+		t.Fatal("receive loop returned while it should be wedged on the remove")
+	default:
+	}
+
+	// Unblock and confirm the wedge was solely the blocked remove: the loop
+	// then advances and consumes the next batch.
+	close(blockRemove)
+	assert.Eventually(t, func() bool {
+		return stream.deliveredCount() >= 2
+	}, 2*time.Second, 5*time.Millisecond,
+		"once the remove unblocks, the loop must advance and consume the next batch")
+}

proxy/server.go

@@ -118,6 +118,9 @@ type Server struct {
 	// The mapping worker waits on this before processing updates.
 	routerReady chan struct{}
 
+	// removePeer defaults to netbird.RemovePeer; overridable in tests.
+	removePeer func(ctx context.Context, accountID types.AccountID, key roundtrip.ServiceKey) error
+
 	// inbound, when non-nil, manages per-account inbound listeners. Set by
 	// initPrivateInbound only when Private is true so the standalone
 	// proxy keeps its zero-overhead default path.
@@ -227,6 +230,10 @@ type Server struct {
 	// Zero means no cap (the proxy honors whatever management sends).
 	// Set via NB_PROXY_MAX_SESSION_IDLE_TIMEOUT for shared deployments.
 	MaxSessionIdleTimeout time.Duration
+	// MappingBatchWatchdog bounds how long a single mapping batch may spend
+	// in processMappings before the receive loop reconnects to resync.
+	// Zero uses defaultMappingBatchWatchdog.
+	MappingBatchWatchdog time.Duration
 }
 
 // clampIdleTimeout returns d capped to MaxSessionIdleTimeout when configured.
@@ -1172,24 +1179,30 @@ func (s *Server) newManagementMappingWorker(ctx context.Context, client proto.Pr
 			s.healthChecker.SetManagementConnected(false)
 		}
 
+		connected := false
+		onConnected := func() { connected = true }
+
 		var streamErr error
 		if syncSupported {
-			streamErr = s.trySyncMappings(ctx, client, &initialSyncDone)
+			streamErr = s.trySyncMappings(ctx, client, &initialSyncDone, onConnected)
 			if isSyncUnimplemented(streamErr) {
 				syncSupported = false
 				s.Logger.Info("management does not support SyncMappings, falling back to GetMappingUpdate")
-				streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone)
+				streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone, onConnected)
 			}
 		} else {
-			streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone)
+			streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone, onConnected)
 		}
 
 		if s.healthChecker != nil {
 			s.healthChecker.SetManagementConnected(false)
 		}
 
-		// Stream established — reset backoff so the next failure retries quickly.
-		bo.Reset()
+		// Reset backoff only when a stream actually connected, so immediate
+		// connect failures still back off instead of spinning.
+		if connected {
+			bo.Reset()
+		}
 
 		if streamErr == nil {
 			return fmt.Errorf("stream closed by server")
@@ -1221,7 +1234,7 @@ func (s *Server) proxyCapabilities() *proto.ProxyCapabilities {
 	}
 }
 
-func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool) error {
+func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool, onConnected func()) error {
 	connectTime := time.Now()
 	mappingClient, err := client.GetMappingUpdate(ctx, &proto.GetMappingUpdateRequest{
 		ProxyId:      s.ID,
@@ -1234,6 +1247,7 @@ func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServ
 		return fmt.Errorf("create mapping stream: %w", err)
 	}
 
+	onConnected()
 	if s.healthChecker != nil {
 		s.healthChecker.SetManagementConnected(true)
 	}
@@ -1242,7 +1256,7 @@ func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServ
 	return s.handleMappingStream(ctx, mappingClient, initialSyncDone, connectTime)
 }
 
-func (s *Server) trySyncMappings(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool) error {
+func (s *Server) trySyncMappings(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool, onConnected func()) error {
 	connectTime := time.Now()
 	stream, err := client.SyncMappings(ctx)
 	if err != nil {
@@ -1263,6 +1277,7 @@ func (s *Server) trySyncMappings(ctx context.Context, client proto.ProxyServiceC
 		return fmt.Errorf("send sync init: %w", err)
 	}
 
+	onConnected()
 	if s.healthChecker != nil {
 		s.healthChecker.SetManagementConnected(true)
 	}
@@ -1307,7 +1322,9 @@ func (s *Server) handleSyncMappingsStream(ctx context.Context, stream proto.Prox
 
 			batchStart := time.Now()
 			s.Logger.Debug("Received mapping update, starting processing")
-			s.processMappings(ctx, msg.GetMapping())
+			if err := s.processMappingsGuarded(ctx, msg.GetMapping()); err != nil {
+				return err
+			}
 			s.Logger.Debug("Processing mapping update completed")
 			tracker.recordBatch(ctx, s, msg.GetMapping(), msg.GetInitialSyncComplete(), batchStart)
 
@@ -1391,7 +1408,9 @@ func (s *Server) handleMappingStream(ctx context.Context, mappingClient proto.Pr
 
 			batchStart := time.Now()
 			s.Logger.Debug("Received mapping update, starting processing")
-			s.processMappings(ctx, msg.GetMapping())
+			if err := s.processMappingsGuarded(ctx, msg.GetMapping()); err != nil {
+				return err
+			}
 			s.Logger.Debug("Processing mapping update completed")
 			tracker.recordBatch(ctx, s, msg.GetMapping(), msg.GetInitialSyncComplete(), batchStart)
 		}
@@ -1456,6 +1475,44 @@ func redactMappingForLog(m *proto.ProxyMapping) *proto.ProxyMapping {
 	return c
 }
 
