ipn/ipnlocal: add advertised and primary route metrics

Updates tailscale/corp#22075

Signed-off-by: Kristoffer Dalby <kristoffer@tailscale.com>

ipn/ipnlocal/local.go

@@ -390,9 +390,18 @@ type updateStatus struct {
 }
 
 type metrics struct {
-	// advertisedRoutes is a metric that counts the number of network routes that are advertised by the local node.
+	// advertisedRoutes is a metric that reports the number of network routes that are advertised by the local node.
 	// This informs the user of how many routes are being advertised by the local node, excluding exit routes.
 	advertisedRoutes *usermetric.Gauge
+
+	// approvedRoutes is a metric that reports the number of network routes served by the local node and approved
+	// by the control server.
+	approvedRoutes *usermetric.Gauge
+
+	// primaryRoutes is a metric that reports the number of primary network routes served by the local node.
+	// A route being a primary route implies that the route is currently served by this node, and not by another
+	// subnet router in a high availability configuration.
+	primaryRoutes *usermetric.Gauge
 }
 
 // clientGen is a func that creates a control plane client.
@@ -441,6 +450,10 @@ func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, lo
 	m := metrics{
 		advertisedRoutes: sys.UserMetricsRegistry().NewGauge(
 			"tailscaled_advertised_routes", "Number of advertised network routes (e.g. by a subnet router)"),
+		approvedRoutes: sys.UserMetricsRegistry().NewGauge(
+			"tailscaled_approved_routes", "Number of approved network routes (e.g. by a subnet router)"),
+		primaryRoutes: sys.UserMetricsRegistry().NewGauge(
+			"tailscaled_primary_routes", "Number of network routes for which this node is a primary router (in high availability configuration)"),
 	}
 
 	b := &LocalBackend{
@@ -5388,6 +5401,11 @@ func (b *LocalBackend) setNetMapLocked(nm *netmap.NetworkMap) {
 	b.setTCPPortsInterceptedFromNetmapAndPrefsLocked(b.pm.CurrentPrefs())
 	if nm == nil {
 		b.nodeByAddr = nil
+
+		// If there is no netmap, the client is going into a "turned off"
+		// state so reset the metrics.
+		b.metrics.approvedRoutes.Set(0)
+		b.metrics.primaryRoutes.Set(0)
 		return
 	}
 
@@ -5408,6 +5426,15 @@ func (b *LocalBackend) setNetMapLocked(nm *netmap.NetworkMap) {
 	}
 	if nm.SelfNode.Valid() {
 		addNode(nm.SelfNode)
+
+		var approved float64
+		for _, route := range nm.SelfNode.AllowedIPs().All() {
+			if !views.SliceContains(nm.SelfNode.Addresses(), route) && !tsaddr.IsExitRoute(route) {
+				approved++
+			}
+		}
+		b.metrics.approvedRoutes.Set(approved)
+		b.metrics.primaryRoutes.Set(float64(tsaddr.WithoutExitRoute(nm.SelfNode.PrimaryRoutes()).Len()))
 	}
 	for _, p := range nm.Peers {
 		addNode(p)

tsnet/tsnet_test.go

@@ -26,6 +26,7 @@ import (
 	"os"
 	"path/filepath"
 	"reflect"
+	"runtime"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -924,6 +925,32 @@ func TestUserMetrics(t *testing.T) {
 	s1.lb.DebugForceNetmapUpdate()
 	s2.lb.DebugForceNetmapUpdate()
 
