[client] Persist route selection (#2810)

client/internal/routemanager/manager.go

@@ -32,7 +32,7 @@ import (
 
 // Manager is a route manager interface
 type Manager interface {
-	Init(*statemanager.Manager) (nbnet.AddHookFunc, nbnet.RemoveHookFunc, error)
+	Init() (nbnet.AddHookFunc, nbnet.RemoveHookFunc, error)
 	UpdateRoutes(updateSerial uint64, newRoutes []*route.Route) (map[route.ID]*route.Route, route.HAMap, error)
 	TriggerSelection(route.HAMap)
 	GetRouteSelector() *routeselector.RouteSelector
@@ -59,6 +59,7 @@ type DefaultManager struct {
 	routeRefCounter      *refcounter.RouteRefCounter
 	allowedIPsRefCounter *refcounter.AllowedIPsRefCounter
 	dnsRouteInterval     time.Duration
+	stateManager         *statemanager.Manager
 }
 
 func NewManager(
@@ -69,6 +70,7 @@ func NewManager(
 	statusRecorder *peer.Status,
 	relayMgr *relayClient.Manager,
 	initialRoutes []*route.Route,
+	stateManager *statemanager.Manager,
 ) *DefaultManager {
 	mCTX, cancel := context.WithCancel(ctx)
 	notifier := notifier.NewNotifier()
@@ -80,12 +82,12 @@ func NewManager(
 		dnsRouteInterval: dnsRouteInterval,
 		clientNetworks:   make(map[route.HAUniqueID]*clientNetwork),
 		relayMgr:         relayMgr,
-		routeSelector:    routeselector.NewRouteSelector(),
 		sysOps:           sysOps,
 		statusRecorder:   statusRecorder,
 		wgInterface:      wgInterface,
 		pubKey:           pubKey,
 		notifier:         notifier,
+		stateManager:     stateManager,
 	}
 
 	dm.routeRefCounter = refcounter.New(
@@ -121,7 +123,7 @@ func NewManager(
 }
 
 // Init sets up the routing
-func (m *DefaultManager) Init(stateManager *statemanager.Manager) (nbnet.AddHookFunc, nbnet.RemoveHookFunc, error) {
+func (m *DefaultManager) Init() (nbnet.AddHookFunc, nbnet.RemoveHookFunc, error) {
 	if nbnet.CustomRoutingDisabled() {
 		return nil, nil, nil
 	}
@@ -137,14 +139,38 @@ func (m *DefaultManager) Init(stateManager *statemanager.Manager) (nbnet.AddHook
 
 	ips := resolveURLsToIPs(initialAddresses)
 
-	beforePeerHook, afterPeerHook, err := m.sysOps.SetupRouting(ips, stateManager)
+	beforePeerHook, afterPeerHook, err := m.sysOps.SetupRouting(ips, m.stateManager)
 	if err != nil {
 		return nil, nil, fmt.Errorf("setup routing: %w", err)
 	}
+
+	m.routeSelector = m.initSelector()
+
 	log.Info("Routing setup complete")
 	return beforePeerHook, afterPeerHook, nil
 }
 
+func (m *DefaultManager) initSelector() *routeselector.RouteSelector {
+	var state *SelectorState
+	m.stateManager.RegisterState(state)
+
+	// restore selector state if it exists
+	if err := m.stateManager.LoadState(state); err != nil {
+		log.Warnf("failed to load state: %v", err)
+		return routeselector.NewRouteSelector()
+	}
+
+	if state := m.stateManager.GetState(state); state != nil {
+		if selector, ok := state.(*SelectorState); ok {
+			return (*routeselector.RouteSelector)(selector)
+		}
+
+		log.Warnf("failed to convert state with type %T to SelectorState", state)
+	}
+
+	return routeselector.NewRouteSelector()
+}
+
 func (m *DefaultManager) EnableServerRouter(firewall firewall.Manager) error {
 	var err error
 	m.serverRouter, err = newServerRouter(m.ctx, m.wgInterface, firewall, m.statusRecorder)
@@ -252,6 +278,10 @@ func (m *DefaultManager) TriggerSelection(networks route.HAMap) {
 		go clientNetworkWatcher.peersStateAndUpdateWatcher()
 		clientNetworkWatcher.sendUpdateToClientNetworkWatcher(routesUpdate{routes: routes})
 	}
+
+	if err := m.stateManager.UpdateState((*SelectorState)(m.routeSelector)); err != nil {
+		log.Errorf("failed to update state: %v", err)
+	}
 }
 