+const defaultMappingBatchWatchdog = 2 * time.Minute
+
+// mappingBatchWatchdog returns the configured batch watchdog or the default.
+func (s *Server) mappingBatchWatchdog() time.Duration {
+	if s.MappingBatchWatchdog > 0 {
+		return s.MappingBatchWatchdog
+	}
+	return defaultMappingBatchWatchdog
+}
+
+// processMappingsGuarded applies a batch under a watchdog, returning an error
+// if processing exceeds the watchdog so the caller reconnects and resyncs
+// instead of wedging silently.
+func (s *Server) processMappingsGuarded(ctx context.Context, mappings []*proto.ProxyMapping) error {
+	batchCtx, cancel := context.WithCancel(ctx)
+	defer cancel()
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		s.processMappings(batchCtx, mappings)
+	}()
+
+	watchdog := s.mappingBatchWatchdog()
+	timer := time.NewTimer(watchdog)
+	defer timer.Stop()
+
+	select {
+	case <-done:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		s.Logger.Errorf("processing mapping batch exceeded %s, cancelling and reconnecting to resync", watchdog)
+		return fmt.Errorf("mapping batch processing stalled after %s", watchdog)
+	}
+}
+
 func (s *Server) processMappings(ctx context.Context, mappings []*proto.ProxyMapping) {
 	debug := s.Logger != nil && s.Logger.IsLevelEnabled(log.DebugLevel)
 	for _, mapping := range mappings {
@@ -1951,7 +2008,11 @@ func (s *Server) updateMapping(ctx context.Context, mapping *proto.ProxyMapping)
 func (s *Server) removeMapping(ctx context.Context, mapping *proto.ProxyMapping) {
 	accountID := types.AccountID(mapping.GetAccountId())
 	svcKey := s.serviceKeyForMapping(mapping)
-	if err := s.netbird.RemovePeer(ctx, accountID, svcKey); err != nil {
+	removePeer := s.removePeer
+	if removePeer == nil {
+		removePeer = s.netbird.RemovePeer
+	}
+	if err := removePeer(ctx, accountID, svcKey); err != nil {
 		s.Logger.WithFields(log.Fields{
 			"account_id": accountID,
 			"service_id": mapping.GetId(),

shared/context/keys.go

@@ -6,4 +6,5 @@ const (
 	RoleKey      = "role"
 	UserIDKey    = "userID"
 	PeerIDKey    = "peerID"
+	UserAgentKey = "userAgent"
 )

shared/management/http/api/openapi.yml

@@ -5107,31 +5107,63 @@ components:
   responses:
     not_found:
       description: Resource not found
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     validation_failed_simple:
       description: Validation failed
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     bad_request:
       description: Bad Request
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     internal_error:
       description: Internal Server Error
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     validation_failed:
       description: Validation failed
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     forbidden:
       description: Forbidden
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     requires_authentication:
       description: Requires authentication
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content: { }
     conflict:
       description: Conflict
+      headers:
+        X-Request-Id:
+          $ref: '#/components/headers/X-Request-Id'
       content:
         application/json:
           schema:
             $ref: '#/components/schemas/ErrorResponse'
+  headers:
+    X-Request-Id:
+      description: |
+        Unique identifier assigned to the request by the server and set on every
+        response. Useful for correlating client requests with server-side logs.
+      schema:
+        type: string
+        example: cot7r4n3l3vh3qj4qveg
   securitySchemes:
     BearerAuth:
       type: http

shared/relay/client/client.go

@@ -9,12 +9,14 @@ import (
 	"net/url"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	log "github.com/sirupsen/logrus"
 
 	auth "github.com/netbirdio/netbird/shared/relay/auth/hmac"
 	"github.com/netbirdio/netbird/shared/relay/client/dialer"
+	netErr "github.com/netbirdio/netbird/shared/relay/client/dialer/net"
 	"github.com/netbirdio/netbird/shared/relay/healthcheck"
 	"github.com/netbirdio/netbird/shared/relay/messages"
 )
@@ -172,6 +174,19 @@ type Client struct {
 	stateSubscription *PeersStateSubscription
 
 	mtu uint16
+
+	// transportFallback, when set, records datagram-too-large failures so a
+	// datagram-sized transport is avoided on subsequent connects. Shared via
+	// the manager.
+	transportFallback *transportFallback
+	// datagramFallbackTriggered guards a single fallback per connection so a
+	// burst of oversized datagrams triggers one reconnect, not many.
+	datagramFallbackTriggered atomic.Bool
+}
+
+// SetTransportFallback wires the shared datagram-transport fallback tracker.
+func (c *Client) SetTransportFallback(tf *transportFallback) {
+	c.transportFallback = tf
 }
 
 // NewClient creates a new client for the relay server. The client is not connected to the server until the Connect
@@ -361,12 +376,13 @@ func (c *Client) Close() error {
 }
 
 func (c *Client) connect(ctx context.Context) (*RelayAddr, error) {
-	dialers := c.getDialers()
+	mode := transportModeFromEnv()
+	dialers := c.getDialers(mode)
 
 	var conn net.Conn
 	if c.serverIP.IsValid() {
 		var err error
-		conn, err = c.dialRaceDirect(ctx, dialers)
+		conn, err = c.dialRaceDirect(ctx, mode, dialers)
 		if err != nil {
 			c.log.Infof("dial via server IP %s failed, falling back to FQDN: %v", c.serverIP, err)
 			conn = nil
@@ -375,6 +391,9 @@ func (c *Client) connect(ctx context.Context) (*RelayAddr, error) {
 
 	if conn == nil {
 		rd := dialer.NewRaceDial(c.log, dialer.DefaultConnectionTimeout, c.connectionURL, dialers...)
+		if mode.sequential() {
+			rd.WithSequential()
+		}
 		var err error
 		conn, err = rd.Dial(ctx)
 		if err != nil {
@@ -382,6 +401,7 @@ func (c *Client) connect(ctx context.Context) (*RelayAddr, error) {
 		}
 	}
 	c.relayConn = conn
+	c.datagramFallbackTriggered.Store(false)
 
 	instanceURL, err := c.handShake(ctx)
 	if err != nil {
@@ -396,7 +416,7 @@ func (c *Client) connect(ctx context.Context) (*RelayAddr, error) {
 }
 
