fix(monitor): refresh maintenance cache after loading sites

Move refreshMaintenanceCache() to after sites are loaded into
liveState. Previously it ran before the site loading loop, so on
the first iteration liveState was empty and the cache resolved
to nothing. This caused groups to briefly show DOWN on startup
for children that were in maintenance, until the next poll cycle
repopulated the cache.

internal/monitor/checks.go

@@ -239,10 +239,24 @@ func (e *Engine) checkGroup(_ context.Context, site models.Site) {
 		status = models.StatusPending
 	}
 
+	var prev models.Status
 	e.applyState(site.ID, func(s *models.Site) {
+		prev = s.Status
 		s.Status = status
+		if status != prev && prev != models.StatusPending {
+			s.StatusChangedAt = time.Now()
+		}
 	})
 	e.recordCheck(site.ID, 0, !status.IsBroken())
+
+	if status != prev && prev != models.StatusPending {
+		e.enqueueWrite(writeStateChange{siteID: site.ID, fromStatus: string(prev), toStatus: string(status)})
+		if status.IsBroken() {
+			e.AddLog(fmt.Sprintf("Group '%s' is %s", site.Name, status))
+		} else if prev.IsBroken() {
+			e.AddLog(fmt.Sprintf("Group '%s' recovered", site.Name))
+		}
+	}
 }
 
 func (e *Engine) EnqueueProbeCheck(siteID int, nodeID string, latencyNs int64, isUp bool) {

internal/monitor/monitor.go

@@ -283,8 +283,6 @@ func (e *Engine) Start(ctx context.Context) {
 			default:
 			}
 
-			e.refreshMaintenanceCache(ctx)
-
 			configs, err := e.db.GetSites(ctx)
 			if err != nil {
 				e.AddLog(fmt.Sprintf("Failed to load sites: %v", err))
@@ -327,6 +325,10 @@ func (e *Engine) Start(ctx context.Context) {
 				}
 			}
 
+			// Refresh after sites load so the cache covers newly added sites.
+			// On first iteration liveState was empty before the loop above.
+			e.refreshMaintenanceCache(ctx)
+
 			e.mu.RLock()
 			var vanished []int
 			for id := range e.liveState {

internal/monitor/monitor_test.go

@@ -445,6 +445,136 @@ func TestEngineStop_Idempotent(t *testing.T) {
 	e.Stop() // must not panic or block
 }
 
+// --- Group 11: Group State History ---
+
+func TestCheckGroup_SavesStateChange(t *testing.T) {
+	ms := newMockStore()
+	e := newTestEngine(ms)
+
+	group := models.Site{
+		SiteConfig: models.SiteConfig{ID: 100, Name: "Infra", Type: "group"},
+		SiteState:  models.SiteState{Status: models.StatusUp},
+	}
+	child := models.Site{
+		SiteConfig: models.SiteConfig{ID: 101, Name: "Server", Type: "ping", ParentID: 100},
+		SiteState:  models.SiteState{Status: models.StatusUp},
+	}
+	injectSite(e, group)
+	injectSite(e, child)
+
+	// Child goes down → group should transition UP→DOWN.
+	child.Status = models.StatusDown
+	injectSite(e, child)
+	e.checkGroup(context.Background(), group)
+
+	s, ok := getSite(e, 100)
+	if !ok {
+		t.Fatal("group not found")
+	}
+	if s.Status != models.StatusDown {
+		t.Errorf("expected DOWN, got %s", s.Status)
+	}
+	if s.StatusChangedAt.IsZero() {
+		t.Error("expected StatusChangedAt to be set")
+	}
+
+	logs := e.GetLogs()
+	found := false
+	for _, l := range logs {
+		if containsStr(l.Message, "Group 'Infra' is DOWN") {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Error("expected log entry for group state change")
+	}
+
+	// Drain the write channel to verify a state change was enqueued.
+	select {
+	case w := <-e.dbWrites:
+		if w.desc() != "state-change" {
+			// Could be a check write; try one more.
+			select {
+			case w2 := <-e.dbWrites:
+				if w2.desc() != "state-change" {
+					t.Errorf("expected state-change write, got %s then %s", w.desc(), w2.desc())
+				}
+			default:
+				t.Errorf("expected state-change write, got only %s", w.desc())
+			}
+		}
+	default:
+		t.Error("no write enqueued for group state change")
+	}
+}
+
+func TestCheckGroup_NoChangeNoWrite(t *testing.T) {
+	ms := newMockStore()
+	e := newTestEngine(ms)
+
+	group := models.Site{
+		SiteConfig: models.SiteConfig{ID: 100, Name: "Infra", Type: "group"},
+		SiteState:  models.SiteState{Status: models.StatusUp},
+	}
+	child := models.Site{
+		SiteConfig: models.SiteConfig{ID: 101, Name: "Server", Type: "ping", ParentID: 100},
+		SiteState:  models.SiteState{Status: models.StatusUp},
+	}
+	injectSite(e, group)
+	injectSite(e, child)
+
+	e.checkGroup(context.Background(), group)
+
+	// Only a check write should be enqueued, no state-change.
+	drainCount := 0
+	for {
+		select {
+		case w := <-e.dbWrites:
+			if w.desc() == "state-change" {
+				t.Error("state-change write enqueued when status didn't change")
+			}
+			drainCount++
+		default:
+			goto done
+		}
+	}
+done:
+	if drainCount == 0 {
+		t.Error("expected at least a check write")
+	}
+}
+
+func TestCheckGroup_RecoveryLog(t *testing.T) {
+	ms := newMockStore()
+	e := newTestEngine(ms)
+
+	group := models.Site{
+		SiteConfig: models.SiteConfig{ID: 100, Name: "Infra", Type: "group"},
+		SiteState:  models.SiteState{Status: models.StatusDown},
+	}
+	child := models.Site{
+		SiteConfig: models.SiteConfig{ID: 101, Name: "Server", Type: "ping", ParentID: 100},
+		SiteState:  models.SiteState{Status: models.StatusUp},
+	}
+	injectSite(e, group)
+	injectSite(e, child)
+
+	e.checkGroup(context.Background(), group)
+
+	logs := e.GetLogs()
+	found := false
+	for _, l := range logs {
+		if containsStr(l.Message, "Group 'Infra' recovered") {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Error("expected recovery log entry for group")
+	}
+}
+
 // --- Utilities ---
 
 func containsStr(s, substr string) bool {