+	wantRoutes := float64(2)
+	if runtime.GOOS == "windows" {
+		wantRoutes = 0
+	}
+
+	// Wait for the routes to be propagated to node 1 to ensure
+	// that the metrics are up-to-date.
+	waitForCondition(t, "primary routes available for node1", 90*time.Second, func() bool {
+		ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+		defer cancel()
+		status1, err := lc1.Status(ctx)
+		if err != nil {
+			t.Logf("getting status: %s", err)
+			return false
+		}
+		if runtime.GOOS == "windows" {
+			// Windows does not seem to support or report back routes when running in
+			// userspace via tsnet. So, we skip this check on Windows.
+			// TODO(kradalby): Figure out if this is correct.
+			return true
+		}
+		// Wait for the primary routes to reach our desired routes, which is wantRoutes + 1, because
+		// the PrimaryRoutes list will contain a exit node route, which the metric does not count.
+		return status1.Self.PrimaryRoutes != nil && status1.Self.PrimaryRoutes.Len() == int(wantRoutes)+1
+	})
+
 	ctxLc, cancelLc := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancelLc()
 	metrics1, err := lc1.UserMetrics(ctxLc)
@@ -951,11 +978,25 @@ func TestUserMetrics(t *testing.T) {
 		t.Errorf("metrics1, tailscaled_advertised_routes: got %v, want %v", got, want)
 	}
 
+	// The control has approved 2 routes:
+	// - 192.0.2.0/24
+	// - 192.0.5.1/32
+	if got, want := parsedMetrics1["tailscaled_approved_routes"], wantRoutes; got != want {
+		t.Errorf("metrics1, tailscaled_approved_routes: got %v, want %v", got, want)
+	}
+
 	// Validate the health counter metric against the status of the node
 	if got, want := parsedMetrics1[`tailscaled_health_messages{type="warning"}`], float64(len(status1.Health)); got != want {
 		t.Errorf("metrics1, tailscaled_health_messages: got %v, want %v", got, want)
 	}
 
+	// The node is the primary subnet router for 2 routes:
+	// - 192.0.2.0/24
+	// - 192.0.5.1/32
+	if got, want := parsedMetrics1["tailscaled_primary_routes"], wantRoutes; got != want {
+		t.Errorf("metrics1, tailscaled_primary_routes: got %v, want %v", got, want)
+	}
+
 	metrics2, err := lc2.UserMetrics(ctx)
 	if err != nil {
 		t.Fatal(err)
@@ -978,8 +1019,28 @@ func TestUserMetrics(t *testing.T) {
 		t.Errorf("metrics2, tailscaled_advertised_routes: got %v, want %v", got, want)
 	}
 
+	// The control has approved 0 routes
+	if got, want := parsedMetrics2["tailscaled_approved_routes"], 0.0; got != want {
+		t.Errorf("metrics2, tailscaled_approved_routes: got %v, want %v", got, want)
+	}
+
 	// Validate the health counter metric against the status of the node
 	if got, want := parsedMetrics2[`tailscaled_health_messages{type="warning"}`], float64(len(status2.Health)); got != want {
 		t.Errorf("metrics2, tailscaled_health_messages: got %v, want %v", got, want)
 	}
+
+	// The node is the primary subnet router for 0 routes
+	if got, want := parsedMetrics2["tailscaled_primary_routes"], 0.0; got != want {
+		t.Errorf("metrics2, tailscaled_primary_routes: got %v, want %v", got, want)
+	}
+}
+
+func waitForCondition(t *testing.T, msg string, waitTime time.Duration, f func() bool) {
+	t.Helper()
+	for deadline := time.Now().Add(waitTime); time.Now().Before(deadline); time.Sleep(1 * time.Second) {
+		if f() {
+			return
+		}
+	}
+	t.Fatalf("waiting for condition: %s", msg)
 }

tstest/integration/testcontrol/testcontrol.go

@@ -366,6 +366,7 @@ func (s *Server) serveMachine(w http.ResponseWriter, r *http.Request) {
 func (s *Server) SetSubnetRoutes(nodeKey key.NodePublic, routes []netip.Prefix) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
+	s.logf("Setting subnet routes for %s: %v", nodeKey.ShortString(), routes)
 	mak.Set(&s.nodeSubnetRoutes, nodeKey, routes)
 }
 
@@ -1018,6 +1019,7 @@ func (s *Server) MapResponse(req *tailcfg.MapRequest) (res *tailcfg.MapResponse,
 
 	s.mu.Lock()
 	defer s.mu.Unlock()
+	res.Node.PrimaryRoutes = s.nodeSubnetRoutes[nk]
 	res.Node.AllowedIPs = append(res.Node.Addresses, s.nodeSubnetRoutes[nk]...)
 
 	// Consume a PingRequest while protected by mutex if it exists