 // dialRaceDirect dials c.serverIP, preserving the original FQDN as the TLS ServerName for SNI.
-func (c *Client) dialRaceDirect(ctx context.Context, dialers []dialer.DialeFn) (net.Conn, error) {
+func (c *Client) dialRaceDirect(ctx context.Context, mode TransportMode, dialers []dialer.DialeFn) (net.Conn, error) {
 	directURL, serverName, err := substituteHost(c.connectionURL, c.serverIP)
 	if err != nil {
 		return nil, fmt.Errorf("substitute host: %w", err)
@@ -406,6 +426,9 @@ func (c *Client) dialRaceDirect(ctx context.Context, dialers []dialer.DialeFn) (
 
 	rd := dialer.NewRaceDial(c.log, dialer.DefaultConnectionTimeout, directURL, dialers...).
 		WithServerName(serverName)
+	if mode.sequential() {
+		rd.WithSequential()
+	}
 	return rd.Dial(ctx)
 }
 
@@ -631,13 +654,53 @@ func (c *Client) writeTo(containerRef *connContainer, dstID messages.PeerID, pay
 	}
 
 	// the write always return with 0 length because the underling does not support the size feedback.
-	_, err = c.relayConn.Write(msg)
+	conn := c.relayConn
+	_, err = conn.Write(msg)
 	if err != nil {
-		c.log.Errorf("failed to write transport message: %s", err)
+		if errors.Is(err, netErr.ErrDatagramTooLarge) {
+			c.onDatagramTooLarge(conn, err)
+		} else {
+			c.log.Errorf("failed to write transport message: %s", err)
+		}
 	}
 	return len(payload), err
 }
 
+// onDatagramTooLarge reacts to a datagram rejected as too large for the path.
+// When a non-datagram transport is available, it records a fallback for this
+// server and closes the connection so the reconnect avoids datagram-sized
+// transports. A single fallback is triggered per connection regardless of how
+// many oversized datagrams arrive. cause carries the datagram size and budget.
+func (c *Client) onDatagramTooLarge(conn net.Conn, cause error) {
+	// Handle one oversized datagram per connection; a burst triggers a single
+	// fallback (and a single log line), not many.
+	if !c.datagramFallbackTriggered.CompareAndSwap(false, true) {
+		return
+	}
+
+	// If the selected mode offers no non-datagram transport (e.g. pinned to a
+	// datagram-sized transport), reconnecting would just re-fail, so leave the
+	// connection up rather than loop.
+	if len(nonDatagramSized(c.baseDialers(transportModeFromEnv()))) == 0 {
+		c.log.Warnf("%s, but no non-datagram transport is available, not falling back", cause)
+		return
+	}
+
+	// Without the shared tracker a reconnect would just select the same
+	// transport again and re-fail, so leave the connection up rather than loop.
+	if c.transportFallback == nil {
+		c.log.Debugf("%s, but no transport fallback configured, leaving connection up", cause)
+		return
+	}
+
+	window := c.transportFallback.recordFailure(c.connectionURL)
+	c.log.Warnf("%s, avoiding datagram-sized transport for %s", cause, window)
+
+	if err := conn.Close(); err != nil {
+		c.log.Debugf("close relay connection for transport fallback: %s", err)
+	}
+}
+
 func (c *Client) listenForStopEvents(ctx context.Context, hc *healthcheck.Receiver, conn net.Conn, internalStopFlag *internalStopFlag) {
 	for {
 		select {

shared/relay/client/dialer/capability.go

@@ -0,0 +1,18 @@
+package dialer
+
+// DatagramSized is implemented by dialers whose connections carry each write in
+// a single datagram, so a write can be rejected when it exceeds the path's
+// datagram budget (e.g. QUIC). Transports without this capability (e.g.
+// WebSocket over TCP) impose no per-write size limit, so the relay client can
+// fall back to them when a datagram-sized transport rejects a write as too
+// large. The capability is advertised per dialer rather than hardcoded, so a
+// new transport only needs to declare whether it is datagram-sized.
+type DatagramSized interface {
+	DatagramSized()
+}
+
+// IsDatagramSized reports whether d produces datagram-sized connections.
+func IsDatagramSized(d DialeFn) bool {
+	_, ok := d.(DatagramSized)
+	return ok
+}

shared/relay/client/dialer/net/err.go

@@ -4,4 +4,9 @@ import "errors"
 
 var (
 	ErrClosedByServer = errors.New("closed by server")
+
+	// ErrDatagramTooLarge is returned when a transport message exceeds the
+	// QUIC datagram size the path to the relay can carry. The relay client
+	// treats it as a signal to fall back to a non-datagram transport.
+	ErrDatagramTooLarge = errors.New("datagram frame too large")
 )

shared/relay/client/dialer/quic/conn.go

@@ -8,7 +8,6 @@ import (
 	"time"
 