 // stopObsoleteClients stops the client network watcher for the networks that are not in the new list

client/internal/routemanager/manager_test.go

@@ -424,9 +424,9 @@ func TestManagerUpdateRoutes(t *testing.T) {
 
 			statusRecorder := peer.NewRecorder("https://mgm")
 			ctx := context.TODO()
-			routeManager := NewManager(ctx, localPeerKey, 0, wgInterface, statusRecorder, nil, nil)
+			routeManager := NewManager(ctx, localPeerKey, 0, wgInterface, statusRecorder, nil, nil, nil)
 
-			_, _, err = routeManager.Init(nil)
+			_, _, err = routeManager.Init()
 
 			require.NoError(t, err, "should init route manager")
 			defer routeManager.Stop(nil)

client/internal/routemanager/mock.go

@@ -21,7 +21,7 @@ type MockManager struct {
 	StopFunc             func(manager *statemanager.Manager)
 }
 
-func (m *MockManager) Init(*statemanager.Manager) (net.AddHookFunc, net.RemoveHookFunc, error) {
+func (m *MockManager) Init() (net.AddHookFunc, net.RemoveHookFunc, error) {
 	return nil, nil, nil
 }
 

client/internal/routemanager/refcounter/refcounter.go

@@ -71,11 +71,14 @@ func New[Key comparable, I, O any](add AddFunc[Key, I, O], remove RemoveFunc[Key
 }
 
 // LoadData loads the data from the existing counter
+// The passed counter should not be used any longer after calling this function.
 func (rm *Counter[Key, I, O]) LoadData(
 	existingCounter *Counter[Key, I, O],
 ) {
 	rm.mu.Lock()
 	defer rm.mu.Unlock()
+	existingCounter.mu.Lock()
+	defer existingCounter.mu.Unlock()
 
 	rm.refCountMap = existingCounter.refCountMap
 	rm.idMap = existingCounter.idMap
@@ -231,6 +234,9 @@ func (rm *Counter[Key, I, O]) MarshalJSON() ([]byte, error) {
 
 // UnmarshalJSON implements the json.Unmarshaler interface for Counter.
 func (rm *Counter[Key, I, O]) UnmarshalJSON(data []byte) error {
+	rm.mu.Lock()
+	defer rm.mu.Unlock()
+
 	var temp struct {
 		RefCountMap map[Key]Ref[O]   `json:"refCountMap"`
 		IDMap       map[string][]Key `json:"idMap"`
@@ -241,6 +247,13 @@ func (rm *Counter[Key, I, O]) UnmarshalJSON(data []byte) error {
 	rm.refCountMap = temp.RefCountMap
 	rm.idMap = temp.IDMap
 
+	if temp.RefCountMap == nil {
+		temp.RefCountMap = map[Key]Ref[O]{}
+	}
+	if temp.IDMap == nil {
+		temp.IDMap = map[string][]Key{}
+	}
+
 	return nil
 }
 

client/internal/routemanager/state.go

@@ -0,0 +1,19 @@
+package routemanager
+
+import (
+	"github.com/netbirdio/netbird/client/internal/routeselector"
+)
+
+type SelectorState routeselector.RouteSelector
+
+func (s *SelectorState) Name() string {
+	return "routeselector_state"
+}
+
+func (s *SelectorState) MarshalJSON() ([]byte, error) {
+	return (*routeselector.RouteSelector)(s).MarshalJSON()
+}
+
+func (s *SelectorState) UnmarshalJSON(data []byte) error {
+	return (*routeselector.RouteSelector)(s).UnmarshalJSON(data)
+}