state: skip node health writes that change nothing

An all-unchanged probe cycle no longer publishes a snapshot.

Updates #3417

hscontrol/state/node_health_test.go

@@ -65,3 +65,27 @@ func TestScanNodeHealthReportsInvalidNameWithoutMutating(t *testing.T) {
 	require.True(t, ok)
 	require.Empty(t, nv.GivenName(), "boot scan must not mutate the stored name")
 }
+
+// TestBatchSetNodeHealthUnchangedSkipsWrite ensures asking for the health
+// value already stored publishes no snapshot — no peersFunc invocation, no
+// peer-map rebuild.
+func TestBatchSetNodeHealthUnchangedSkipsWrite(t *testing.T) {
+	_, s, nodeID := persistTestSetup(t)
+	t.Cleanup(func() { _ = s.Close() })
+
+	// Use the State's nodeStore directly to confirm the node exists.
+	_, exists := s.nodeStore.GetNode(nodeID)
+	require.True(t, exists)
+
+	initialSnapshot := s.nodeStore.data.Load()
+
+	// The node starts with Unhealthy=false. Asking for "healthy=true"
+	// means Unhealthy stays false — no change.
+	for range 20 {
+		changed := s.BatchSetNodeHealth(map[types.NodeID]bool{nodeID: true})
+		require.False(t, changed,
+			"unchanged health write must not report a route-primary change")
+		require.Same(t, initialSnapshot, s.nodeStore.data.Load(),
+			"unchanged health write must not publish a new snapshot")
+	}
+}

hscontrol/state/state.go

@@ -33,6 +33,8 @@ import (
 	"github.com/juanfont/headscale/hscontrol/util"
 	"github.com/juanfont/headscale/hscontrol/util/zlog"
 	"github.com/juanfont/headscale/hscontrol/util/zlog/zf"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
 	"github.com/puzpuzpuz/xsync/v4"
 	"github.com/rs/zerolog"
 	"github.com/rs/zerolog/log"
@@ -1368,11 +1370,25 @@ func (s *State) BatchSetNodeHealth(updates map[types.NodeID]bool) bool {
 
 	prevRoutes := s.nodeStore.PrimaryRoutes()
 
+	// Skip writes that would not change anything so an all-unchanged
+	// probe cycle publishes no snapshot. healthSetter stays authoritative
+	// for the candidacy and race checks under the writer.
 	fns := make(map[types.NodeID]UpdateNodeFunc, len(updates))
+
 	for id, healthy := range updates {
+		if nv, ok := s.nodeStore.GetNode(id); ok && nv.Unhealthy() == !healthy {
+			haHealthUpdates.WithLabelValues("unchanged").Inc()
+
+			continue
+		}
+
 		fns[id] = healthSetter(healthy)
 	}
 
+	if len(fns) == 0 {
+		return false
+	}
+
 	s.nodeStore.UpdateNodes(fns)
 
 	return !maps.Equal(prevRoutes, s.nodeStore.PrimaryRoutes())
@@ -1383,15 +1399,27 @@ func (s *State) BatchSetNodeHealth(updates map[types.NodeID]bool) bool {
 // an unhealthy mark only sticks when the node is still online and
 // still advertises approved routes, so a node that left HA candidacy
 // between probe dispatch and result does not carry a stale bit.
+// Bounded labels only: every requested health update lands in exactly one
+// outcome, so the series can be summed to the probe request count.
+var haHealthUpdates = promauto.NewCounterVec(prometheus.CounterOpts{
+	Namespace: prometheusNamespace,
+	Name:      "ha_health_updates_total",
+	Help:      "HA health updates by outcome: unchanged and skipped, applied, or rejected because the node left candidacy.",
+}, []string{"result"})
+
 func healthSetter(healthy bool) UpdateNodeFunc {
 	return func(n *types.Node) {
 		if !healthy {
 			online := n.IsOnline != nil && *n.IsOnline
 			if !online || len(n.AllApprovedRoutes()) == 0 {
+				haHealthUpdates.WithLabelValues("rejected").Inc()
+
 				return
 			}
 		}
 
+		haHealthUpdates.WithLabelValues("changed").Inc()
+
 		n.Unhealthy = !healthy
 	}
 }