 	"github.com/quic-go/quic-go"
-	log "github.com/sirupsen/logrus"
 
 	netErr "github.com/netbirdio/netbird/shared/relay/client/dialer/net"
 )
@@ -52,11 +51,8 @@ func (c *Conn) Read(b []byte) (n int, err error) {
 }
 
 func (c *Conn) Write(b []byte) (int, error) {
-	err := c.session.SendDatagram(b)
-	if err != nil {
-		err = c.remoteCloseErrHandling(err)
-		log.Errorf("failed to write to QUIC stream: %v", err)
-		return 0, err
+	if err := c.session.SendDatagram(b); err != nil {
+		return 0, c.writeErrHandling(err, len(b))
 	}
 	return len(b), nil
 }
@@ -95,3 +91,15 @@ func (c *Conn) remoteCloseErrHandling(err error) error {
 	}
 	return err
 }
+
+// writeErrHandling normalizes SendDatagram errors. A datagram that exceeds the
+// path's QUIC packet budget is mapped to ErrDatagramTooLarge (annotated with the
+// datagram size and path budget) so the relay client can fall back to a
+// non-datagram transport.
+func (c *Conn) writeErrHandling(err error, size int) error {
+	var tooLarge *quic.DatagramTooLargeError
+	if errors.As(err, &tooLarge) {
+		return fmt.Errorf("%w: %d byte datagram over path budget %d", netErr.ErrDatagramTooLarge, size, tooLarge.MaxDatagramPayloadSize)
+	}
+	return c.remoteCloseErrHandling(err)
+}

shared/relay/client/dialer/quic/quic.go

@@ -9,6 +9,7 @@ import (
 	"time"
 
 	"github.com/quic-go/quic-go"
+	"github.com/quic-go/quic-go/logging"
 	log "github.com/sirupsen/logrus"
 
 	nbnet "github.com/netbirdio/netbird/client/net"
@@ -23,6 +24,12 @@ func (d Dialer) Protocol() string {
 	return Network
 }
 
+// DatagramSized marks QUIC as a datagram-sized transport: relay traffic is
+// carried in QUIC DATAGRAM frames, which must fit a single packet.
+func (d Dialer) DatagramSized() {
+	// Intentional marker method; presence is the capability signal.
+}
+
 func (d Dialer) Dial(ctx context.Context, address, serverName string) (net.Conn, error) {
 	quicURL, err := prepareURL(address)
 	if err != nil {
@@ -47,6 +54,7 @@ func (d Dialer) Dial(ctx context.Context, address, serverName string) (net.Conn,
 		MaxIdleTimeout:    4 * time.Minute,
 		EnableDatagrams:   true,
 		InitialPacketSize: nbRelay.QUICInitialPacketSize,
+		Tracer:            connectionTracer(quicURL),
 	}
 
 	udpConn, err := nbnet.ListenUDP("udp", &net.UDPAddr{Port: 0})
@@ -74,6 +82,28 @@ func (d Dialer) Dial(ctx context.Context, address, serverName string) (net.Conn,
 	return conn, nil
 }
 
+// connectionTracer returns a QUIC tracer that logs the DPLPMTUD result and the
+// reason a relay connection closed, so the path MTU settled on and teardown
+// cause are visible in logs. Lines carry the relay address as a structured
+// field, matching the rest of the relay client logging.
+func connectionTracer(addr string) func(context.Context, logging.Perspective, quic.ConnectionID) *logging.ConnectionTracer {
+	relayLog := log.WithField("relay", addr)
+	return func(context.Context, logging.Perspective, quic.ConnectionID) *logging.ConnectionTracer {
+		return &logging.ConnectionTracer{
+			UpdatedMTU: func(mtu logging.ByteCount, done bool) {
+				if done {
+					relayLog.Infof("QUIC path MTU settled at %d", mtu)
+					return
+				}
+				relayLog.Debugf("QUIC path MTU probing at %d", mtu)
+			},
+			ClosedConnection: func(err error) {
+				relayLog.Debugf("QUIC connection closed: %v", err)
+			},
+		}
+	}
+}
+
 func prepareURL(address string) (string, error) {
 	var host string
 	var defaultPort string

shared/relay/client/dialer/race_dialer.go

@@ -32,6 +32,7 @@ type RaceDial struct {
 	serverName        string
 	dialerFns         []DialeFn
 	connectionTimeout time.Duration
+	sequential        bool
 }
 
 func NewRaceDial(log *log.Entry, connectionTimeout time.Duration, serverURL string, dialerFns ...DialeFn) *RaceDial {
@@ -53,7 +54,21 @@ func (r *RaceDial) WithServerName(serverName string) *RaceDial {
 	return r
 }
 
+// WithSequential makes Dial try the dialers in order, falling back to the next
+// only when one fails to connect, instead of racing them concurrently.
+//
+// Mutates the receiver and is not safe for concurrent reconfiguration; a
+// RaceDial is intended to be constructed per dial and discarded.
+func (r *RaceDial) WithSequential() *RaceDial {
+	r.sequential = true
+	return r
+}
+
 func (r *RaceDial) Dial(ctx context.Context) (net.Conn, error) {
+	if r.sequential {
+		return r.dialSequential(ctx)
+	}
+
 	connChan := make(chan dialResult, len(r.dialerFns))
 	winnerConn := make(chan net.Conn, 1)
 	abortCtx, abort := context.WithCancel(ctx)
@@ -72,6 +87,30 @@ func (r *RaceDial) Dial(ctx context.Context) (net.Conn, error) {
 	return conn, nil
 }
 
