Allow other states for windows neighbor network monitor

client/internal/networkmonitor/monitor_windows.go

@@ -33,12 +33,8 @@ func checkChange(ctx context.Context, nexthopv4, nexthopv6 systemops.Nexthop, ca
 			return fmt.Errorf("get neighbors: %w", err)
 		}
 
-		if n, ok := initialNeighbors[nexthopv4.IP]; ok {
-			neighborv4 = &n
-		}
-		if n, ok := initialNeighbors[nexthopv6.IP]; ok {
-			neighborv6 = &n
-		}
+		neighborv4 = assignNeighbor(nexthopv4, initialNeighbors)
+		neighborv6 = assignNeighbor(nexthopv6, initialNeighbors)
 	}
 	log.Debugf("Network monitor: initial IPv4 neighbor: %v, IPv6 neighbor: %v", neighborv4, neighborv6)
 
@@ -58,6 +54,16 @@ func checkChange(ctx context.Context, nexthopv4, nexthopv6 systemops.Nexthop, ca
 	}
 }
 
+func assignNeighbor(nexthop systemops.Nexthop, initialNeighbors map[netip.Addr]systemops.Neighbor) *systemops.Neighbor {
+	if n, ok := initialNeighbors[nexthop.IP]; ok &&
+		n.State != unreachable &&
+		n.State != incomplete &&
+		n.State != tbd {
+		return &n
+	}
+	return nil
+}
+
 func changed(
 	nexthopv4 systemops.Nexthop,
 	neighborv4 *systemops.Neighbor,
@@ -127,7 +133,7 @@ func neighborChanged(nexthop systemops.Nexthop, neighbor *systemops.Neighbor, ne
 
 	// TODO: consider non-local nexthops, e.g. on point-to-point interfaces
 	if n, ok := neighbors[nexthop.IP]; ok {
-		if n.State != reachable && n.State != permanent {
+		if n.State == unreachable || n.State == incomplete {
 			log.Infof("network monitor: neighbor %s (%s) is not reachable: %s", neighbor.IPAddress, neighbor.LinkLayerAddress, stateFromInt(n.State))
 			return true
 		} else if n.InterfaceIndex != neighbor.InterfaceIndex {

client/internal/routemanager/dynamic/route.go

@@ -23,7 +23,8 @@ import (
 const (
 	DefaultInterval = time.Minute
 
-	minInterval = 2 * time.Second
+	minInterval     = 2 * time.Second
+	failureInterval = 5 * time.Second
 
 	addAllowedIP = "add allowed IP %s: %w"
 )
@@ -160,7 +161,12 @@ func (r *Route) startResolver(ctx context.Context) {
 	ticker := time.NewTicker(interval)
 	defer ticker.Stop()
 
-	r.update(ctx)
+	if err := r.update(ctx); err != nil {
+		log.Errorf("Failed to resolve domains for route [%v]: %v", r, err)
+		if interval > failureInterval {
+			ticker.Reset(failureInterval)
+		}
+	}
 
 	for {
 		select {
@@ -168,17 +174,28 @@ func (r *Route) startResolver(ctx context.Context) {
 			log.Debugf("Stopping dynamic route resolver for domains [%v]", r)
 			return
 		case <-ticker.C:
-			r.update(ctx)
+			if err := r.update(ctx); err != nil {
+				log.Errorf("Failed to resolve domains for route [%v]: %v", r, err)
+				// Use a lower ticker interval if the update fails
+				if interval > failureInterval {
+					ticker.Reset(failureInterval)
+				}
+			} else if interval > failureInterval {
+				// Reset to the original interval if the update succeeds
+				ticker.Reset(interval)
+			}
 		}
 	}
 }
 
-func (r *Route) update(ctx context.Context) {
+func (r *Route) update(ctx context.Context) error {
 	if resolved, err := r.resolveDomains(); err != nil {
-		log.Errorf("Failed to resolve domains for route [%v]: %v", r, err)
+		return fmt.Errorf("resolve domains: %w", err)
 	} else if err := r.updateDynamicRoutes(ctx, resolved); err != nil {
-		log.Errorf("Failed to update dynamic routes for [%v]: %v", r, err)
+		return fmt.Errorf("update dynamic routes: %w", err)
 	}
+
+	return nil
 }
 
 func (r *Route) resolveDomains() (domainMap, error) {