fix(route): insert route tracker entry only after AddRoute succeeds

Signed-off-by: Maximilian Rink <maximilian.rink@telekom.de>

pkg/route/manager.go

@@ -49,9 +49,6 @@ func (m *Manager) Add(object string, r route, precheck, update bool) error {
 	itm, exists := m.tracker[key]
 
 	if !exists {
-		m.tracker[key] = newItem(r)
-		itm = m.tracker[key]
-
 		added, err := r.AddRoute(precheck)
 		if err != nil {
 			if update && errors.Is(err, syscall.EEXIST) && update {
@@ -70,6 +67,9 @@ func (m *Manager) Add(object string, r route, precheck, update bool) error {
 			}
 		}
 
+		itm = newItem(r)
+		m.tracker[key] = itm
+
 		if added {
 			log.Debug("[RT] added route", "path", key, "object", object)
 		}

pkg/route/manager_atomic_test.go

@@ -0,0 +1,29 @@
+package route
+
+import (
+	"errors"
+	"testing"
+)
+
+var errTransientRouteAdd = errors.New("transient route add failure")
+
+func TestManagerRetriesRouteAfterInitialAddFailure(t *testing.T) {
+	m := NewManager()
+	r := &mockRoute{hash: "retry", addErr: errTransientRouteAdd}
+
+	if err := m.Add("service", r, false, true); !errors.Is(err, errTransientRouteAdd) {
+		t.Fatalf("first add error = %v, want %v", err, errTransientRouteAdd)
+	}
+
+	// A transient netlink failure (for example while the link is being recreated)
+	// must not poison the in-memory tracker. The next reconciliation has to retry
+	// the kernel operation.
+	r.addErr = nil
+	r.added = true
+	if err := m.Add("service", r, false, true); err != nil {
+		t.Fatalf("retry add failed: %v", err)
+	}
+	if r.addCalls != 2 {
+		t.Fatalf("AddRoute called %d times, want 2", r.addCalls)
+	}
+}

pkg/route/manager_test.go

@@ -186,6 +186,7 @@ func Test_MultipleRoutesAddDel(t *testing.T) {
 
 type mockRoute struct {
 	added     bool
+	addCalls  int
 	addErr    error
 	updated   bool
 	updateErr error
@@ -195,6 +196,7 @@ type mockRoute struct {
 }
 
 func (mr *mockRoute) AddRoute(_ bool) (bool, error) {
+	mr.addCalls++
 	return mr.added, mr.addErr
 }