+// dialSequential tries each dialer in order, returning the first connection and
+// falling back to the next on failure.
+func (r *RaceDial) dialSequential(ctx context.Context) (net.Conn, error) {
+	for _, dfn := range r.dialerFns {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		attemptCtx, cancel := context.WithTimeout(ctx, r.connectionTimeout)
+		r.log.Infof("dialing Relay server via %s", dfn.Protocol())
+		conn, err := dfn.Dial(attemptCtx, r.serverURL, r.serverName)
+		cancel()
+		if err != nil {
+			if errors.Is(err, context.Canceled) {
+				return nil, err
+			}
+			r.log.Errorf("failed to dial via %s: %s", dfn.Protocol(), err)
+			continue
+		}
+		r.log.Infof("successfully dialed via: %s", dfn.Protocol())
+		return conn, nil
+	}
+	return nil, errors.New("failed to dial to Relay server on any protocol")
+}
+
 func (r *RaceDial) dial(dfn DialeFn, abortCtx context.Context, connChan chan dialResult) {
 	ctx, cancel := context.WithTimeout(abortCtx, r.connectionTimeout)
 	defer cancel()

shared/relay/client/dialer/race_dialer_test.go

@@ -250,3 +250,66 @@ func TestRaceDialFirstSuccessfulDialerWins(t *testing.T) {
 		}
 	}
 }
+
+func TestRaceDialSequentialFallback(t *testing.T) {
+	logger := logrus.NewEntry(logrus.New())
+	serverURL := "test.server.com"
+
+	var firstDialed, secondDialed bool
+	preferred := &MockDialer{
+		protocolStr: "quic",
+		dialFunc: func(ctx context.Context, address string) (net.Conn, error) {
+			firstDialed = true
+			return nil, errors.New("quic unreachable")
+		},
+	}
+	fallbackConn := &MockConn{remoteAddr: &MockAddr{network: "ws"}}
+	fallback := &MockDialer{
+		protocolStr: "ws",
+		dialFunc: func(ctx context.Context, address string) (net.Conn, error) {
+			secondDialed = true
+			return fallbackConn, nil
+		},
+	}
+
+	rd := NewRaceDial(logger, DefaultConnectionTimeout, serverURL, preferred, fallback).WithSequential()
+	conn, err := rd.Dial(context.Background())
+	if err != nil {
+		t.Fatalf("expected fallback to succeed, got %v", err)
+	}
+	if conn != fallbackConn {
+		t.Errorf("expected fallback connection, got %v", conn)
+	}
+	if !firstDialed || !secondDialed {
+		t.Errorf("expected both dialers attempted in order, first=%v second=%v", firstDialed, secondDialed)
+	}
+}
+
+func TestRaceDialSequentialPreferredWins(t *testing.T) {
+	logger := logrus.NewEntry(logrus.New())
+	serverURL := "test.server.com"
+
+	preferredConn := &MockConn{remoteAddr: &MockAddr{network: "quic"}}
+	preferred := &MockDialer{
+		protocolStr: "quic",
+		dialFunc: func(ctx context.Context, address string) (net.Conn, error) {
+			return preferredConn, nil
+		},
+	}
+	fallback := &MockDialer{
+		protocolStr: "ws",
+		dialFunc: func(ctx context.Context, address string) (net.Conn, error) {
+			t.Errorf("fallback dialer must not be tried when preferred succeeds")
+			return nil, errors.New("should not happen")
+		},
+	}
+
+	rd := NewRaceDial(logger, DefaultConnectionTimeout, serverURL, preferred, fallback).WithSequential()
+	conn, err := rd.Dial(context.Background())
+	if err != nil {
+		t.Fatalf("expected preferred to succeed, got %v", err)
+	}
+	if conn != preferredConn {
+		t.Errorf("expected preferred connection, got %v", conn)
+	}
+}

shared/relay/client/dialers_generic.go

@@ -9,11 +9,42 @@ import (
 	"github.com/netbirdio/netbird/shared/relay/client/dialer/ws"
 )
 
-// getDialers returns the list of dialers to use for connecting to the relay server.
-func (c *Client) getDialers() []dialer.DialeFn {
-	if c.mtu > 0 && c.mtu > iface.DefaultMTU {
-		c.log.Infof("MTU %d exceeds default (%d), forcing WebSocket transport to avoid DATAGRAM frame size issues", c.mtu, iface.DefaultMTU)
-		return []dialer.DialeFn{ws.Dialer{}}
+// getDialers returns the ordered dialers for connecting to the relay server. It
+// applies the datagram fallback generically: if this server recently rejected a
+// datagram-sized transport, those dialers are dropped, leaving the rest.
+func (c *Client) getDialers(mode TransportMode) []dialer.DialeFn {
+	dialers := c.baseDialers(mode)
+
+	if c.transportFallback != nil && c.transportFallback.avoidDatagramSized(c.connectionURL) {
+		if filtered := nonDatagramSized(dialers); len(filtered) > 0 {
+			c.log.Infof("relay recently rejected a datagram-sized transport, avoiding it")
+			return filtered
+		}
 	}
-	return []dialer.DialeFn{quic.Dialer{}, ws.Dialer{}}
+	return dialers
+}
+
+// baseDialers returns the ordered dialers for the mode, before any datagram
+// fallback filtering. For racing modes (auto) the order is irrelevant; for
+// prefer modes the first entry is tried before falling back to the second.
+func (c *Client) baseDialers(mode TransportMode) []dialer.DialeFn {
+	switch mode {
+	case TransportModeWS:
+		c.log.Infof("%s=ws, using WebSocket transport", EnvRelayTransport)
+		return []dialer.DialeFn{ws.Dialer{}}
+	case TransportModeQUIC:
+		c.log.Infof("%s=quic, using QUIC transport", EnvRelayTransport)
+		return []dialer.DialeFn{quic.Dialer{}}
+	}
+
+	all := []dialer.DialeFn{quic.Dialer{}, ws.Dialer{}}
+	if mode == TransportModePreferWS {
+		all = []dialer.DialeFn{ws.Dialer{}, quic.Dialer{}}
+	}
+
+	if c.mtu > 0 && c.mtu > iface.DefaultMTU {
+		c.log.Infof("MTU %d exceeds default (%d), avoiding datagram-sized transports", c.mtu, iface.DefaultMTU)
+		return nonDatagramSized(all)
+	}
+	return all
 }

shared/relay/client/dialers_generic_test.go

@@ -0,0 +1,101 @@
+//go:build !js
+
+package client
+
+import (
+	"os"
+	"testing"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
+
+	"github.com/netbirdio/netbird/client/iface"
+	"github.com/netbirdio/netbird/shared/relay/client/dialer"
+	netErr "github.com/netbirdio/netbird/shared/relay/client/dialer/net"
+	"github.com/netbirdio/netbird/shared/relay/client/dialer/quic"
+	"github.com/netbirdio/netbird/shared/relay/client/dialer/ws"
+)
+
+// TestDatagramSizedCapability locks the capability the generic fallback relies
+// on: QUIC is datagram-sized, WebSocket is not.
+func TestDatagramSizedCapability(t *testing.T) {
+	assert.True(t, dialer.IsDatagramSized(quic.Dialer{}), "QUIC must advertise datagram-sized")
+	assert.False(t, dialer.IsDatagramSized(ws.Dialer{}), "WebSocket must not advertise datagram-sized")
+}
+
+func protocols(dialers []dialer.DialeFn) []string {
+	out := make([]string, len(dialers))
+	for i, d := range dialers {
+		out[i] = d.Protocol()
+	}
+	return out
+}
+
+func TestGetDialers(t *testing.T) {
+	const url = "rels://relay.example:443"
+
+	tests := []struct {
+		name     string
+		mode     string
+		mtu      uint16
+		preferWS bool
+		want     []string
+	}{
+		{name: "auto races quic and ws", mode: "auto", mtu: iface.DefaultMTU, want: []string{"quic", "WS"}},
+		{name: "ws pinned", mode: "ws", mtu: iface.DefaultMTU, want: []string{"WS"}},
+		{name: "quic pinned", mode: "quic", mtu: iface.DefaultMTU, want: []string{"quic"}},
+		{name: "prefer-quic orders quic first", mode: "prefer-quic", mtu: iface.DefaultMTU, want: []string{"quic", "WS"}},
+		{name: "prefer-ws orders ws first", mode: "prefer-ws", mtu: iface.DefaultMTU, want: []string{"WS", "quic"}},
+		{name: "mtu above default forces ws", mode: "auto", mtu: iface.DefaultMTU + 100, want: []string{"WS"}},
+		{name: "sticky fallback forces ws in auto", mode: "auto", mtu: iface.DefaultMTU, preferWS: true, want: []string{"WS"}},
+		{name: "sticky fallback forces ws in prefer-quic", mode: "prefer-quic", mtu: iface.DefaultMTU, preferWS: true, want: []string{"WS"}},
+		{name: "quic pin overrides sticky fallback", mode: "quic", mtu: iface.DefaultMTU, preferWS: true, want: []string{"quic"}},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv(EnvRelayTransport, tc.mode)
+			if tc.mode == "" {
+				os.Unsetenv(EnvRelayTransport)
+			}
+
+			tf := newTransportFallback()
+			if tc.preferWS {
+				tf.recordFailure(url)
+			}
+
+			c := &Client{
+				log:               log.WithField("test", t.Name()),
+				connectionURL:     url,
+				mtu:               tc.mtu,
+				transportFallback: tf,
+			}
+
+			assert.Equal(t, tc.want, protocols(c.getDialers(transportModeFromEnv())))
+		})
+	}
+}
+
+// TestStickyFallbackAfterDatagramTooLarge verifies the full chain: an oversized
+// datagram records a fallback that makes the next dial pick WebSocket, the way a
+// reconnect would after the connection is closed.
+func TestStickyFallbackAfterDatagramTooLarge(t *testing.T) {
+	const url = "rels://relay.example:443"
+	t.Setenv(EnvRelayTransport, string(TransportModeAuto))
+
+	c := &Client{
+		log:               log.WithField("test", t.Name()),
+		connectionURL:     url,
+		mtu:               iface.DefaultMTU,
+		transportFallback: newTransportFallback(),
+	}
+
+	// First dial races both transports.
+	assert.Equal(t, []string{"quic", "WS"}, protocols(c.getDialers(transportModeFromEnv())))
+
+	// An oversized datagram records the fallback for this server.
+	c.onDatagramTooLarge(&closeTrackingConn{}, netErr.ErrDatagramTooLarge)
+
+	// The reconnect now sticks to WebSocket.
+	assert.Equal(t, []string{"WS"}, protocols(c.getDialers(transportModeFromEnv())))
+}

shared/relay/client/dialers_js.go

@@ -7,7 +7,11 @@ import (
 	"github.com/netbirdio/netbird/shared/relay/client/dialer/ws"
 )
 
-func (c *Client) getDialers() []dialer.DialeFn {
+func (c *Client) getDialers(_ TransportMode) []dialer.DialeFn {
 	// JS/WASM build only uses WebSocket transport
 	return []dialer.DialeFn{ws.Dialer{}}
 }
+
+func (c *Client) baseDialers(_ TransportMode) []dialer.DialeFn {
+	return []dialer.DialeFn{ws.Dialer{}}
+}

shared/relay/client/manager.go

@@ -79,23 +79,30 @@ type Manager struct {
 
 	cleanupInterval      time.Duration
 	keepUnusedServerTime time.Duration
+
+	// transportFallback is shared across home and foreign relay clients so a
+	// datagram-too-large failure makes that server avoid datagram-sized transports across reconnects.
+	transportFallback *transportFallback
 }
 
 // NewManager creates a new manager instance.
 // The serverURL address can be empty. In this case, the manager will not serve.
 func NewManager(ctx context.Context, serverURLs []string, peerID string, mtu uint16, opts ...ManagerOption) *Manager {
 	tokenStore := &relayAuth.TokenStore{}
+	tf := newTransportFallback()
 
 	m := &Manager{
-		ctx:        ctx,
-		peerID:     peerID,
-		tokenStore: tokenStore,
-		mtu:        mtu,
+		ctx:               ctx,
+		peerID:            peerID,
+		tokenStore:        tokenStore,
+		mtu:               mtu,
+		transportFallback: tf,
 		serverPicker: &ServerPicker{
 			TokenStore:        tokenStore,
 			PeerID:            peerID,
 			MTU:               mtu,
 			ConnectionTimeout: defaultConnectionTimeout,
+			TransportFallback: tf,
 		},
 		relayClients:            make(map[string]*RelayTrack),
 		onDisconnectedListeners: make(map[string]*list.List),
@@ -287,6 +294,7 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	m.relayClientsMutex.Unlock()
 
 	relayClient := NewClientWithServerIP(serverAddress, serverIP, m.tokenStore, m.peerID, m.mtu)
+	relayClient.SetTransportFallback(m.transportFallback)
 	err := relayClient.Connect(m.ctx)
 	if err != nil {
 		rt.err = err

shared/relay/client/picker.go

@@ -29,6 +29,7 @@ type ServerPicker struct {
 	PeerID            string
 	MTU               uint16
 	ConnectionTimeout time.Duration
+	TransportFallback *transportFallback
 }
 
 func (sp *ServerPicker) PickServer(parentCtx context.Context) (*Client, error) {
@@ -70,6 +71,7 @@ func (sp *ServerPicker) PickServer(parentCtx context.Context) (*Client, error) {
 func (sp *ServerPicker) startConnection(ctx context.Context, resultChan chan connResult, url string) {
 	log.Infof("try to connecting to relay server: %s", url)
 	relayClient := NewClient(url, sp.TokenStore, sp.PeerID, sp.MTU)
+	relayClient.SetTransportFallback(sp.TransportFallback)
 	err := relayClient.Connect(ctx)
 	resultChan <- connResult{
 		RelayClient: relayClient,

shared/relay/client/transport.go

@@ -0,0 +1,129 @@
+package client
+
+import (
+	"os"
+	"strings"
+	"sync"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/relay/client/dialer"
+)
+
+// EnvRelayTransport pins the relay transport. Valid values: "auto" (default,
+// race QUIC and WebSocket), "quic" (QUIC only), "ws" (WebSocket only),
+// "prefer-quic" / "prefer-ws" (try the preferred transport first, fall back to
+// the other only if it fails to connect; no race). The prefer modes trade a
+// slower connect when the preferred transport is blackholed for deterministic
+// transport selection.
+const EnvRelayTransport = "NB_RELAY_TRANSPORT"
+
+const (
+	// transportFallbackBase is the initial window a relay server avoids
+	// datagram-sized transports after a datagram is rejected as too large.
+	transportFallbackBase = 10 * time.Minute
+	// transportFallbackMax caps the pinned window when failures repeat.
+	transportFallbackMax = 60 * time.Minute
+)
+
+// TransportMode selects which relay dialers are used.
+type TransportMode string
+
+const (
+	TransportModeAuto       TransportMode = "auto"
+	TransportModeQUIC       TransportMode = "quic"
+	TransportModeWS         TransportMode = "ws"
+	TransportModePreferQUIC TransportMode = "prefer-quic"
+	TransportModePreferWS   TransportMode = "prefer-ws"
+)
+
+// transportModeFromEnv reads EnvRelayTransport, defaulting to auto for an empty
+// or unrecognized value.
+func transportModeFromEnv() TransportMode {
+	switch TransportMode(strings.ToLower(strings.TrimSpace(os.Getenv(EnvRelayTransport)))) {
+	case "", TransportModeAuto:
+		return TransportModeAuto
+	case TransportModeQUIC:
+		return TransportModeQUIC
+	case TransportModeWS:
+		return TransportModeWS
+	case TransportModePreferQUIC:
+		return TransportModePreferQUIC
+	case TransportModePreferWS:
+		return TransportModePreferWS
+	default:
+		log.Warnf("invalid %s value %q, using %q", EnvRelayTransport, os.Getenv(EnvRelayTransport), TransportModeAuto)
+		return TransportModeAuto
+	}
+}
+
+// sequential reports whether the mode tries dialers in order with fallback
+// instead of racing them concurrently.
+func (m TransportMode) sequential() bool {
+	return m == TransportModePreferQUIC || m == TransportModePreferWS
+}
+
+// transportFallback tracks relay servers that have rejected a datagram-sized
+// transport (a write too large for the path) and should temporarily avoid such
+// transports. It is shared across the relay manager so the preference survives
+// client recreation (foreign relay clients are evicted and rebuilt on
+// disconnect). Entries are keyed by server URL and expire after a window that
+// grows on repeated failures.
+type transportFallback struct {
+	mu      sync.Mutex
+	entries map[string]*fallbackEntry
+}
+
+type fallbackEntry struct {
+	until    time.Time
+	duration time.Duration
+}
+
+func newTransportFallback() *transportFallback {
+	return &transportFallback{entries: make(map[string]*fallbackEntry)}
+}
+
+// avoidDatagramSized reports whether serverURL is currently within a window
+// where datagram-sized transports should be avoided.
+func (f *transportFallback) avoidDatagramSized(serverURL string) bool {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	e := f.entries[serverURL]
+	return e != nil && time.Now().Before(e.until)
+}
+
+// recordFailure makes serverURL avoid datagram-sized transports for a window:
+// transportFallbackBase on the first failure, doubling up to transportFallbackMax
+// when a datagram transport fails again after a previous window expired. It
+// returns the active window duration.
+func (f *transportFallback) recordFailure(serverURL string) time.Duration {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	now := time.Now()
+	e := f.entries[serverURL]
+	switch {
+	case e == nil:
+		e = &fallbackEntry{duration: transportFallbackBase}
+		f.entries[serverURL] = e
+	case now.Before(e.until):
+		return time.Until(e.until)
+	default:
+		e.duration = min(e.duration*2, transportFallbackMax)
+	}
+	e.until = now.Add(e.duration)
+	return e.duration
+}
+
+// nonDatagramSized returns the dialers from in that are not datagram-sized,
+// preserving order.
+func nonDatagramSized(in []dialer.DialeFn) []dialer.DialeFn {
+	out := make([]dialer.DialeFn, 0, len(in))
+	for _, d := range in {
+		if !dialer.IsDatagramSized(d) {
+			out = append(out, d)
+		}
+	}
+	return out
+}

shared/relay/client/transport_test.go

@@ -0,0 +1,140 @@
+package client
+
+import (
+	"net"
+	"os"
+	"testing"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	netErr "github.com/netbirdio/netbird/shared/relay/client/dialer/net"
+)
+
+// closeTrackingConn records whether Close was called; only Close is exercised.
+type closeTrackingConn struct {
+	net.Conn
+	closed bool
+}
+
+func (c *closeTrackingConn) Close() error {
+	c.closed = true
+	return nil
+}
+
+func TestTransportModeFromEnv(t *testing.T) {
+	tests := []struct {
+		value string
+		want  TransportMode
+	}{
+		{"", TransportModeAuto},
+		{"auto", TransportModeAuto},
+		{"quic", TransportModeQUIC},
+		{"QUIC", TransportModeQUIC},
+		{"ws", TransportModeWS},
+		{" Ws ", TransportModeWS},
+		{"prefer-quic", TransportModePreferQUIC},
+		{"prefer-ws", TransportModePreferWS},
+		{"garbage", TransportModeAuto},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.value, func(t *testing.T) {
+			t.Setenv(EnvRelayTransport, tc.value)
+			if tc.value == "" {
+				os.Unsetenv(EnvRelayTransport)
+			}
+			assert.Equal(t, tc.want, transportModeFromEnv())
+		})
+	}
+}
+
+func TestTransportFallbackRecordAndExpiry(t *testing.T) {
+	const url = "rels://relay.example:443"
+	f := newTransportFallback()
+
+	assert.False(t, f.avoidDatagramSized(url), "no fallback recorded yet")
+
+	d := f.recordFailure(url)
+	assert.Equal(t, transportFallbackBase, d, "first failure pins for the base window")
+	assert.True(t, f.avoidDatagramSized(url), "datagram-sized transport avoided within the window")
+
+	// A second failure while still inside the window must not grow the window.
+	d = f.recordFailure(url)
+	assert.LessOrEqual(t, d, transportFallbackBase, "still within the active window")
+	require.NotNil(t, f.entries[url])
+	assert.Equal(t, transportFallbackBase, f.entries[url].duration, "duration unchanged inside window")
+
+	// Expire the window: datagram-sized transport allowed again.
+	f.entries[url].until = time.Now().Add(-time.Second)
+	assert.False(t, f.avoidDatagramSized(url), "window expired, datagram-sized transport allowed")
+}
+
+func TestTransportFallbackGrowsOnRepeat(t *testing.T) {
+	const url = "rels://relay.example:443"
+	f := newTransportFallback()
+
+	want := transportFallbackBase
+	for i := range 6 {
+		d := f.recordFailure(url)
+		assert.Equal(t, want, d, "window after %d expiries", i)
+
+		// expire the window so the next failure is treated as a repeat
+		f.entries[url].until = time.Now().Add(-time.Second)
+
+		want = min(want*2, transportFallbackMax)
+	}
+
+	assert.Equal(t, transportFallbackMax, f.entries[url].duration, "window caps at the max")
+}
+
+func TestOnDatagramTooLargeAuto(t *testing.T) {
+	const url = "rels://relay.example:443"
+	t.Setenv(EnvRelayTransport, string(TransportModeAuto))
+
+	tf := newTransportFallback()
+	c := &Client{
+		log:               log.WithField("test", t.Name()),
+		connectionURL:     url,
+		transportFallback: tf,
+	}
+	conn := &closeTrackingConn{}
+
+	c.onDatagramTooLarge(conn, netErr.ErrDatagramTooLarge)
+
+	assert.True(t, conn.closed, "connection closed to force reconnect")
+	assert.True(t, tf.avoidDatagramSized(url), "fallback recorded for the server")
+
+	// A second oversized datagram on the same connection must not re-close.
+	conn.closed = false
+	c.onDatagramTooLarge(conn, netErr.ErrDatagramTooLarge)
+	assert.False(t, conn.closed, "single fallback per connection")
+}
+
+func TestOnDatagramTooLargeQUICPinned(t *testing.T) {
+	const url = "rels://relay.example:443"
+	t.Setenv(EnvRelayTransport, string(TransportModeQUIC))
+
+	tf := newTransportFallback()
+	c := &Client{
+		log:               log.WithField("test", t.Name()),
+		connectionURL:     url,
+		transportFallback: tf,
+	}
+	conn := &closeTrackingConn{}
+
+	c.onDatagramTooLarge(conn, netErr.ErrDatagramTooLarge)
+
+	assert.False(t, conn.closed, "QUIC pin keeps the connection, no fallback redial")
+	assert.False(t, tf.avoidDatagramSized(url), "QUIC pin records no fallback")
+}
+
+func TestTransportFallbackPerServer(t *testing.T) {
+	f := newTransportFallback()
+	f.recordFailure("rels://a.example:443")
+
+	assert.True(t, f.avoidDatagramSized("rels://a.example:443"))
+	assert.False(t, f.avoidDatagramSized("rels://b.example:443"), "fallback is scoped to one server")
+}